Computer Science Review: Leila Cristina C. Bergamasco, Fátima L.S. Nunes

Computer Science Review 31 (2019) 19–38
Contents lists available at ScienceDirect
Computer Science Review

journal homepage: www.elsevier.com/locate/cosrev
Intelligent retrieval and classification in three-dimensional

biomedical images — A systematic mapping✩
∗
Leila Cristina C. Bergamasco a , , Fátima L.S. Nunes a,b
a
Escola Politécnica da Universidade de São Paulo, Professor Luciano Gualberto Avenue - 3, São Paulo - SP, 05508-010, Brazil
b
School of Arts, Sciences and Humanities, Universidade de São Paulo, Arlindo Béttio Avenue - 1000, São Paulo - SP, 03828-000, Brazil
highlights
• A systematic mapping covering intelligent retrieval for 3D medical images.

• Descriptors focusing in shape are widely developed.
• 3D medical images use computing techniques from 2D domain, limiting the results.
article info a b s t r a c t
Article history: The massive generation of data has raised new research topics, such as methods to store and to retrieve
Received 30 March 2018 large volumes of information. Some medical image modalities, such as Magnetic Resonance Imaging,
Accepted 25 October 2018 generate hundreds images series and many research groups have presented studies to develop intelligent
Available online xxxx
techniques to classify and to retrieve this information. However, these studies are dispersed in several
Keywords: databases, and cataloged by using different terms. In this paper we present an analysis of these studies,
CBIR through a Systematic Mapping that identifies methods and techniques currently being used in this
Classification scenario. In addition, we provide a perspective about the type of scientific literature the researchers
Descriptors have been disclosed their studies as well as impact on the scientific literature in dissemination of this
Information retrieval knowledge domain. Some challenges and research opportunities are also highlighted in order to propitiate
Medical images
advances in the area.
© 2018 Elsevier Inc. All rights reserved.
Contents
1. Introduction......................................................................................................................................................................................................................... 20
2. Research method................................................................................................................................................................................................................. 21
2.1. Planning .................................................................................................................................................................................................................. 21
2.2. Conduction and data extraction phases................................................................................................................................................................ 21
3. Global analysis .................................................................................................................................................................................................................... 22
4. Descriptors .......................................................................................................................................................................................................................... 25
4.1. Spherical Harmonics .............................................................................................................................................................................................. 25
4.2. 3D Moment Invariants ........................................................................................................................................................................................... 25
4.3. 3D Wavelet Transform ........................................................................................................................................................................................... 26
5. Indexing methods ............................................................................................................................................................................................................... 26
5.1. Graphs-based indexing methods ......................................................................................................................................................................... 26
5.2. K-d trees.................................................................................................................................................................................................................. 27
6. Similarity comparison methods......................................................................................................................................................................................... 27
7. Databases............................................................................................................................................................................................................................. 27
8. Applications......................................................................................................................................................................................................................... 28
9. Evaluation metrics .............................................................................................................................................................................................................. 29
✩ No author associated with this paper has disclosed any potential or pertinent conflicts which may be perceived to have impending conflict with this work. For full
disclosure statements refer to https://doi.org/10.1016/j.cosrev.2018.10.003.
∗ Corresponding author.
E-mail addresses: [email protected] (L.C.C. Bergamasco), [email protected] (F.L.S. Nunes).
https://doi.org/10.1016/j.cosrev.2018.10.003
1574-0137/© 2018 Elsevier Inc. All rights reserved.
20 L.C.C. Bergamasco and F.L.S. Nunes / Computer Science Review 31 (2019) 19–38
10. Discussion............................................................................................................................................................................................................................ 29
10.1. General definitions ................................................................................................................................................................................................. 29
10.2. Papers quality ......................................................................................................................................................................................................... 30
10.3. Descriptors.............................................................................................................................................................................................................. 30
10.4. Indexing methods................................................................................................................................................................................................... 30
10.5. Similarity comparison ............................................................................................................................................................................................ 30
10.6. Databases ................................................................................................................................................................................................................ 31
10.7. Applications ............................................................................................................................................................................................................ 31
10.8. Evaluation metrics.................................................................................................................................................................................................. 31
10.9. Opportunities.......................................................................................................................................................................................................... 31
11. Conclusion ........................................................................................................................................................................................................................... 32
Acknowledgments .............................................................................................................................................................................................................. 32
References ........................................................................................................................................................................................................................... 32
advance in CBIR techniques applied to medical images, with special

1. Introduction attention to the 3D domain, but focusing on search of key slices
of the same volume. Finally, with the advent of deep learning in
The amount of data generated by our society has grown ex- medical image processing, Litjens et al. [11] provide a survey about
ponentially. From 2005 to 2020 it will increase from 130 exabytes this technique and how it is used in different image processing
to 40,000 exabytes, or 40 trillion gigabytes [1]. In Healthcare, this domains — segmentation, registration, classification, and retrieval.
scenario is similar: in 2009 it was estimated about 434 pentabytes This Systematic Mapping is innovative since it provides an
data storage only related to Healthcare in United States, and unlike updated review of processing and retrieval from a 3D perspective
other sectors where big data is represented mostly by text, this sec- considering the medical image domain. Besides including more
tor also stores medical images, generating individual exams with recent studies, a Systematic Mapping was conducted, which allows
large volume of data [2]. In 2020, about 25.000 petabytes of infor-
reproducing and auditing the search. Additionally, different from
mation from Healthcare, related to digital data and patient records
previous reviews about this subject, we also performed an analysis
are expected [3]. Considering this scenario, many researchers have
about the studies quality using some well-known metrics such
proposed intelligent methods to process, classify or retrieve these
as H5-index impact factor and the scientific literature where the
data to make them available immediately wherever and whenever
studies were published.
necessary.
Many of the intelligent retrieval systems implement methods Furthermore, we provide a critical perspective about this sub-
that use metadata-based approaches to retrieve objects from im- ject since we made an effort to answer these questions: are the
age databases. This type of retrieval requires that the user employ researchers producing relevant studies in this area? Where are the
keywords which will be used as parameters. Consequently, this researchers publishing their studies? Is the Healthcare area really
method may contain flaws, due poorly chosen keywords, sub- being benefited from the retrieval and classification techniques?
jectivity of the content searched, among others. Systems dealing For achieve the answers of these questions we analyzed the
with medical image content can be an alternative for meta-data included papers through different perspectives: (1) papers quality
searches. The ‘‘content’’ in this scenario consider characteristics through h5-index metric and by the papers content such as quality
extracted from the image related to aspects such as color, texture of methodology description and results reported; (2) descriptors
and shape [4]. used; (3) similarity comparison methods applied; (4) indexing
Medical exams are currently more complex and can generate a methods used to store the feature vectors; (5) databases used
large volume of data to be analyzed by experts. Computed Tomog- by the author to validate their approach; (6) the proposal of the
raphy (CT), Magnetic Resonance Imaging (MRI), Positron Emission techniques developed, and (7) the evaluation metrics used in the
Tomography (PET) and Single-photon Emission Computed Tomog- included papers.
raphy (SPECT) are examples of medical modalities that can gener- Besides CBIR studies, we also investigated classification prob-
ate hundreds of slices in a single exam [5]. In order to make a cor- lems that produce innovative descriptors and different methods to
rect diagnosis, a careful investigation is usually conducted on each classify the 3D medical objects or images. When it is mentioned
slice. The analysis of these two-dimensional (2D) images, even 3D graphic structures, i.e., surface or volume resultant of a recon-
using an automated process, can be highly time-consuming. To struction process, we refer them as 3D medical objects. When the
minimize this problem when the process is automated, two main graphical objects are related to 2D images (slices from 3D exams,
approaches have been frequently used: optimizing the feature
such as MRI or CT) we refer them as medical images.
extraction from 2D-slice sets or extracting features directly from
This paper is organized as follows: Section 2 explains our re-
three-dimensional (3D) objects [4]. The latter approach results
search method — the Systematic Mapping, presenting some topics
in systems named 3D Content-based Image Retrieval (3D CBIR),
considered in this research process. Section 3 presents a quanti-
model retrieval, 3D object retrieval or simply 3D image intelligent
tative and qualitative global analysis about the papers related to
retrieval.
Some previous reviews were published regarding this theme. In classification and retrieval in 3D medical objects or images, and a
a time perspective, the first work published was Müller et al. [6], high level view about the studies quality. Sections 4–6, respectively
which provided an overview of medical images retrieval, exploring discuss descriptors, indexing methods, and methods for measuring
2D features, and pointing out 3D domain as ‘‘future trend’’. In Datta the similarity developed in the articles included along the research.
et al. [7], the focus was on CBIR in several domains, such as medical Section 7 presents the medical images databases found in the
images, engineering, and arts. The authors offer an overview of the papers selected. Section 8 discuss the main problems solved by
publications of this subject and tried to map the main techniques computational techniques, and Section 9 details evaluation metrics
used for this purpose. Akgul et al. [8] and Hwang et al. [9] presented cited by the studies included. Section 10 discusses the results found
the state of art of CBIR in medical images with emphasis mainly on and maps some trends for this knowledge field. Finally, Section 11
2D descriptors and similarity functions. Kumar et al. [10] offer an presents the final conclusion.
L.C.C. Bergamasco and F.L.S. Nunes / Computer Science Review 31 (2019) 19–38 21
Table 1
Strings used in each scientific library and amount of retrieval papers.
Database Search string Results
similarity retrieval AND 3D and medical 25
CBIR AND MRI 22
CBIR AND CT 32
EE
descriptor medical 3D 53
IE image retrieval AND medical AND 3D 171
3D medical volume retrieval 30
3D medical surface retrieval 37
(‘‘model retrieval’’) OR ‘‘3D CBIR’’ OR (‘‘image 34
ed
retrieval’’[Title/Abstract] AND 3D)

bM
Pu
descriptor AND 3D AND retrieval 25

similarity retrieval AND 3D 63
(image retrieval AND medical) and TITLE-ABSTR-KEY(3D). 82
CBIR AND (CT or MRI) 353
t
ec
ir
TITLE-ABSTR-KEY(3D medical ) and (volume retrieval) 84

D
ce
TITLE-ABSTR-KEY(3D medical ) and (surface retrieval) 65

n
ie
TITLE-ABSTR-KEY(descriptor) and (medical 3D) 428

Sc
‘‘similarity retrieval’’ AND 3D AND medical 69

(Abstract:image and Abstract:retrieval and Abstract:medical and 26
Abstract:3D)
CBIR AND (MRI or CT) 97
M
AC
‘‘similarity retrieval’’ and 3D and medical 94

( medical 3D) and (Title:descriptor) 37
(Abstract:3D and Abstract:medical and Abstract:volume and 8
Abstract:retrieval)
(Abstract:3D and Abstract:medical and Abstract:surface and 11
Abstract:retrieval)
ABSTRACT:(3D) AND ABSTRACT:(medical) AND 27
ABSTRACT:(retrieval)
IE
SP
ABSTRACT:(medical descriptor) AND ABSTRACT:(3D) 11

ABSTRACT:(medical) AND (‘‘volume retrieval’’) 1
(‘‘similarity retrieval’’) AND ABSTRACT:(medical) 1
‘‘image retrieval’’ or ‘‘model retrieval’’ AND (medical) AND NOT 47
e r
ng
(Overview or semantic)
ri
Sp
3D AND medical AND (‘‘image retrieval’’ or ‘‘model retrieval’’) 29
2. Research method 5. Which are the practical applications these techniques are
developed for?
Systematic Mapping, also called scoping review is defined by 6. Is there relationships between the techniques developed
Petticrew and Roberts [12] as a review that evaluates the types of and the medical images or anomalies?
studies, the date on which they were made, and where they are
The databases selected for this Systematic Mapping were: IEEE
located. It presents a methodology which can be reproduced and
Xplore, ACM Digital Library, Science Direct, SPIE library, Springer-
audited by other researchers, as well as provide a general overview
Link and Pubmed. We did not limit a period for the search but
of complex subjects. Three phases are usually defined: planning,
the newest papers returned in this Systematic Mapping are from
conduction and data extraction.
January 2018. We considered only papers written in English. The
The planning phase is very similar to the Systematic Review
keywords, search strings and the number of results found in each
[13]. At this stage some topics are defined such as review questions,
database are presented in Table 1.
keywords, scientific libraries, inclusion and exclusion criteria. In
the Conduction stage each paper is evaluated accordingly to the
inclusion and exclusion criteria defined in the protocol. Finally, 2.2. Conduction and data extraction phases
during the extraction phase the studies are deeply analyzed aiming
at classifying and categorizing data about the subject. After retrieving the papers from the databases, we applied the
following Inclusion (I) and Exclusion (E) criteria listed in Table 2.
To evaluate whether an article would be included in this study,
2.1. Planning
the following rules were elaborated considering the type of article
and the inclusion and exclusion criteria presented:
At this stage, we defined the main topics to outline the research,
such as the questions that this study proposes to answer: • (I1 OR I2 OR I3) AND (NOT(E1 OR E2 OR E3))
• (I4 AND (NOT (E1))
1. Which descriptors and methods for measuring similarity are
being used to retrieve and to store 3D medical objects or During this phase, we followed some guidelines in the papers
images? evaluation process: firstly, the studies that investigate molecules
2. Which data structures are used for descriptors representa- shape classification and retrieval are not included because these
tion? 3D objects are not reconstructed from medical images. In this
3. What methods are used to evaluate these processing and sense, we considered ‘‘3D medical object’’ only the sets of 2D slices,
retrieval techniques? such as from MRI and CT, the 3D object meshes with the surface
4. Which types of medical images are being analyzed? reconstructed or the 3D object represented in a voxels grid. Thus,
Table 2
List of Inclusion and Exclusion criteria.
Criterion Description
I1 Conceptualizes descriptors or similarity
comparison methods for 3D medical objects or
medical images classification and retrieval.
I2 Conceptualize techniques to improve results of
the content-based searches.
I3 Evaluates methods for 3D medical objects
classification and retrieval.
I4 Details medical databases and state of the art or
reviews about this subject.
E1 Focus is different from 3D medical objects.
E2 Papers that do not develop classification and
retrieval techniques.
E3 Uses retrieval and classification techniques for
tracking or video retrieval.
this type of studies were excluded by the first exclusion criterion

(E1). Secondly, as the purpose of this paper is to evaluate the
retrieval and classification tasks, papers that performed segmen- Fig. 1. Relationship between Publication type and h5-index distribution. Journals
tation, reconstruction, tracking and modeling were excluded by E3 in which the papers selected were published present higher h5-index when com-
criterion. Finally, we analyzed individually papers that executed pared to Conferences.
face recognition and retrieval, in order to validate if the object
was reconstructed from MRI or CT images; papers that extracted
features from a face multi-views images were also excluded. The 107 studies included were distributed into categories con-
We also defined quality criteria applied to the papers included. sidering the retrieval and classification techniques. The distribu-
The main goal of these criteria is to analyze characteristics of the tion of each technique was 62% for classification, 32% for CBIR, 1%
papers included and to score them both to compose an additional exploits databases for 3D medical objects and other papers such as
exclusion criterion, and support us on the analysis of the relation- review ones represent 5% of the papers selected.
ship between the article and the publication quality. Besides understanding how computational techniques are ap-
If an article completely satisfies a quality criterion, it gets value plied to store and retrieve 3D medical objects or images, we also
1 for the item; if the criterion is partially satisfied, the article verified what kind of scientific literature the researchers are pub-
receives value 0.5 for the item, and the value zero is assigned if the lishing their works in. Fig. 1 shows a difference between Sympo-
criterion is not satisfied for that item. A weight is assigned for each siums/Conference and Journal publications when the h5-index is
criterion so that the sum of an article can range from 0 to 4. Table 3 considered. A percentage of 53% of the papers selected were pub-
presents the quality criteria used in this Systematic Mapping. lished in Symposiums and Conferences, while 47% were published
We found 1962 papers, from which 520 were duplicated (the in Journals. The h5-index average of the articles included2 was
same papers were found from different search strings), one was higher in Journals (average of 38) than in Symposiums (average of
not in English. After applying the Inclusion/Exclusion criteria, we 17).
obtained 121 papers selected for the Data Extraction phase. At this In addition, we ranked the top 10 scientific literature according
stage, we applied the quality criteria and papers with score below to their h5-index and (Table 4). Both Fig. 1 and Table 4 indicates
3.5 were excluded.1 Thus, we selected 107 studies that will be that studies on this knowledge area have attracted the scientific
presented and discussed in the next sections. community and generated relevant contributions, since we note
papers published in the scientific literature with high h5-index in
3. Global analysis Engineering, Physics, Computer Science and Mathematics knowl-
edge domains [14].
Before starting the studies analysis it is important to highlight The application of quality criteria in the papers included
the differences between CBIR and classification techniques, since showed that the publication with the higher h5-index (Neuroim-
these differences guided us on the papers categorization. age, with a h5-index of 120) had a article published that was
For image classification problems, there are labeled classes in evaluated with the maximum score (4) in the quality criteria
which objects retrieved are classified. In CBIR there are not labeled indicating that the studies have well defined goals and the results
classes and the algorithm efficiency usually is calculated based were evaluated suitably. (See Fig. 2.)
on the first images retrieved and their similarity with the image As mentioned in Section 1, this Systematic Mapping focus on
given as query [4]. For example: if the user gives a flower image as CBIR and classification of 3D medical objects structure. Fig. 3 shows
query, we can consider a perfect classification if the result return all each of these techniques related to their respective h5-index aver-
flowers images stored in the database. In this example, the scope age. We verified that articles dealing with classification problems
is correctly retrieve flowers and no-flowers. For retrieval purposes
we consider a perfect retrieval if the first images retrieved are 2 In this study we used the h5-index as metric instead of the Journal Citation
flowers and they must have the same attributes (color, shape, etc.). Report [14], which is well-known among researchers. The study of [15] pointed
In retrieval problems the search context is more specific. out a strong correlation between the two metrics using Spearman’s correlation
(0.718 (p < 0.000)). The advantage of h5-index is that it can be calculated for both
Conferences and Journals. The h5-index metric is composed by the total number of
1 The list of each paper evaluated following the inclusion and quality criteria publications and the total number of citations to those works. For example: if an
defined below can be find in the URL: http://www.lapis.each.usp.br/Bergamasco_ conference has h5-index = 30 it means that this conference has 30 papers with at
SM/SystematicMapping.xls. least 30 citations in the last 5 years.
Table 3
Quality criteria and weights.
Criteria Description Weight
QC1 The study proposal is clearly described 1
QC2 There is a clear technique explanation including 1
implementation, evaluation method and/or review
methodology.
QC3 The study describes the goal or the importance of its research. 1
QC4 The study presents evaluation of the technique or method 0.4
developed.
QC5 The results are reported 0.6
Total 4
Table 4
Ranking of top 10 publication type and their respective h5-index. Periodicals and Conferences that published four or more of the studies
were included in this list) (J = Journal and S = Symposium or Conference).
Publication title h5-index Frequency (# of Type
papers included)
Neuroimage 122 3 J
Information Sciences 75 1 J
International Journal of Computer Vision 68 1 J
IEEE International Conference on Acoustics, 67 1 S
Speech and Signal Processing
Medical Image Analysis 49 3 J
Pattern Recognition Letters 46 2 J
Computer Methods and Programs in Biomedicine 38 2 J
IEEE International Conference on Image 38 2 S
Processing (ICIP)
Academic Radiology 35 4 J
Medical image computing and 34 4 S
computer-assisted intervention (MICCAI)
area because this field is older as research topic, while CBIR is more
recent. Thus, it could be natural that there are less high impact
specific periodicals for the latter subject. However, we noted that
the initial year of publication in high h5-index periodicals of the
papers included was 2009 for both topics (h5-index greater than
40).
The authors of the included studies focused their researches
mainly on descriptors development and analysis: 89% of the pa-
pers dealt with new descriptor proposals [16,17], or performance
analysis of different descriptors for a particular scope [18,19]. Thus,
we can conclude that the descriptors extraction and analysis is an
important task for both classification and retrieval techniques.
In medical image researches, it is possible to apply descriptors
in 2D or 3D domains. Here, we consider 3D descriptors only those
implemented for extracting features of 3D medical objects (with
surface or volume reconstructed). Descriptors applied to a set of
slices of the same volumetric object are considered 2D because
they usually process 2D information (area, contour, pixel gradient,
among others) [20,21].
Fig. 4 shows that papers related to classification and retrieval
of 3D medical objects has increased year by year from 2009 until
2013, as well as the number of specific papers presenting 3D
Fig. 2. Relationship between the score obtained in the quality criteria evaluation descriptors. However, two-dimensional descriptors remain impor-
and the h5-index distribution.
tant to this research area: about 30% of the 2D descriptors devel-
oped were applied mainly for slice retrieval [22] and alignment. In
succeed into publishing in scientific publication with higher h5- the case of brain images, it is common use 2D descriptors to detect
index. About 44% of classification papers were published in scien- tumors and Alzheimer disease [23,24,18]. Despite these numbers,
tific journals or symposiums with h5-index higher than 30, while some authors reported that 2D approaches can be highly time-
papers focused on CBIR techniques have about 21% of publications consuming, since it is necessary to analyze many slices to have the
in this h5-index range. Regarding these findings, our main hypoth- complete information about the problem under investigation [25–
esis is that there are more well cited periodicals in the classification 27].
Table 5
Descriptors categories according to [4] and their respective advantages and disadvantages.
Category Advantages Disadvantages Papers classification
Statistical-based - Faster, invariant to - Spatial information not [28–52,17,53,54]
rotation and robustness preserved
- Robust against noise - Small differences are not
distinguished
Geometrical-based - Good representation of - Difficulty in creating [55–59,47,60,17]
3D surface or voxels in efficient indexing
the object analyzed methods
- Easy implementation - Sensitive to small noise
Signal-based - Easy implementation - Requires pre-processing [61–68,16,69–71,59,72–77]
before feature extraction
Topology-based - Intuitive representation - Time-consuming [78,48,79]
of 3D object shape
structure
Appearance-based - High level description of - This information is not [80,29–33,38]
surface color and texture enough to differentiate
3D objects. Usually needs
to be complemented by
other descriptors
Visual-based - Reducing the -Time-consuming (one [81–87,20,88,27,89–104,18,105–
descriptors development 3D object could have 126,24,127–156,23,157–169,26,170–176]
difficulty, since this several projections/slices)
descriptor works with 2D
information
Fig. 4. Relationship between the descriptor type domain type studied in the papers
Fig. 3. Techniques distribution per h5-index value: CBIR and classification. selected and their respective year of publication.
data volume grown in the last decade, it is expected that the repos-
Due to medical imaging technology advances, volumetric ex- itories to store 2D and 3D objects have also increased. However, in
ams such as CT, MRI and PET modalities became more popular the medical image scenario we deal with information related to the
and, consequently, we observed a large increase of data to be patient health and their identity; there are hence ethical questions
stored. In this review, 35% of the papers used CT images, 36% that can make it difficult release information to allows validate the
results. This can explain the fact that 72% of the papers selected use
used MRI images, 11% PET, and 8% combined different types of
particular medical images, provided by some hospitals and health
image modalities. More than 85% developed techniques consider-
centers. This can hinder the comparison among the techniques
ing information from the 3D domain, indicating the relevancy of
developed by different research groups. From the included papers,
developing techniques that deal with this considerable volume of 42% studied a limited number of patients, not processing more
data generated by these image modalities in a fast and accurately than 50 cases [173,16,44]. Though we noted important databases
procedure. which are used as benchmarks for the studies, highlighting the
Another topic of interest in medical image classification and
retrieval is related to the database used in the researches. Since the
Alzheimer’s Disease Neuroimaging Initiative (ADNI),3 and the Lung of the human body represented), but we noted that 40% of the
Image Database Consortium (LIDC) [177]. papers using visual-based descriptors were focused on brain im-
Since the aim of medical images classification and retrieval sys- ages [87,88]. Some descriptors were used more frequently, such as
tems is commonly related to aid diagnosis, clinicians are an impor- Spherical Harmonics in the Signal-based category, the 3D Moment
tant piece during their development and evaluation. Meanwhile, Invariants in the Statistical-based category, and Wavelet Trans-
just 4% of the included papers reported evaluations performed form in the Visual-based category. These techniques are detailed
with experts [76,54]. Additionally, only 1% of the papers [153,127] in the next sections.
reported the use of the technique developed in the daily clinical
routine. 4.1. Spherical Harmonics
4. Descriptors Spherical Harmonics (SH) is a frequency-based space that rep-

resents functions over a sphere. The SH is defined as the angu-
Descriptors are important pieces for both retrieval and classi- lar portion of a set of solutions to Laplace’s equation using the
fication problems. They are algorithms that extract characteristics spherical coordinates of the 3D object, organized by their angular
of 2D or 3D objects in order to represent them in a manner that al- frequency. For the specific case of Laplace’s equation solution, the
lows posterior comparison. For 2D images, we have a well-known SH forms an orthogonal system [4].
taxonomy that establishes three categories: shape-based, texture- Eq. (1) represents this case, where Yℓm (θ, ϕ ) is the Laplace’s
based, and color-based. For 3D objects, many of the descriptors de- equation with respect to spherical coordinates (θ, ϕ ); m and ℓ are
veloped are focused on extracting shape and texture information, complex integer numbers, and f is a square-integrable function.
mainly related to 3D object’s structure and topology. Fig. 5 shows a set of SH possibilities related to the values of m and
Specifically for medical images, researchers have predomi- ℓ.
nantly worked with shape and texture features, but other ap- ℓ
∞ ∑
∑
proaches are available in the literature. In [4] a descriptors clas- f (θ, ϕ ) = fℓm Yℓm (θ, ϕ ) (1)
sification is provided, defining the following categories: ℓ=0 m=−ℓ
• Statistical-based: extracts sampling regions of the 3D objects In [70], the authors used SH to map the left ventricle structure,
and analyzes characteristics as curvature distribution, rela- reconstructed from SPECT images, into a sphere. Then, by using the
tionship among vertices, and moments of various orders; normalized and invariant sphere, they compare the left ventricles
• Geometrical-based: works directly with 3D objects data struc- in two different states (dilated and relaxed).
ture, and extracts information from their elementary unit The authors of [74] use the same idea to compare the volume
such as vertex, polygon or voxel. It also considers the spatial and shape in the amygdala–hippocampal between patients diag-
domain, since it can separate 3D objects in sets based on their nosed with schizophrenia and normal subjects. The researchers
coordinates. Descriptors in this category include distance discovered a closer correspondence between the shapes of the left
maps, ray-based, weighted point sets, among others; and right amygdala–hippocampal complex in normal controls than
• Signal-based: analyzes 3D objects from the frequency domain they observed in patients with schizophrenia.
perspective. Examples are Fourier descriptors and Spherical In [76] SH is applied to deformable 3D organs. In the first phase
Harmonics; of the process, the deformation subspaces are identified in the
• Topology-based: consists in a high-level representation of SH domain. Then, they represent the deformed surface by a block
how vertices are connected. The most common descriptors in sparse vector in a structured dictionary. The researchers compare
this category are based on graph data structure and skeletons; the method considering different aspects such as computing time
• Visual-based: the main idea of this category is mapping the and accuracy, reaching a precision greater than 70% using Haus-
3D object on a set of 2D images. This is a popular technique in dorff distance.
the medical field because a number of 3D medical objects are
composed of slices; 2D descriptors are less complex and there 4.2. 3D Moment Invariants
are many descriptors already developed for the 2D domain;
• Appearance-based: descriptors of this category extract fea- 3D Moment Invariants are statistical descriptors that calculate
tures based on surface color and texture. This category is nor- shape characteristics (such as area and volume) of a 3D object
mally applied together to shape descriptors to complement regarding its translation and rotation [28]. 3D Moment Invariants
the information. can be related to both volume and 3D surfaces. For both cases,
Eq. (2) is applied. This equation is directly inherited from 2D
Table 5 provides summarized information regarding the de- Moment Invariants, where M is the 3D object and the term Mpqr
scriptors classification found in the studies included. Statistical- is the (p, q, r)th 3D moment. Each 3D object has a unique set of 3D
based are predominant in the 3D domain. In [178], for example, moments [4].
the brain surface contours in Teichmüller space is used to analyze ∫ ∞ ∫ ∞ ∫ ∞
morphology alterations due to Alzheimer disease. Author of [38] Mpqr = xp yq z r f (x, y, z) dx dy dz (2)
proposes a Shape Context descriptor that calculates, for each vertex −∞ −∞ −∞
on the surface, the set of vectors originating from this vertex to In this context, in [42] this technique is applied in 3D bones
all the other vertices forming a unique configuration of the entire micro-CT images to detect the distribution of osteocyte lacunae re-
surface of this vertex. Texture and shape descriptors are found in sponsible to indicate several bones diseases. The authors compared
several works in the 2D domain. In [179] a good overview about their method with manual segmentation and statistical volume
descriptors in this domain is provided. calculation. According to them, results show that the automatic
No relationship was found between the descriptor used and the extraction using 3D Moment Invariants reached 92.8% of similarity
characteristics of the medical image (anomaly investigated or part comparing with the ground truth.
Also, the study [49] investigates the correlation between the
3 Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni. shapes of the cortical sulci and the sex and handedness of 142
usc.edu). brains exams generated by CT scan. They applied 12 first Moments
Fig. 5. Spherical Harmonics demonstration with complex constants variations.
3
Invariants to cortical sulci and found that moment invariant I33 4% [30,120,171]. In the next sections, we detail graphs and k-d
shows that the sulcus of the right-handed population is deeper trees, since they are more complex structures that can provide an
backwards, near the central sulcus than forward, in the frontal part, efficient way for storing features that must be quickly retrieved.
when compared to left-handed subjects. The authors also noted
a significant correlation between sex and the distribution of the
fourth moment invariant, in which the curvature associated to the
S-shape of the sulcus is more significant for females subjects.
4.3. 3D Wavelet Transform 5.1. Graphs-based indexing methods
The 3D Wavelet Transform decomposes an image into different

domains, such as frequency and time. Thus, the wavelet functions
A possible solution for indexing features of an image is based on
are powerful tools for signal processing [4]. It is broadly used in
graphs, as shown in Definition 1 [183]. This indexing method has
Visual-based descriptors to analyze the image texture, being useful
been employed since the image structure, as well as the character-
to detect anomalies in which the appearance of normal tissues
istics extracted, have been increasingly complex. In the 3D domain
is different from that of abnormal tissues, such as tumors and
for example, the information about the 3D object could be related
Alzheimer disease.
to different aspects, such as volume, area, spatial distribution of
The Wavelet Transform decomposes a 3D signal into four sub-
voxels, and vertices. Thus, this largest range of options generates
bands regarding to columns and row images. These filters create a
descriptors with more characteristics and manipulating this infor-
possible combination for each pair of column and row.
mation can be easier using graphs [4].
Eq. (3) presents the formulation of a wavelet coefficient of signal
In a graph, nodes and edges can be associated to attributes,
t Ψ (t), a refers to scaling and b is the time. The multiplication by
called labels. In the 3D domain, we can use the vertices/voxels to
the factor √1a normalizes the energy in different scales.
represent the graph nodes and the edges as spatial relationship
1
(
t −b
) between the nodes, for example.
Ψa,b (t) = √ Ψ a ̸ = 0, b ∈ ℜ (3)
a a
Definition 1. Let V (g) the vertices set of a graph g and E(g) the
In the included studies of the present review, the 3D wavelet edges set. A label function, l, maps a vertex or an edge to a label.
transform was used in different types of applications. In [21], The size of a graph is defined by the number of edges and denoted
authors investigate the interstitial lung disease in CT images us- as |g |. The set of graphs g generating a graph G.
ing gray level histograms and wavelet-based features such as B- G = {g1 , g2 , . . . , gn };
spline wavelet. Also, for the investigation of pulmonary embolism, g = {V1 (g), V2 (g), . . . , Vn (g), E1 (g), E2 (g), . . . , En (g)}
researchers of [180] use the Difference of Gaussians and visual
words in CT lung images from University Hospitals of Geneva.
They reported more than 60% of precision in the cases retrieved. To compute the dissimilarity distance between graphs, opti-
In [64] images from PET images of Royal Prince Alfred Hospital are mized search techniques, such as bubble sort, hashing and match-
processed by 3D Gabor wavelet filter and authors aimed to analyze ing functions are used [4]. Although these techniques are time-
different dementia diseases such as Alzheimer, vascular dementia consuming, they present good accuracy.
and frontotemporal dementia. An approach using graphs was applied in [32], which stored the
lung tumors information (volume, indication of malignity, among
5. Indexing methods others) in nodes, and their spatial distribution on PET scans images
is represented by edges. The authors reported an average precision
For indexing features, we found the traditional vectors applied of 60%–70% in the exams retrieval based on the position of lung
to 83% of the papers [181,70,182,28], while graphs represents tumors and their structure information.
5.2. K-d trees
A K-dimensional tree (k-d tree) is a complex type of indexing

structure, in which partitions of the data structure are in the n-
dimensional space [4]. In the image retrieval domain, each im-
age can be represented by a features vector, which can be high-
dimensional. Thus, each feature vector is mapped into a k-d tree
space, and the retrieval can be performed using specific algorithms,
such as round-robin [183].
A k-d tree divides the features vector space into (d-1)-
dimensional perpendicular hyperplanes to a specific coordinate
axis. Every node has at least two children and the manner in which
we structure the information can be customized [184]. We can, Fig. 6. SVM approaches: (a) represent a linear SVM; (b) represent a non-linear SVM.
for example, specify a threshold value to detect whether a new
features vector is a ‘‘child’’.
The nodes can be arranged according to some spatial infor-
mation of the 3D models, as proposed in [147], which stores the
information regarding curved shape of heart structures provided K (xi , xj ) = Φ (xi ) · Φ (xj ) (6)
by MRI images in the nodes. The goal of the authors is to retrieve Approaches using SVM are often cited in the studies included.
the medical slices most similar to the one given as query, ordering In [115] a SVM with radial basis function (RBF) kernel is used to
the database according to the slices similarity. detect abnormal shape and texture patterns in 2D brain MRI, and
in [75] SVM with RBF kernel is applied to classify supraspinatus
6. Similarity comparison methods muscle disorders.
Finally, the KNN is one of the simplest methods in the Machine
The technique used to compare the similarity among graphical Learning area, which aims to determine whether a new element
objects is also a key factor in classification and retrieval. These belongs to a specific class. It computes the distance between the
techniques have influence in the final results, since they define new element and the elements previously classified. The group of
how the objects must be compared, classified or clustered. In this elements with the shortest distance is chosen as the ‘‘owner’’ of the
context, some studies focus only on evaluating which similarity new element [4].
functions or classifiers best suit certain contexts [143,108,124]. Regarding the application of these methods to similarity com-
Some techniques were most frequently used by authors for parison between the feature vectors, both methods (Distances
measuring similarity. We can mention, for example, distances of functions and Machine Learning) are equally applied between 2D
the Lp space (Definition 2), which were used in 32% of the papers and 3D object analysis. In both techniques for retrieval information
such as Minkowski distances [185,186]. (CBIR and classification), Artificial Intelligence (AI) techniques with
the predominance of SVM are most used. They are present in 70%
Definition 2. The distance of order p between any two points of the papers classified in the CBIR group and in 85% of the papers
X = (x1 , x2 , . . . , xn ) and Y = (y1 , y2 , . . . , yn ) ∈ Rn is defined by: of the classification category. Some limitations noted concerning
( )1/p the application of AI techniques to compare the similarity among
n
∑
p feature vectors are related mainly to the need of combining several
Distancep = |xi − yi | (4)
parameters of the classifiers to optimize the result [68]. Addition-
i=1
ally, as highlighted in [93], the need of new processing is frequent,
When p = 1, Manhattan Distance is defined, while p = 2, Euclidean since the SVM classifier training dataset must be recalculated for
Distance is defined and when p reaches infinity, we obtain the each new object introduced in the database.
Chebyshev distance.
7. Databases
Machine Learning-based techniques are also commonly used
in retrieval and classification methods. In this category, we can An important challenge for studies related to medical images
highlight the Support Vector Machine (SVM) and the K-Nearest is how to obtain the necessary data to evaluate techniques. Be-
Neighbor (KNN) implemented in 23% and 3% of the papers in- sides the medical images themselves, information about specific
cluded, respectively [34,155]. diseases (medical reports, and indication of specific slices, for
SVM aims to generate dichotomies, i.e., separating example) must to be obtained and this can make the evaluation
n-dimensional data (x1 , x2 , .., xn ) into well defined classes through of techniques a very difficult task. Thus, public medical images
at least one hyperplane. SVM can be classified as linear or non- databases are very important, but their formation is not easy.
linear. When there is a linear SVM, it is possible to split the set of As aforementioned in Section 3, few included papers (28%) cited
data into separated groups. Classifiers that allow such division are the medical image databases used for evaluating their approaches.
illustrated in Fig. 6(a). They are defined by Eq. (5), where w · x is Most of them worked with specific sets of images provided by
the scalar multiplication between vectors w and x, w is the normal partners (hospitals and medical experts). When the papers cited
vector in relation to the hyperplane and b is the bias [187,188]. that the research used medical image databases, we verified a pre-
Fig. 6(b) presents an example of non-linear SVM. In this approach, dominance of public databases such as ADNI [189], which provides
the data types cannot be separated by the linear approach. Thus, it brain MRI and PET images, and LIDC [177], which provides lung CT
is necessary to map the original data into a higher space using the and Computed Radiography (CR) images. Table 6 presents all the
inner product between any points of original data (xi , xj ), as shown public medical databases found and the respective studies where
in Eq. (6). they were cited.
The databases presented in Table 6 are available for everyone
+1 if w · xi + b > 0,
{
yi = (5) and some of them contain cases of specific diseases (Alzheimer
−1 if w · xi + b < 0. and Oncology problems, for example), specific human organs, and
Table 6
List of medical images databases used in the included papers.
Database Description Images modalities Papers
Alzheimer’s Disease Exams of Alzheimer’s disease MRI and PET images, genetics, [126,39,65,40,66,190]
Neuroimaging Initiative patients, mild cognitive cognitive tests, CSF and blood
(ADNI) impairment subjects and elderly biomarkers
controls.
BrainWeb [191] Set of realistic brain MRI data MRI images [142,87,22,89,172]
volumes produced by a MRI
simulator, and images of Multi
Sclerosis disease.
Diabetic Retinopathy Benchmark of diabetic Digital fundus camera [99,114]
Database [192] retinopathy detection from
digital images.
CLEF Cross Language Image Different images (generic and MRI, CT, X-ray, CR, SPECT and [109,174,165]
Retrieval Track medical) for computational PET
(ImageCLEF)[193] challenges to evaluate the best
techniques for image retrieval
task including segmentation,
filtering and others image
processing techniques.
Image Retrieval in Medical Focused on CBIR of X-ray [140,153,165]
Applications (IRMA)[194] radiographies.
Lung Image Database Lung lesions. CT [45,34,111,31,157,19,182]
Consortium (LIDC)[195]
National Health and Surveys into USA population MRI, CT, X-ray, PET and SPECT [119,173]
Nutrition Examination health and habits besides
Survey (NHANES)[196] medical exams.
OncoPet [197] Distributed database of realistic PET images [29]
simulated whole body PET
images focused on Oncology.
specific image modalities (PET, MRI or CT). Some databases are

more generic, but usually medical reports are not available, such
as IRMA [194] and ImageClef [193], which hinder the evaluation
of techniques and the comparison among different approaches.
Another important database is the Cancer Imaging Archive [198]
used in [199], which since 2014 groups several databases to help
in cancer diagnosis through 2D and 3D object analysis. In this
repository we can found images related to different organs and
diseases such as breast, brain and prostate cancer.
8. Applications
The scenario of medical images classification and retrieval em-

phasizes different subjects. Techniques are applied to differentiate
human organs, structures or diseases. Fig. 7 illustrates the dis-
tribution of the included papers. Some considerations about this
distribution must be highlighted.
Firstly, brain and lung images are the most frequent subjects of
study. Brains were analyzed in 48% , lung in 17% and heart in 21% of
the included papers. More specifically, most of the papers aimed at
Fig. 7. Disease distribution per human body structure.
identifying and/or classifying Alzheimer disease using MRI brain
images, followed by detection of lung tumors. Both human body
structures have public medical database available (LIDC and ADNI).
In addition to brain and lung images, we found several works
dealing with heart, intestinal and bones structures. For heart im- Related to intestinal medical images analysis, authors are fo-
ages the most studied topic aims to analyze the heart structure as a cuses in disease classification and retrieval. An example is the
whole. In [122], for example, the objective is to quantify global and study [200], which applies a texture descriptor based on gray level
regional left ventricular systolic function to discriminate segments co-occurrence matrix to classify suspicious regions that present
showing an abnormal regional wall motion. The researchers used polyps. The authors achieved more than 75% of average accuracy,
statistical shape modeling technique getting an area under a ROC using a set of 20 patients exams.
curve of 0.87. In [46], the authors applied a set of shape descrip- A common problem investigated considering all types of images
tors (Regional Surface Changing, Global Surface Curvature, Surface is the structure of the human organs. In this context, in [201] the
Distance, Normalized surface distance, and Effective Radius) to
quantify the left ventricle structure.
Table 7
Evaluation metrics cited in the studies included (TP: True Positive, FP: False Positive,
TN: True Negative, TP: True Positive, di : distance between ranks, R: sum of the ranks,
n: sample size, cov: covariance, σ : standard deviation).
Metric Equation
tp
Precision tp+fp
tp
Recall or Sensitivity tp+fn
tp+tn
Accuracy tp+fp+tn+fn
tp
Specificity fp+tn
Precision∗Recall
F-measure 2∗ Precision+Recall
d2i
∑
6
Spearman’s rank 1− n(n2 −1)
ni (ni +1)
Mann–Whitney U test Ri − 2
cov (X ,Y )
Pearson Correlation σx σy
have to critically analyze this knowledge domain to identify the

trends and challenges to be overcome.
In a general analysis, the following points were observed: in
relation to the quality of included studies, the h5-index of the
Fig. 8. Relationship between the evaluation metric analyzed in the included papers papers included is higher for Journals than for Conferences, and it is
and their respective distribution per technique. also higher for classification problem articles than for CBIR articles.
There is a predominance of proprietary medical databases for tests
as well as a great difficult in performing tests with a considerable
number of cases. However, we found some papers working with
a large set of images and 3D objects, mainly from lung and brain
authors analyze the topology of 3D vascular trees. In the same con- focused in cancer and dementia diagnosis, which can contribute to
text, in [202] the lateral pterygoid muscle deformation is detected the greater interest in these knowledge fields. Public databases in
and quantified based on mandible MRI exams. these areas facilitate the techniques evaluation.
From the analysis conducted, advances are observed in classi-
9. Evaluation metrics fication retrieval of 3D medical objects and images. The gaps we
identified in the literature analyzed can be explored as opportuni-
Methods to evaluate and to compare techniques to store and ties for new research lines, as described below.
to retrieve medical images are essential to validate the results
found and to verify if the research hypotheses can be accepted or 10.1. General definitions
rejected. Fig. 8 shows an overview of the evaluation metrics and
tests performed by the included papers. Table 7 presents a list of The first item to point out regards to the differentiation be-
equations of the most frequent evaluation metrics found. tween CBIR and classification problems. In some studies authors
Accuracy is the most used among all the metrics (25% of the state that classification can be associated to CBIR since a classifier
studies), mainly cited in papers of the classification category. In can be applied to measure the similarity to improve the system
the studies in the CBIR category, Precision vs. Recall is the most accuracy [203]. However, the CBIR main goal remains: retrieving
widely used, presented in 51% of the papers. Considering the total the most similar images to an image given as a query parameter,
of papers, Precision vs. Recall is present in 20%. which can imply several images as a result. However, in our study,
Another representative metric is specificity, used mainly in we found many papers focused only in the classification problem,
papers of the classification category, representing 15% of all the without the concept of query by example and retrieval of similar
metrics. Finally, the ROC curve is also frequent in classification images as a result. These papers only analyze how the images
problems. Even though some papers used Specificity and Sensi- can be classified, usually using two classes: normal and abnormal
tivity metrics, most of researchers does not use these metrics to cases. All these approaches are being considered in our mapping.
compose ROC curve, since we just found this metric in only 5% of Therefore, this is one of the challenges of the area: to standardize
the papers [45,116,120]. the nomenclature of these techniques both for avoiding misunder-
In addition to these metrics, we also found some non- standing and for finding related studies when a research is being
parametric tests, such as Mann–Whitney and Spearman’s rank [49, conducted.
28], as well as parametric ones, such as Pearson Correlation [86], Fig. 4 shows an interesting behavior related to the amount of
but they were not frequent enough to be presented in the graph of publications that considers medical or 3D object image processing:
Fig. 8. there is a decrement number of publications after 2014, but it is
wrong to affirm that this knowledge domain is losing relevance.
10. Discussion As shown in [204], this subject is very active research field and
authors pointed out some future trends such as the application
From the data presented in the previous sections, the classifi- of multi-features to discriminate properly the image analyzed and
cation and retrieval of medical images can be observed to have the performance required for classification and retrieval process.
increased over the years, producing numerous studies. Besides Also, in [11] it is highlighted the ‘‘new era’’ of deep learning and
understanding which techniques are applied to which goals, we how this concept is few applied mainly on CBIR domain, which
opens up new opportunities in future coming. We observed that to handle this variety. Also, the shape variety of specific diseases
papers related to the utilization of deep learning in CBIR started represents an additional challenge to implement a ‘‘standard’’ de-
on 2007 and was more disseminated after 2015, at 2017 we noted scriptor for 3D medical objects [4].
more than 66 papers dealing with deep learning techniques CBIR To deal with the time-consuming problem related to 2D slices,
problems. However only in 2013 we found studies that applied several techniques have been developed aiming to extract dis-
this approach in medical images. In 2017, just 14 papers were tinctive information from several slices, mainly associated to the
published considering deep learning for CBIR applied in medical shape and texture of the entire 3D medical object. In spite of this
images and deep learning.4 The result presented above indicates approach has several studies with good results, some limitations
that deep learning and AI-based techniques ‘‘ wave’’ are coming to persist, such as the need of defining a Region of Interest (ROI),
the medical image domain and it is very plausible that in the next the need of techniques to apply shape, texture or color descriptors
months/years we note a increment on this topic in the publications to different image resolutions, parametrization to run segmenta-
related to CBIR and classification. tion techniques and problems related to slices alignment, which
Beyond the trends identified by [11], we analyzed several works may influence the results [132,135,165,140,143,148,93,155]. In
on this Systematic Mapping and we did not find studies which this knowledge domain, the 3D texture descriptor is pretty much
deal with both 3D objects (reconstructed and set of slices). Con- used and the work of [206] performed a good review of the litera-
sequently, it was not possible to verify if the time to process the ture focused on 3D solid texture in medical images.
queries and the efficiency were better for any specific approach. Still about descriptors, this Systematic Mapping not covered
The evaluation of these two approaches, considering the same dis- semantic descriptors which is a relevant sub-domain of classifica-
ease or structure, could allow different tests of time performance tion and retrieval techniques. This type of descriptor labels the 2D
and retrieval systems efficiency. However, this type of evaluation images and 3D object regions accordingly to the innate humans
requires descriptors development, similarity functions definition, visual understanding about this region. In 3D domain this type
as well suitable evaluation metrics for both 2D and 3D domains, of descriptor is used connected with several AI-based techniques
besides techniques for processing both types of objects. This is not such as neural networks, and SVMs [11].
a trivial task, but could generate important results for the area.
10.2. Papers quality

10.4. Indexing methods
We verified that it is usual for the same group of researches
to have more than a paper with the same object of study, just
changing one descriptor or a similarity comparison method for As mentioned in Section 3, we noted a predominance of tradi-
each paper. The evaluation method and the cases are usually the tional vectors during the features indexing. However, other index-
same. This hinders understanding the whole problem under in- ing structures such as graphs and K-d trees are offering promising
vestigation, since different papers can be in different databases results.
and some of them may be missed. Additionally, partial results are The authors that used these alternative data structures point
usually published in scientific literature of smaller impact and the out that the biggest advantage is due to the possibility of offering, in
article is often not complete, which can result in an excluded paper addition to the features characteristics organization and indexing,
in researches as the one presented here. Thus, publishing complete the spatial information about the region where these features were
papers instead of prioritizing the number of publications is still a extracted and, in some cases, how each feature is associated to each
challenge to be overcome not only in this field of investigation, but other [30,120].
in all research areas. Besides these data structures give more possibilities related to
Papers that evaluated their techniques with experts were in- the features data organization, they offer new methods to compute
terestingly published in Journals and Conferences with high h5- the dissimilarity between them such as flow-based dissimilarity
index (average of h5-index equal to 35). This can be indicative that methods (min-cut and maximum flow, for example) and neural
this type of evaluation can add value to the research. However, networks [207] .
interdisciplinary research requires extra-efforts such as the proper
involvement and commitment of all members of the research
group in the project, establishing protocols and methods for eval-
uating the techniques developed. Moreover, the result found in 10.5. Similarity comparison
this Systematic Mapping highlights the benefits of evaluation with
experts: publications with greater relevance, more robust tests
and more accurate evaluations. Also, this kind of questioning was Minkowski distance was the most used similarity function an-
discussed in [204] which pointed out the importance of bring the alyzed the included papers, followed by SVM technique. This in-
end-user to the final evaluation of CBIR systems. formation is coherent due to the fact that the most used indexing
methods was traditional vectors and Minkowski distance is very
10.3. Descriptors common function used to compare this type of data structure.
However we noted some limitation in both approaches. The first
one, related to Minkowski distance, is based on the fact that this
The greatest advantage of processing objects in the 3D domain
technique is noise-sensitive, i.e, if the features vector has outliers in
is the larger amount of information that they can provide, mainly
some features, this information will impact negatively the results.
related to the 3D object shape. However, it can be also the weak-
Additionally it is not possible extract local information from the
ness of this approach, since not all the 3D representations have well
similarity result (unless you divide the features vectors in small
defined geometry. In addition, the variety of rotation, translation
pieces associated to different regions of the 3D object analyzed).
and scale properties of 3D object requires preprocessing methods
Regarding SVM, the main limitation is the fact of this technique
to align the 3D object or development of more complex descriptors
demands several parametrizations and a training dataset, normally
the research that used this approach performs the algorithm eval-
4 This search considered the IEEE Xplore Library [205]. uation in a very controlled scenario. When the system receives
a different input: new disease to be investigated or images from 10.8. Evaluation metrics
different body structures, the system needs a complete new set of
adjustments. This frequent changes on the parametrization is the For evaluation metrics we verified that Precision vs. Recall is the
major disadvantage of this approach. most used metric for evaluation of CBIR systems and accuracy for
classification studies. These metrics are very consolidated on this
10.6. Databases knowledge domain and could provide a good evaluation related to
the techniques performance.
We verified that more than 70% of the included papers did not However, if we consider the clinicians’ evaluation of these tech-
report the source of medical images they used. We cannot conclude niques developed, we noted only few papers reporting it, as men-
if this is miss information during the methodology description or if tioned in Section 3. This raises a question: why? Time and schedule
the medical images are from particular databases, since we noted problems of clinicians? Evaluation with health professionals is
a strong relationship between Medical Schools and Computer Sci- not considered important by the researchers who developed the
ence research groups. techniques?
However, we noted there are important public databases, as We believe that it is a problem related to synergy between the
mentioned in Section 7, such as LUNG Image Database [195], two different groups (health experts and designers) than a wrong
ADNI [189] and Cancer Imaging Archive [198], which provide a impression that is not important the clinicians evaluation. It is clear
larger number of different types of medical images examinations for us, based on the paper quality information, that authors that
for different diseases. Coincidentally, as we noted in Section 8, the provide results with both evaluation (traditional metrics and clin-
most diseases investigated in the included papers belongs to one icians evaluation) publish the results in more qualified scientific
of these public databases. literature.
Despite not being a determinant factor to increase the re-
searches in the area, this could favor the evaluation of the tech- 10.9. Opportunities
niques as well as the comparison of results with those obtained by
other researches.
In summary, the main challenges and opportunities in this field
can be briefly listed as follows:
10.7. Applications
• optimizing the parametrization and indexing methods to
We did not find a relationship between the computational tech- provide faster search/classification. This includes develop-
nique (CBIR, classification) and specific parts of the human body ing methods with less parametrization needs and indexing
or a specific disease. All these categories are equally distributed methods suitable to problems dealing with local and global
in the selected papers. Yet we found a relationship between the characteristics;
some diseases, such as Alzheimer and Lung tumors and the devel- • understanding the anatomy of human organs as well as their
opment of texture descriptors for 2D slices. As it is known, these deformation due to a specific disease, and designing suitable
diseases are investigated mainly through CT examinations, which descriptors to aid the diagnosis. Shape and texture descrip-
generate slices with high resolution and very detailed images about tors have a great potential for clinical problems. We have a
these body structures. Thus, researches of this expertise, identified good number of diseases that cause shape deformation in the
that texture descriptors are being efficient to identify the diseases some organ structures. Anomalies, such as aneurysm, infarc-
patterns on those images, since it can be found analyzing the tion, and even tumors can be analyzed as benign and malig-
homogeneity, coarseness and granularity of the region. nant based on their shape. Other diseases, such as Alzheimer,
In our study, we also investigated information about the adher- do not cause shapes deformation but visually we note differ-
ence of the industry to the techniques for medical image classifi- ent patterns on the examinations, which can be differentiated
cation and retrieval. We also noted only few papers reporting that through texture descriptors.
the techniques were applied in clinical routine. Our hypothesis is • exploring the local perspective of descriptors in the 3D do-
firstly related to the difficulty in approving computational systems main. Global and local descriptors have their own advantages
to be used in daily Healthcare routine or turn them into commercial and disadvantages, mainly related to noise sensitivity. From
products. Secondly, the absence of this type of information in the our perspective, descriptors that act locally and, at the same
studies can be related to privacy restrictions, but the papers do time, support some noise in the information, perform well
not mention that. Finally, the authors may still be dealing with in several problems. This kind of technique is still underex-
technical problems, such as the need to improve their techniques, plored in 3D medical objects and images classification and
involving parametrization dependent on medical images sets, ROIs retrieval.
extraction, spatial and contrast resolutions. • developing or improvement of comparison methods are di-
These questions can generate non-intuitive systems, which are rectly related to system performance, and we noted that
difficult for health experts to use. However, it is worth highlighting AI-based methods are being used more extensively in the
that in many papers we found this aspiration — the method being last years. Besides these techniques improve the system ef-
applied to clinical daily routine as ‘‘future work’’. We believe that ficiency, we noted a common concern about the amount
through a development of protocols, pilot studies and support of of parametrization and the necessity of training datasets to
business incubators, this goal have some advances in this scenario adjust properly those techniques. In addition, as we previ-
in the next years. ously discussed, these several adjustments could impact on
In the mean time, we are observing the advances in medical im- the system utilization by end-users. Thus, it is very welcome
ages modalities, such as MRI, CT, PET, SPECT and Ultrasound. Thus, researches to deal with optimization or automation of these
a possibility for research is changing our mindset about the design parameters.
of systems to aid the diagnosis, which should be an intelligent • developing new indexing methods we note a predominance
information provider in the machine while performing the image of traditional vectors in the included paper. Despite this, some
acquisition. This approach favors the development of embedded authors identified that others structures can be a good as an
systems and some companies have heavily spent resources in this alternative, such as graphs and k-d trees. Thus, it is important
direction [2]. and an opportunity, mainly for 3D domain that provides a
large amount of information. Testing and validating alterna- of Science and Technology Medicine Assisted by Scientific Comput-
tive indexing structures could improve the systems perfor- ing.
mance.
• improving medical databases availability, which is crucial for References
comparing the techniques;
• conducting studies for develop methodologies to improve the [1] J. Gantz, D. Reinsel, The digital universe in 2020: Big data, bigger digital
synergy between research groups, aimed to reach end-to- shadows, and biggest growth in the far east, IDC iView: IDC Anal. Future 2007
(2012) 1–16.
end evaluation both for computational and end-user perspec-
[2] N.T. Issa, S.W. Byers, S. Dakshanamurthy, Big data: the next frontier for
tives. innovation in therapeutics and healthcare, Expert Rev. Clin. Pharmacol. 7 (3)
(2014) 293–298.
11. Conclusion [3] R. Hermon, P.A. Williams, Big data in healthcare: What is it used for? in: 3rd
Australian eHealth Informatics and Security Conference, 2014, pp. 40–49.
[4] F. Yu, Z. Lu, H. Luo, P. Wang, Three-Dimensional Model Analysis and Pro-
This Systematic Mapping aimed to provide a general overview cessing, Springer Science & Business Media, 2011, p. 500, http://dx.doi.org/
of studies aimed to classify and to retrieve medical 3D objects, 10.1007/978-3-642-12651-2.
besides a critical view about some aspects of these papers. We [5] I.N. Bankman, S. Morcovescu, Handbook of medical imaging. Processing and
analysis, Med. Phys. 29 (1) (2002) 107.
noted a growing development of methods for 3D objects, but
[6] H. M"uller, N. Michoux, D. Bandon, A. Geissbuhler, A review of content-based
many of them are adaptations of techniques from the 2D domain, image retrieval systems in medical applications–clinical benefits and future
for both descriptors and methods for similarity comparison and directions, Int. J. Med. Inf. 73 (1) (2004) 1–23.
evaluation. Therefore, we concluded that specific descriptors and [7] R. Datta, D. Joshi, J. Li, J.Z. Wang, Image retrieval: Ideas, influences, and
specific evaluation methods for the 3D domain are still a gap to be trends of the new age, ACM Comput. Surv. 40 (2) (2008) 5:1–5:60, http:
//dx.doi.org/10.1145/1348246.1348248.
covered in this area.
[8] C.B. Akg"ul, D.L. Rubin, S. Napel, C.F. Beaulieu, H. Greenspan, B. Acar, Content-
We also found few public databases, with limited scope con- based image retrieval in radiology: current status and future directions, J.
cerning diseases represented in the medical images as well as Digit. Imaging 24 (2) (2011) 208–222.
modalities included in the database. The establishment of re- [9] K.H. Hwang, H. Lee, D. Choi, Medical image retrieval: past and present,
searchers networks aiming at forming public databases with Healthc. Inf. Res. 18 (1) (2012) 3–9.
enough information to allow complete evaluations of techniques [10] A. Kumar, J. Kim, W. Cai, M. Fulham, D. Feng, Content-based medical image
retrieval: a survey of applications to multidimensional and multimodality
constitute an opportunity still underexplored in the literature. data, J. Digit. Imaging 26 (6) (2013) 1025–1039.
The low adherence of health professionals – or the lack of this [11] G. Litjens, T. Kooi, B.E. Bejnordi, A.A.A. Setio, F. Ciompi, M. Ghafoo-
information in the papers included – indicates that another oppor- rian, J.A.W.M. van der Laak, B. van Ginneken, C.I. Sánchez, A survey on
tunity is gathering researches aiming at developing new processes deep learning in medical image analysis, Med. Image Anal. 42 (Suppl. C)
to transfer technology to the industry. (2017) 60–88, http://dx.doi.org/10.1016/j.media.2017.07.005, URL http://
www.sciencedirect.com/science/article/pii/S1361841517301135.
This study aimed to answer some questions, presented in Sec- [12] M. Petticrew, H. Roberts, Systematic Reviews in the Social Sciences: A
tion 1. Some of them were answered along the paper, but we would Practical Guide, John Wiley & Sons, 2008.
like to highlight those more specific here: [13] J. Bailey, D. Budgen, M. Turner, B. Kitchenham, P. Brereton, S. Linkman,
Evidence relating to object-oriented software design: A survey, in: Empirical
1. are the researchers producing relevant studies in this area? Software Engineering and Measurement, 2007, ESEM 2007, First Interna-
Yes. As discussed in Section 3, most of the 107 papers in- tional Symposium on, 2007, pp. 482–484, http://dx.doi.org/10.1109/ESEM.
cluded were published in periodicals and conferences with 2007.58.
[14] T. Reuters, J. Cit. Rep. (2015)
high h5-index;
[15] A.-W. Harzing, R. Van Der Wal, A google scholar h-index for journals: An
2. where the researchers are publishing? As mentioned in alternative metric to measure journal impact in economics and business, J.
Section 3, the publication are divided almost equally be- Assoc. Inf. Sci. Technol. 60 (1) (2009) 41–46.
tween conference and journals. Both conferences and jour- [16] A.B. Abdallah, F. Ghorbel, K. Chatti, H. Essabbah, M.H. Bedoui, A new uni-
nals present high impact; form parameterization and invariant 3D spherical harmonic shape descrip-
tors for shape analysis of the heart’s left ventricle – A pilot study, Pat-
3. has the Healthcare area really benefited from the classifi-
tern Recognit. Lett. 31 (13) (2010) 1981–1990, http://dx.doi.org/10.1016/j.
cation and retrieval knowledge field? Yes. Although only patrec.2010.06.009, URL http://www.sciencedirect.com/science/article/pii/
few studies reported applying the techniques developed S0167865510001960.
into daily medical routine (we pointed out some reasons in [17] Y. Su, L. Zhong, C.W. Lim, D. Ghista, T. Chua, R.-S. Tan, A geometrical approach
Section 10) we noted a strong partnership between Medical for evaluating left ventricular remodeling in myocardial infarct patients,
Comput. Methods Programs Biomed. 108 (2) (2012) 500–510, http://dx.
Schools and research groups (since many experiments were doi.org/10.1016/j.cmpb.2011.03.008, URL http://www.sciencedirect.com/
made with images provided by internal collaboration and science/article/pii/S0169260711000733.
not by public medical databases, as discussed in Section 7). [18] W. Wan Ahmad, M.F.A. Fauzi, Comparison of different feature extraction
This can indicates the interest of both areas in the research. techniques in content-based image retrieval for CT brain images, in: Mul-
timedia Signal Processing, 2008 IEEE 10th Workshop on, 2008, pp. 503–508,
Finally, we highlight the importance of this area for the next http://dx.doi.org/10.1109/MMSP.2008.4665130.
years, as complex data and information will be more explored and [19] J.R. Ferreira, P.M. de Azevedo-Marques, M.C. Oliveira, Selecting relevant 3D
image features of margin sharpness and texture for lung nodule retrieval,
Healthcare is considered a mainstream sector to be benefit from
Int. J. Comput. Assist. Radiol. Surg. 12 (3) (2017) 509–517, http://dx.doi.org/
that. Therefore, this mapping intends to provide a contribution 10.1007/s11548-016-1471-7.
to help present and future researchers to quickly locate general [20] P.V. Ingole, K.D. Kulat, A chain code based fuzzy features for brain MRI
information about this topic. retrieval, in: Emerging Trends in Engineering and Technology, ICETET,
2012 Fifth International Conference on, 2012, 185–190, http://dx.doi.org/10.
1109/ICETET.2012.50.
Acknowledgments [21] A. Depeursinge, A. Vargas, A. Platon, A. Geissbuhler, P. Poletti, H. Müller, 3D
Case-Based Retrieval for Interstitial Lung Diseases, Springer, Berlin, Heidel-
berg, 2010, pp. 39–48, http://dx.doi.org/10.1007/978-3-642-11769-5_4.
This work is supported by Brazilian National Council of Sci-
[22] A. Quddus, O. Basir, Wavelet based MR 2D slice retrieval in 3D volumes,
entific and Technological Development, the - Brazilian Federal in: Science and Technology for Humanity, TIC-STH, 2009 IEEE Toronto In-
Agency for Support and Evaluation of Graduate Education (CAPES), ternational Conference, 2009, pp. 545–550, http://dx.doi.org/10.1109/TIC-
Sao Paulo Research Foundation (FAPESP) and the National Institute STH.2009.5444439.
[23] Y. Zhuang, N. Jiang, Z. Wu, Q. Li, D.K.W. Chiu, H. Hu, Efficient and robust large [41] H. Khosravi, M. Mohammadkhani, H. Soltanian-Zadeh, Sparse MEG source
medical image retrieval in mobile cloud computing environment, Inform. localization on parameterized cortical medial surface, in: Biomedical Engi-
Sci. 263 (0) (2014) 60–86, http://dx.doi.org/10.1016/j.ins.2013.10.013, URL neering, ICBME, 2012 19th Iranian Conference of, 2012, pp. 107–112, http:
http://www.sciencedirect.com/science/article/pii/S002002551300738X. //dx.doi.org/10.1109/ICBME.2012.6519667.
[24] J. Chen, J. Tian, Real-time multi-modal rigid registration based on a novel [42] P. Dong, S. Haupert, P.-J.. Gouttenoire, F. Peyrin, Efficient extraction of 3D
symmetric-SIFT descriptor, Prog. Nat. Sci. 19 (5) (2009) 643–651, http:// bone cells descriptors from micro-CT images, in: Biomedical Imaging, ISBI,
dx.doi.org/10.1016/j.pnsc.2008.06.029, URL http://www.sciencedirect.com/ 2013 IEEE 10th International Symposium on, 2013, pp. 1492–1495, http:
science/article/pii/S1002007109000240. //dx.doi.org/10.1109/ISBI.2013.6556817.
[25] Y.-Y. Liu, M. Chen, H. Ishikawa, G. Wollstein, J.S. Schuman, J.M. Rehg, Au- [43] X. Gao, Feature wise representation for both still and motion 3D medical im-
tomated macular pathology diagnosis in retinal {OCT} images using multi- ages, in: Image Analysis and Interpretation, SSIAI, 2014 IEEE Southwest Sym-
scale spatial pyramid and local binary patterns in texture and shape encod- posium on, 2014, pp. 1–4, http://dx.doi.org/10.1109/SSIAI.2014.6806014.
ing, Med. Image Anal. 15 (5) (2011) 748–759, http://dx.doi.org/10.1016/j. [44] L.C.C. Bergamasco, F.L.S. Nunes, A new local feature extraction approach
media.2011.06.005, URL http://www.sciencedirect.com/science/article/pii/ for content-based 3d medical model retrieval using shape descriptor, in:
S1361841511000922. Proceedings of the 29th Annual ACM Symposium on Applied Computing,
[26] R. Munbodh, H.D. Tagare, Z. Chen, D.A. Jaffray, D.J. Moseley, J.P.S. Knisely, SAC ’14, ACM, New York, NY, USA, 2014, pp. 902–907, http://dx.doi.org/10.
J.S. Duncan, 2D-3D registration for prostate radiation therapy based on a 1145/2554850.2554873.
statistical model of transmission images, Med. Phys. 36 (10) (2009) 4555– [45] W.-J. Choi, T.-S. Choi, Automated pulmonary nodule detection based on
4568. three-dimensional shape-based feature descriptor, Comput. Methods Prog.
[27] S. Roy, Y. Chi, J. Liu, S.K. Venkatesh, M.S. Brown, Three-dimensional spa- Biomed. 113 (1) (2014) 37–54, http://dx.doi.org/10.1016/j.cmpb.2013.08.
tiotemporal features for fast content-based retrieval of focal liver lesions, 015.
IEEE Trans. Biomed. Eng. 61 (11) (2014) 2768–2778, http://dx.doi.org/10. [46] J.-R. Duann, S.-H. Chiang, S.-B. Lin, C.-C. Lin, J.-H. Chen, J.-L. Su, Assessment
1109/TBME.2014.2329057. of left ventricular cardiac shape by the use of volumetric curvature analysis
[28] M.E. Gonzalez, K. Dinelle, N. Vafai, N. Heffernan, J. McKenzie, S. Appel- from 3D echocardiography, Comput. Med. Imaging Graph. 23 (2) (1999) 89–
Cresswell, M.J. McKeown, A.J. Stoessl, V. Sossi, Novel spatial analy- 101, http://dx.doi.org/10.1016/S0895-6111(98)00065-2, URL http://www.
sis method for {PET} images using 3D moment invariants: Applica- sciencedirect.com/science/article/pii/S0895611198000652.
tions to Parkinson’s disease, Neuroimage 68 (0) (2013) 11–21, http://dx. [47] B. HajGhanbari, G. Hamarneh, N. Changizi, A.D. Ward, W.D. Reid, MRI-
doi.org/10.1016/j.neuroimage.2012.11.055, URL http://www.sciencedirect. based 3D shape analysis of thigh muscles: Patients with chronic ob-
com/science/article/pii/S1053811912011676. structive pulmonary disease versus healthy adults, Acad. Radiol. 18 (2)
[29] A. Kumar, J. Kim, D. Feng, M. Fulham, Graph-based retrieval of multi- (2011) 155–166, http://dx.doi.org/10.1016/j.acra.2010.09.008, http://www.
modality medical images: A comparison of representations using simulated sciencedirect.com/science/article/pii/S1076633210005179.
images, in: Computer-Based Medical Systems (CBMS), 2012 25th Interna- [48] A. Lauric, E. Miller, S. Frisken, A.M. Malek, Automated detection of intracra-
tional Symposium on, 2012, http://dx.doi.org/10.1109/CBMS.2012.6266295. nial aneurysms based on parent vessel 3D analysis, Med. Image Anal. 14 (2)
[30] A. Kumar, J. Kim, L. Wen, D. Feng, A graph-based approach to the retrieval of (2010) 149–159, http://dx.doi.org/10.1016/j.media.2009.10.005, URL http:
volumetric PET-CT lung images, in: Proceedings of the Annual International //www.sciencedirect.com/science/article/pii/S1361841509001212.
Conference of the IEEE Engineering in Medicine and Biology Society, EMBS [49] J.-F. Mangin, F. Poupon, E. Duchesnay, D. Rivière, A. Cachia, D.L. Collins,
2012, pp. 5408–5411, http://dx.doi.org/10.1109/EMBC.2012.6347217. A.C. Evans, J. Régis, Brain morphometry using 3D moment invariants,
[31] A. Kumar, F. Nette, K. Klein, M. Fulham, J. Kim, A visual analytics approach Med. Image Anal. 8 (3) (2004) 187–196, http://dx.doi.org/10.1016/j.
using the exploration of multi-dimensional feature spaces for content-based media.2004.06.016, URL http://www.sciencedirect.com/science/article/pii/
medical image retrieval, IEEE J. Biomed. Health Inf. PP (99) (2014) 1, http: S1361841504000210.
//dx.doi.org/10.1109/JBHI.2014.2361318. [50] A. Mishra, P.K. Dutta, M.K. Ghosh, Fuzzy shape based motion evaluation
[32] A. Kumar, J. Kim, M. Fulham, D. Feng, Efficient PET-CT image retrieval using of left ventricle using genetic algorithm, Image Vis. Comput. 24 (5) (2006)
graphs embedded into a vector space, in: Engineering in Medicine and 436–446, http://dx.doi.org/10.1016/j.imavis.2006.01.010, URL http://www.
Biology Society, EMBC, 2014 36th Annual International Conference of the sciencedirect.com/science/article/pii/S0262885606000436.
IEEE, 2014, pp. 1901–1904, http://dx.doi.org/10.1109/EMBC.2014.6943982. [51] L.C. Bergamasco, F.L. Nunes, {A}pplying {d}istance {h}istogram to retrieve
[33] J. Feulner, S. Zhou, E. Angelopoulou, S. Seifert, A. Cavallaro, J. Hornegger, 3{D} cardiac medical models, in: AMIA Annu. Symp Proc, vol. 2013, 2013,
D. Comaniciu, Comparing axial {CT} slices in quantized N-dimensional pp. 112–121.
{SURF} descriptor space to estimate the visible body region, Comput. [52] R. Sparks, A. Madabhushi, Explicit shape descriptors: Novel morpho-
Med. Imaging Graph. 35 (3) (2011) 227–236, http://dx.doi.org/10.1016/ logic features for histopathology classification, Med. Image Anal. 17 (8)
j.compmedimag.2010.11.004, URL http://www.sciencedirect.com/science/ (2013) 997–1009, http://dx.doi.org/10.1016/j.media.2013.06.002, URL http:
article/pii/S0895611110001126. //www.sciencedirect.com/science/article/pii/S136184151300087X.
[34] L.B. Nascimento, A.C. de Paiva, A.C. Silva, Lung nodules classification in CT im- [53] T. Syeda-Mahmood, C.B. Compas, D. Beymer, R. Kumar, Similarity retrieval
ages using shannon and simpson diversity indices and SVM, in: Proceedings of angiogram images BASED on a flexible shape model, in: Lecture Notes
of the 8th International Conference on Machine Learning and Data Mining in Computer Science (including subseries Lecture Notes in Artificial Intel-
in Pattern Recognition, MLDM’12, Springer-Verlag, Berlin, Heidelberg, 2012, ligence and Lecture Notes in Bioinformatics), FIMH’13, in: LNCS, vol. 7945,
pp. 454–466, http://dx.doi.org/10.1007/978-3-642-31537-436. Springer-Verlag, Berlin, Heidelberg, 2013, pp. 97–105, http://dx.doi.org/10.
[35] Y. Rigolle, W. Puentes, M. Giordani, C. Roux, Evaluation of 3D anatomical 1007/978-3-642-38899-6_12.
surfaces indexing for surgery planning, in: Engineering in Medicine and [54] A.D. Ward, G. Hamarneh, R. Ashry, M.E. Schweitzer, 3D shape analysis of
Biology Society, 2006, EMBS ’06, 28th Annual International Conference of the the supraspinatus muscle: A clinical study of the relationship between
IEEE, 2006, pp. 3349–3352, http://dx.doi.org/10.1109/IEMBS.2006.259968. shape and pathology, Acad. Radiol. 14 (10) (2007) 1229–1241, http://
[36] G. González, E. Turetken, F. Fleuret, P. Fua, Delineating trees in noisy 2D dx.doi.org/10.1016/j.acra.2007.06.014, URL http://www.sciencedirect.com/
images and 3D image-stacks, in: Computer Vision and Pattern Recognition, science/article/pii/S1076633207003376.
CVPR, 2010 IEEE Conference on, IEEE Computer Society, San Francisco, USA, [55] F. O’Sullivan, O. Eary, S. Roy, J. O’Sullivan, C. Vernon, An examination of the
2010, pp. 2799–2806, http://dx.doi.org/10.1109/CVPR.2010.5540010. statistical reliability and prognostic potential of a measure of 3D spatial
[37] S. Liu, W. Cai, L. Wen, S. Eberl, M.J. Fulham, D. Feng, A robust volumetric heterogeneity developed for volumetric PET images of human sarcomas, in:
feature extraction approach for 3D neuroimaging retrieval, in: Engineering Nuclear Science Symposium Conference Record, 2002 IEEE, vol. 2, 2002, pp.
in Medicine and Biology Society, EMBC, 2010 Annual International Con- 1131–1134, http://dx.doi.org/10.1109/NSSMIC.2002.1239521.
ference of the IEEE, 2010, pp. 5657–5660, http://dx.doi.org/10.1109/IEMBS. [56] H. Wu, J. Kim, W. Cai, D. Feng, Volume of interest (VOI) feature representation
2010.5627900. and retrieval of multi-dimensional dynamic positron emission tomography
[38] H. Zhang, H. Zhong, E. Finol, A content-based 3D shape retrieval system images, in: Intelligent Multimedia, Video and Speech Processing, 2004.
for Abdominal Aortic Aneurysm rupture risk prediction, in: Biomedical Proceedings of 2004 International Symposium on, 2004, pp. 639–642, http:
Imaging: From Nano to Macro, 2011 IEEE International Symposium on, 2011, //dx.doi.org/10.1109/ISIMP.2004.1434145.
pp. 697–701, http://dx.doi.org/10.1109/ISBI.2011.5872502. [57] S. Liu, W. Cai, L. Wen, S. Eberl, M.J. Fulham, D.D. Feng, Generalized regional
[39] J.S. Faruque, D.L. Rubin, C.F. Beaulieu, J. Rosenberg, A. Kamaya, G. Tye, S. disorder-sensitive-weighting scheme for 3D neuroimaging retrieval, in: En-
Napel, R.M. Summers, A scalable reference standard of visual similarity for a gineering in Medicine and Biology Society, EMBC, 2011 Annual International
content-based image retrieval system, in: Healthcare Informatics, Imaging Conference of the IEEE, 2011, pp. 7009–7012, http://dx.doi.org/10.1109/
and Systems Biology, HISB, 2011 First IEEE International Conference on, IEMBS.2011.6091772.
2011, pp. 158–165,. [58] U. Castellani, P. Mirtuono, V. Murino, M. Bellani, G. Rambaldelli, M. Tansella,
[40] R. Guerrero, L. Pizarro, R. Wolz, D. Rueckert, Landmark localisation in P. Brambilla, A new shape diffusion descriptor for brain classification, in:
brain MR images using feature point descriptors based on 3D local self- Proceedings of the 14th International Conference on Medical Image Com-
similarities, in: Biomedical Imaging, ISBI, 2012 9th IEEE International puting and Computer-assisted Intervention - Volume Part II, MICCAI’11,
Symposium on, 2012, pp. 1535–1538, http://dx.doi.org/10.1109/ISBI.2012. Springer-Verlag, Berlin, Heidelberg, 2011, pp. 426–433, URL http://dl.acm.
6235865. org/citation.cfm?id=2044749.2044805.
[59] J. Goñi, O. Sporns, H. Cheng, M. Aznárez-Sanado, Y. Wang, S. Josa, G. [76] D. Wang, A.H. Tewfik, Y. Zhang, Y. Shen, Sparse representation of deformable
Arrondo, V.P. Mathews, T.A. Hummer, W.G. Kronenberger, A. Avena- 3D organs with spherical harmonics and structured dictionary, J. Biomed.
Koenigsberger, A.J. Saykin, M.A. Pastor, Robust estimation of fractal mea- Imaging 2011 (2011) 6:1–6:17, http://dx.doi.org/10.1155/2011/658930.
sures for characterizing the structural complexity of the human brain: Op- [77] S. Liu, W. Cai, L. Wen, S. Eberl, M.J. Fulham, D. Feng, Localized functional
timization and reproducibility, Neuroimage 83 (2013) 646–657, http://dx. neuroimaging retrieval using 3D discrete curvelet transform, in: Biomedical
doi.org/10.1016/j.neuroimage.2013.06.072, URL http://www.sciencedirect. Imaging: From Nano to Macro, 2011 IEEE International Symposium on, 2011,
com/science/article/pii/S1053811913007106. pp. 1877–1880, http://dx.doi.org/10.1109/ISBI.2011.5872774.
[60] J. Feng, H.H. Ip, {I}terative 3{D} point-set registration based on hierarchical [78] M.R. Ogiela, R. Tadeusiewicz, M. Trzupek, Graph-based linguistic formalisms
vertex signature ({H}{V}{S}), Med. Image Comput. Comput Assist. Interv. 8 in spatial modelling of 3d coronary vessels, in: Imaging Systems and Tech-
(Pt 2) (2005) 279–286. niques, 2007, IST ’07, IEEE International Workshop on, 2007, PP. 1–6, http:
[61] W.B.H. Khelifa, A. Ben Abdallah, F. Ghorbel, Three dimensional modeling of //dx.doi.org/10.1109/IST.2007.379567.
the left ventricle of the heart using spherical harmonic analysis, in: Biomed- [79] A. Pepe, L. Brandolini, M. Piastra, J. Koikkalainen, J. Hietala, J. Tohka, Simpli-
ical Imaging: From Nano to Macro, 2008. ISBI 2008. 5th IEEE International fied reeb graph as effective shape descriptor for the striatum, in: Proceedings
Symposium on, 2008, pp. 1275–1278, http://dx.doi.org/http://dx.doi.org/10. of the 2012 International Conference on Mesh Processing in Medical Image
1109/ISBI.2008.4541236. Analysis, MeshMed’12, Springer-Verlag, Berlin, Heidelberg, 2012, pp. 134–
[62] F. Maffessanti, C. Corsi, R. Lang, E. Caiani, Development of a method for left 146, http://dx.doi.org/10.1007/978-3-642-33463-4_14.
ventricular shape evaluation based on surfaces obtained by real-time 3D [80] W. Hong, F. Qiu, A. Kaufman, A pipeline for computer aided polyp detection,
echocardiographic images, in: Computers in Cardiology, vol. 34, 2007, pp. IEEE Trans. Vis. Comput. Graphics 12 (5) (2006) 861–868, http://dx.doi.org/
1–4, http://dx.doi.org/10.1109/CIC.2007.4745406. 10.1109/TVCG.2006.112.
[63] V. Nandakumar, X. An, Y. Wang, R. Johnson, D. Meldrum, Conformal mapping [81] J.M. Aman, J. Yao, R.M. Summers, Content-based image retrieval on CT
of nuclei in 3D tomographic cell images to assess shape heterogeneity, in: colonography using rotation and scale invariant features and bag-of-words
Biomedical Imaging, ISBI, 2012 9th IEEE International Symposium on, 2012, model, in: Biomedical Imaging: From Nano to Macro, 2010 IEEE Interna-
pp. 222–225, http://dx.doi.org/10.1109/ISBI.2012.6235524. tional Symposium on, 2010, pp. 1357–1360, http://dx.doi.org/10.1109/ISBI.
[64] S. Liu, W. Cai, L. Wen, D. Feng, Multiscale and multiorientation feature 2010.5490249.
extraction with degenerative patterns for 3D neuroimaging retrieval, in: [82] M. Yu, Z. Lu, Q. Feng, W. Chen, Liver CT image retrieval based on non-tensor
Image Processing, ICIP, 2012 19th IEEE International Conference on, 2012, product wavelet, in: Medical Image Analysis and Clinical Applications, MI-
pp. 1249–1252, http://dx.doi.org/10.1109/ICIP.2012.6467093. ACA, 2010 International Conference on, 2010, pp. 67–70, http://dx.doi.org/
[65] P. Morgado, M. Silveira, J. Marques, Extending local binary patterns to 3D for 10.1109/MIACA.2010.5528419.
the diagnosis of Alzheimer’s Disease, in: Biomedical Imaging, ISBI, 2013 IEEE
[83] Y. Song, W. Cai, S. Eberl, M.J. Fulham, D. Feng, Thoracic image case retrieval
10th International Symposium on, 2013, pp. 117–120, http://dx.doi.org/10.
with spatial and contextual information, in: Biomedical Imaging: From Nano
1109/ISBI.2013.6556426.
to Macro, 2011 IEEE International Symposium on, 2011, pp. 1885–1888,
[66] S. Liu, W. Cai, L. Wen, D. Feng, Multi-channel brain atrophy pattern analysis http://dx.doi.org/10.1109/ISBI.2011.5872776.
in neuroimaging retrieval, in: Biomedical Imaging, ISBI, 2013 IEEE 10th
[84] B.G. Prasad, A.N. Krishna, Statistical texture feature-based retrieval and
International Symposium on, 2013, pp. 202–205, http://dx.doi.org/10.1109/
performance evaluation of CT brain images, in: Electronics Computer Tech-
ISBI.2013.6556447.
nology, ICECT, 2011 3rd International Conference on, vol. 2, 2011, pp. 289–
[67] S. Dong, K. Dong, L. Yin, The application of local structure classification in 293, http://dx.doi.org/10.1109/ICECTECH.2011.5941703.
content-based 3d medical image retrieval, in: Image and Graphics, ICIG,
[85] Y. Song, W. Cai, S. Eberl, M.J. Fulham, D.D. Feng, Thoracic image matching
2013 Seventh International Conference on, 2013, pp. 639–642, http://dx.doi.
with appearance and spatial distribution, in: Engineering in Medicine and
org/10.1109/ICIG.2013.132.
Biology Society, EMBC, 2011 Annual International Conference of the IEEE,
[68] T. Majtner, D. Svoboda, Texture analysis using 3D Gabor features and 3D
2011, pp. 4469–4472, http://dx.doi.org/10.1109/IEMBS.2011.6091108.
MPEG-7 Edge Histogram descriptor in fluorescence microscopy, in: 3D
[86] A.K. Mishra, K. Desok, I. Andayana, Development of three dimensional
Imaging, IC3D, 2014 International Conference on, 2014, pp. 1–7, http://dx.
binary patterns for local bone structure analysis, in: IEEE International
doi.org/10.1109/IC3D.2014.7032576.
Conference on Bioinformatics and Biomedicine Workshops, BIBMW, 2011,
[69] O.S. Al-Kadi, D.Y.F. Chung, R.C. Carlisle, C.C. Coussios, J.A. Noble, Quan-
pp. 1006–1008, http://dx.doi.org/10.1109/BIBMW.2011.6112540, URL http:
tification of ultrasonic texture intra-heterogeneity via volumetric stochas-
//ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=6112540.
tic modeling for tissue characterization, Med. Image Anal. 21 (1) (2015)
[87] A. Quddus, O. Basir, Semantic image retrieval in magnetic resonance brain
59–71, http://dx.doi.org/10.1016/j.media.2014.12.004, URL http://www.
volumes, IEEE Trans. Inf. Technol. Biomed. 16 (3) (2012) 348–355, http:
sciencedirect.com/science/article/pii/S1361841514001881.
//dx.doi.org/10.1109/TITB.2012.2189439.
[70] R. Ayari, A.B. Abdallah, R. Sfar, F. Ghorbel, M.H. Bedoui, Analysis of re-
gional deformation of the heart’s left ventricle using invariant {SPHARM} [88] S. Mohanapriya, M. Vadivel, Automatic retrival of MRI brain image using
descriptors, IRBM 35 (5) (2014) 226–232, http://dx.doi.org/10.1016/j. multiqueries system, in: Information Communication and Embedded Sys-
irbm.2014.05.002, URL http://www.sciencedirect.com/science/article/pii/ tems, ICICES, 2013 International Conference on, 2013, pp. 1099–1103, http:
S1959031814000773. //dx.doi.org/10.1109/ICICES.2013.6508214.
[71] J.L. Bernal-Rusiel, M. Atienza, J.L. Cantero, Detection of focal changes in [89] S. Kim, Y.-W. Tai, Hierarchical nonrigid model for 3D medical image regis-
human cortical thickness: Spherical wavelets versus Gaussian smoothing, tration, in: Image Processing, ICIP, 2014 IEEE International Conference on,
Neuroimage 41 (4) (2008) 1278–1292, http://dx.doi.org/10.1016/j. 2014, pp. 3562–3566, http://dx.doi.org/10.1109/ICIP.2014.7025723.
neuroimage.2008.03.022, URL http://www.sciencedirect.com/science/ [90] E.-S.A. El-Dahshan, H.M. Mohsen, K. Revett, A.-B.M. Salem, Computer-aided
article/pii/S1053811908002486. diagnosis of human brain tumor through MRI: A survey and a new algorithm,
[72] S. Li, Q. Zhao, S. Wang, T. Hou, A. Hao, H. Qin, A novel material-aware feature Expert Syst. Appl. 41 (11) (2014) 5526–5545, http://dx.doi.org/10.1016/
descriptor for volumetric image registration in diffusion tensor space, in: j.eswa.2014.01.021, URL http://www.sciencedirect.com/science/article/pii/
Lecture Notes in Computer Science (including subseries Lecture Notes in S0957417414000426.
Artificial Intelligence and Lecture Notes in Bioinformatics), ECCV’12, in: [91] U.R. Acharya, S.V. Sree, M.M.R. Krishnan, F. Molinari, R. Garberoglio, J.S. Suri,
LNCS, vol. 7575, (PART 4) Springer-Verlag, Berlin, Heidelberg, 2012, pp. 502– Non-invasive automated 3D thyroid lesion classification in ultrasound: A
515, http://dx.doi.org/10.1007/978-3-642-33765-9_36. class of ThyroScanTM systems, Ultrasonics 52 (4) (2012) 508–520, http://dx.
[73] K. Rajakumar, S. Muttan, Medical image retrieval using modified {DCT}, doi.org/10.1016/j.ultras.2011.11.003, URL http://www.sciencedirect.com/
Procedia Comput. Sci. 2 (0) (2010) 298–302, http://dx.doi.org/10.1016/ science/article/pii/S0041624X11002472.
j.procs.2010.11.038, URL http://www.sciencedirect.com/science/article/pii/ [92] M.S. Sudhakar, K.B. Bagan, An effective biomedical image retrieval frame-
S1877050910003686. work in a fuzzy feature space employing phase congruency and Geo-
[74] M.E. Shenton, G. Gerig, R.W. McCarley, G. Székely, R. Kikinis, Amygdala- SOM, Appl. Soft Comput. 22 (2014) 492–503, http://dx.doi.org/10.1016/
hippocampal shape differences in schizophrenia: the application j.asoc.2014.04.029, URL http://www.sciencedirect.com/science/article/pii/
of 3D shape models to volumetric {MR} data, Psychiatry Res. S156849461400194X.
Neuroimaging 115 (1–2) (2002) 15–35, http://dx.doi.org/10.1016/S0925- [93] Y. Xu, E.J.R. van Beek, Y. Hwanjo, J. Guo, G. McLennan, E.A. Hoffman,
4927(02)00025-2, URL http://www.sciencedirect.com/science/article/pii/ Computer-aided classification of interstitial lung diseases via MDCT: 3D
S0925492702000252. Adaptive Multiple Feature Method (3D AMFM), Acad. Radiol. 13 (8) (2006)
[75] O. van Kaick, G. Hamarneh, A.D. Ward, M. Schweitzer, H. Zhang, Learn- 969–978, http://dx.doi.org/10.1016/j.acra.2006.04.017, URL http://www.
ing Fourier descriptors for computer-aided diagnosis of the supraspina- sciencedirect.com/science/article/pii/S1076633206002716.
tus, Acad. Radiol. 17 (8) (2010) 1040–1049, http://dx.doi.org/10.1016/ [94] X.W. Gao, Y. Qian, R. Hui, {T}he state of the art of medical imaging tech-
j.acra.2010.04.006, URL http://www.sciencedirect.com/science/article/pii/ nology: from creation to archive and back, Open Med. Inform. J. 5 (Suppl 1)
S1076633210001868. (2011) 73–85.
[95] W. Cai, S. Liu, L. Wen, S. Eberl, M.J. Fulham, D. Feng, 3D neurological image [115] J. Gutierrez-Cáceres, C. Portugal-Zambrano, C. Beltrán-Castañón, Computer
retrieval with localized pathology-centric CMRGlc patterns, in: Image Pro- aided medical diagnosis tool to detect normal/abnormal studies in dig-
cessing, ICIP, 2010 17th IEEE International Conference on, 2010, pp. 3201– ital MR brain images, in: 2014 IEEE 27th International Symposium on
3204, http://dx.doi.org/10.1109/ICIP.2010.5651869. Computer-Based Medical Systems, 2014, pp. 501–502, http://dx.doi.org/10.
[96] W. Zhang, J. Noble, J. Brady, {A}daptive non-rigid registration of real time 1109/CBMS.2014.110.
3{D} ultrasound to cardiovascular {M}{R} images, Inf. Process. Med. Imaging [116] F. Ciompi, C. Jacobs, E.T. Scholten, M.M.W. Wille, P.A. de Jong, M. Prokop,
20 (2007) 50–61. B. van Ginneken, Bag-of-frequencies: a descriptor of pulmonary nodules in
[97] A. Depeursinge, A. Vargas, A. Platon, A. Geissbuhler, P.-A. Poletti, H. Müller, computed tomography images, IEEE Trans. Med. Imaging 34 (4) (2015) 962–
Building a reference multimedia database for interstitial lung diseases, 973, http://dx.doi.org/10.1109/TMI.2014.2371821.
Comput. Med. Imaging Graph. 36 (3) (2012) 227–238, http://dx.doi.org/ [117] J. Esther, M. Sathik, Retrieval of brain image using soft computing technique,
10.1016/j.compmedimag.2011.07.003, URL http://www.sciencedirect.com/ in: Intelligent Computing Applications, ICICA, 2014 International Conference
science/article/pii/S0895611111001017. on, 2014, pp. 285–290, http://dx.doi.org/10.1109/ICICA.2014.104.
[98] P. Mountney, G.-Z. Yang, Context specific descriptors for tracking deforming [118] A.F. Abate, M. Nappi, G. Tortora, M. Tucci, IME: an image management envi-
tissue, Med. Image Anal. 16 (3) (2012) 550–561, http://dx.doi.org/10.1016/j. ronment with content-based access, Image Vis. Comput. 17 (13) (1999) 967–
media.2011.02.010, URL http://www.sciencedirect.com/science/article/pii/ 980, http://dx.doi.org/10.1016/S0262-8856(98)00186-3, URL http://www.
S1361841511000570. sciencedirect.com/science/article/pii/S0262885698001863.
[99] G. Quellec, M. Lamard, G. Cazuguel, B. Cochener, C. Roux, Wavelet [119] S. Antani, D.J. Lee, L.R. Long, G.R. Thoma, Evaluation of shape similarity mea-
optimization for content-based image retrieval in medical databases, surement methods for spine X-ray images, J. Vis. Commun. Image Represent.
Med. Image Anal. 14 (2) (2010) 227–241, http://dx.doi.org/10.1016/j. 15 (3) (2004) 285–302, http://dx.doi.org/10.1016/j.jvcir.2004.04.005, URL
media.2009.11.004, URL http://www.sciencedirect.com/science/article/pii/ http://www.sciencedirect.com/science/article/pii/S1047320304000367.
S1361841509001418. [120] P.R. Bakic, D. Kontos, V. Megalooikonomou, M.A. Rosen, A.D.A. Maidment,
[100] Y. Liu, F. Dellaert, A classification based similarity metric for 3D image Comparison of methods for classification of breast ductal branching pat-
retrieval, in : Computer Vision and Pattern Recognition, 1998. Proceedings, terns, in: Proceedings of the 8th International Conference on Digital Mam-
1998 IEEE Computer Society Conference on, 1998, pp. 800–805, http://dx. mography, IWDM’06, Springer-Verlag, Berlin, Heidelberg, 2006, pp. 634–
doi.org/10.1109/CVPR.1998.698695. 641, http://dx.doi.org/10.1007/11783237_85.
[101] A. Kak, C. Pavlopoulou, Content-based image retrieval from large medical [121] S. Baloch, C. Davatzikos, Morphological appearance manifolds in
databases, in: 3D Data Processing Visualization and Transmission, 2002. computational anatomy: Groupwise registration and morphological
Proceedings. First International Symposium on, 2002, pp. 138–147, http: analysis, Neuroimage 45 (1, Suppl. 1) (2009) S73 – S85, http://dx.doi.org/
//dx.doi.org/10.1109/TDPVT.2002.1024053. 10.1016/j.neuroimage.2008.10.048, URL http://www.sciencedirect.com/
[102] F. Kruggel, V. Kovalev, Classifying brain shapes, in: Proceedings of the 9th science/article/pii/S1053811908011695.
International Conference on Neural Information Processing, 2002, ICONIP
[122] J. Bermejo, J. Timperley, R.G. Odreman, M. Mulet, J.A. Noble, A.P. Banning, R.
’02, vol. 5, 2002, pp. 2678–2682, http://dx.doi.org/10.1109/ICONIP.2002.
Yotti, E. Pérez-David, J. Declerck, H. Becher, M.A. García-Fernández, Objec-
1201982.
tive quantification of global and regional left ventricular systolic function by
[103] J. Moustakas, K. Marias, S. Dimitriadis, S. Orphanoudakis, A two-level CBIR
endocardial tracking of contrast echocardiographic sequences, Int. J. Cardiol.
platform with application to brain MRI retrieval, in: Multimedia and Expo,
124 (1) (2008) 47–56, http://dx.doi.org/10.1016/j.ijcard.2006.12.091, URL
2005. ICME 2005. IEEE International Conference on, 2005, pp. 1278–1281,
http://www.sciencedirect.com/science/article/pii/S016752730700335X.
http://dx.doi.org/10.1109/ICME.2005.1521662.
[123] L. Bocchi, J. Nori, Shape analysis of microcalcifications using Radon
[104] M. Najjar, C. Ambroise, J.-R. Cocquerez, A. Cotten, M. Eltabach, A content-
transform, Med. Eng. Phys. 29 (6) (2007) 691–698, http://dx.doi.org/
based image retrieval system for osteo-articular applications, in: Informa-
10.1016/j.medengphy.2006.07.012, URL http://www.sciencedirect.com/
tion and Communication Technologies, 2006, ICTTA ’06, 2nd. vol. 1, 2006,
science/article/pii/S1350453306001640.
pp. 1726–1731, http://dx.doi.org/10.1109/ICTTA.2006.1684646.
[124] F. Calle-Alonso, C.J. Pérez, J.P. Arias-Nicolás, J. Martín, Computer-aided di-
[105] S. Jiang, S. Yang, X. Zhou, W. Chen, Fuzzy region content based image
agnosis system: A Bayesian hybrid classification method, Comput. Meth-
retrieval and relevance feedback for medical cerebral image, in: Biomedical
ods Programs Biomed. 112 (1) (2013) 104–113, http://dx.doi.org/10.1016/
Engineering and Informatics, BMEI, 2010 3rd International Conference on,
j.cmpb.2013.05.029, URL http://www.sciencedirect.com/science/article/pii/
vol. 1, 2010, pp. 282–285, http://dx.doi.org/10.1109/BMEI.2010.5639479.
S0169260713001910.
[106] R. Liu, S. Li, C.L. Tan, B.C. Pang, C.C.T. Lim, C.K. Lee, Q. Tian, Z. Zhang, Fast
[125] J. Carballido-Gamio, J.S. Bauer, R. Stahl, K.-Y. Lee, S. Krause, T.M. Link,
traumatic brain injury CT slice indexing via anatomical feature classification,
S. Majumdar, Inter-subject comparison of {MRI} knee cartilage thick-
in: Image Processing, ICIP, 2010 17th IEEE International Conference on, 2010,
ness, Med. Image Anal. 12 (2) (2008) 120–135, http://dx.doi.org/10.1016/j.
pp. 4377–4380, http://dx.doi.org/10.1109/ICIP.2010.5652317.
[107] Y. Song, W. Cai, S. Eberl, M.J. Fulham, D. Feng, Structure-adaptive feature media.2007.08.002, URL http://www.sciencedirect.com/science/article/pii/
extraction and representation for multi-modality lung images retrieval, S1361841507000783.
in: Digital Image Computing: Techniques and Applications, DICTA, 2010 [126] R. Chaves, J. Ramírez, J.M. Górriz, Integrating discretization and asso-
International Conference on, 2010, pp. 152–157, http://dx.doi.org/10.1109/ ciation rule-based classification for Alzheimer’s disease diagnosis, Ex-
DICTA.2010.37. pert Syst. Appl. 40 (5) (2013) 1571–1578, http://dx.doi.org/10.1016/j.
[108] W. Yang, Q. Feng, Z. Lu, W. Chen, Metric learning for maximizing MAP eswa.2012.09.003, URL http://www.sciencedirect.com/science/article/pii/
and its application to content-based medical image retrieval, in: Biomedical S0957417412010573.
Imaging: From Nano to Macro, 2011 IEEE International Symposium on, 2011, [127] P. Dankerl, A. Cavallaro, A. Tsymbal, M.J. Costa, M. Suehling, R. Janka, M.
pp. 1901–1904, http://dx.doi.org/10.1109/ISBI.2011.5872780. Uder, M. Hammon, A retrieval-based computer-aided diagnosis system for
[109] M.M. Rahman, S.K. Antani, G.R. Thoma, A learning-based similarity fusion the characterization of liver lesions in {CT} scans, Acad. Radiol. 20 (12) (2013)
and filtering approach for biomedical image retrieval using svm classifica- 1526–1534, http://dx.doi.org/10.1016/j.acra.2013.09.001, URL http://www.
tion and relevance feedback, IEEE Trans. Inf. Technol. Biomed. 15 (4) (2011) sciencedirect.com/science/article/pii/S1076633213004091.
640–646, http://dx.doi.org/10.1109/TITB.2011.2151258. [128] A. Depeursinge, D. Racoceanu, J. Iavindrasana, G. Cohen, A. Platon, P.A.
[110] Y. Dong, Multi-feature based medical image retrieval, in: Electrical Elec- Poletti, H. Müller, Fusing visual and clinical information for lung tissue
tronics Engineering, EEESYM, 2012 IEEE Symposium on, 2012, pp. 522–524, classification in high-resolution computed tomography, Artif. Intell. Med.
http://dx.doi.org/10.1109/EEESym.2012.6258708. 50 (1) (2010) 13–21, http://dx.doi.org/10.1016/j.artmed.2010.04.006, URL
[111] A.-M. Giuca, K.A. Seitz, J. Furst, D. Raicu, Expanding diagnostically labeled http://www.sciencedirect.com/science/article/pii/S0933365710000382.
datasets using content-based image retrieval, in: Image Processing, ICIP, [129] A. Depeursinge, A. Foncubierta-Rodriguez, D.V.D. Ville, H. Müller, Three-
2012 19th IEEE International Conference on, 2012, pp. 2397–2400, http: dimensional solid texture analysis in biomedical imaging: Review and op-
//dx.doi.org/10.1109/ICIP.2012.6467380. portunities, Med. Image Anal. 18 (1) (2014) 176–196, http://dx.doi.org/
[112] F. Maiorana, A medical content based image retrieval system with eye 10.1016/j.media.2013.10.005, URL http://www.sciencedirect.com/science/
tracking relevance feedback, in: Computer-Based Medical Systems, CBMS, article/pii/S1361841513001497.
2013 IEEE 26th International Symposium on, 2013, pp. 297–302, http://dx. [130] L. Dettori, L. Semler, A comparison of wavelet, ridgelet, and curvelet-based
doi.org/10.1109/CBMS.2013.6627805. texture classification algorithms in computed tomography, Comput. Biol.
[113] S. Murala, Q.M.J. Wu, Local mesh patterns versus local binary patterns: Med. 37 (4) (2007) 486–498, http://dx.doi.org/10.1016/j.compbiomed.
Biomedical image indexing and retrieval, IEEE J. Biomed. Health Inf. 18 (3) 2006.08.002, URL http://www.sciencedirect.com/science/article/pii/
(2014) 929–938, http://dx.doi.org/10.1109/JBHI.2013.2288522. S0010482506001302.
[114] S. Sivakamasundari, G. Kavitha, V. Natarajan, S. Ramakrishnan, Proposal [131] I. Dimitrovski, D. Kocev, I. Kitanovski, S. Loskovska, S. Džeroski, Improved
of a content based retinal Image Retrieval system using Kirsch template medical image modality classification using a combination of visual and tex-
based edge detection, in: 2014 International Conference on Informatics, tual features, Comput. Med. Imaging Graph. 39 (2015) 14–26, http://dx.doi.
Electronics Vision, ICIEV, 2014, pp. 1–5, http://dx.doi.org/10.1109/ICIEV. org/10.1016/j.compmedimag.2014.06.005, URL http://www.sciencedirect.
2014.6850744. com/science/article/pii/S0895611114000986.
[132] D.R. Ericeira, A.C. Silva, A.C. de Paiva, M. Gattass, Detection of masses based [148] O. Sertel, J. Kong, H. Shimada, U.V. Catalyurek, J.H. Saltz, M.N. Gurcan,
on asymmetric regions of digital bilateral mammograms using spatial Computer-aided prognosis of neuroblastoma on whole-slide images:
description with variogram and cross-variogram functions, Comput. Biol. Classification of stromal development, Pattern Recognit. 42 (6)
Med. 43 (8) (2013) 987–999, http://dx.doi.org/10.1016/j.compbiomed. (2009) 1093–1103, http://dx.doi.org/10.1016/j.patcog.2008.08.027, URL
2013.04.019, URL http://www.sciencedirect.com/science/article/pii/ http://www.sciencedirect.com/science/article/pii/S0031320308003439.
S0010482513001169. [149] H. Shen, D. Ma, Y. Zhao, H. Sun, S. Sun, R. Ye, L. Huang, B. Lang, Y.
[133] M. Faraki, M.T. Harandi, A. Wiliem, B.C. Lovell, Fisher tensors for classifying Sun, MIAPS: A web-based system for remotely accessing and present-
human epithelial cells, Pattern Recognit. 47 (7) (2014) 2348–2359, http://dx. ing medical images, Comput. Methods Programs Biomed. 113 (1) (2014)
doi.org/10.1016/j.patcog.2013.10.011, URL http://www.sciencedirect.com/ 266–283, http://dx.doi.org/10.1016/j.cmpb.2013.09.008, URL http://www.
science/article/pii/S0031320313004214. sciencedirect.com/science/article/pii/S0169260713003076.
[134] A.V. Faria, K. Oishi, S. Yoshida, A. Hillis, M.I. Miller, S. Mori, Content- [150] C.-R. Shyu, C. Pavlopoulou, A.C. Kak, C.E. Brodley, L.S. Broderick, Using
based image retrieval for brain MRI: An image-searching engine and human perceptual categories for content-based retrieval from a medical
population-based analysis to utilize past clinical data for future diagno- image database, Comput. Vis. Image Underst. 88 (3) (2002) 119–151, http:
sis, NeuroImage: Clinical 7 (0) (2015) 367–376, http://dx.doi.org/10.1016/ //dx.doi.org/10.1006/cviu.2002.0972, URL http://www.sciencedirect.com/
j.nicl.2015.01.008, URL http://www.sciencedirect.com/science/article/pii/ science/article/pii/S1077314202909720.
S2213158215000091. [151] T. Syeda-Mahmood, D. Ponceleon, J. Yang, Validating cardiac echo diagnosis
[135] M.M. Fernández-Carrobles, G. Bueno, O. Déniz, J. Salido, M. García-Rojo, through video similarity, in: Proceedings of the 13th Annual ACM Interna-
L. Gonzández-López, Frequential versus spatial colour textons for breast tional Conference on Multimedia, MULTIMEDIA ’05, ACM, New York, NY,
{TMA} classification, Comput. Med. Imaging Graph. 42 (0) (2015) 25–37, USA, 2005, pp. 527–530, http://dx.doi.org/10.1145/1101149.1101268.
http://dx.doi.org/10.1016/j.compmedimag.2014.11.009, URL http://www. [152] T. Glatard, J. Montagnat, I.E. Magnin, Texture based medical image indexing
sciencedirect.com/science/article/pii/S0895611114001840. and retrieval: application to cardiac imaging, in: Proceedings of the 6th ACM
[136] R. Gallea, E. Ardizzone, R. Pirrone, O. Gambino, Three-dimensional Fuzzy SIGMM international workshop on Multimedia information retrieval, MIR
Kernel Regression framework for registration of medical volume data, ’04, ACM, New York, NY, USA, 2004, pp. 135–142, http://dx.doi.org/10.1145/
Pattern Recognit. 46 (11) (2013) 3000–3016, http://dx.doi.org/10.1016/j. 1026711.1026734.
patcog.2013.03.025, URL http://www.sciencedirect.com/science/article/pii/ [153] P. Welter, B. Fischer, R.W. Günther, T.M.D.n. Lehmann, Generic integration of
S0031320313001556. content-based image retrieval in computer-aided diagnosis, Comput. Meth-
[137] M. Häfner, M. Liedlgruber, A. Uhl, A. Vécsei, F. Wrba, Color treatment in ods Programs Biomed. 108 (2) (2012) 589–599, http://dx.doi.org/10.1016/
endoscopic image classification using multi-scale local color vector pat- j.cmpb.2011.08.010, URL http://www.sciencedirect.com/science/article/pii/
terns, Med. Image Anal. 16 (1) (2012) 75–86, http://dx.doi.org/10.1016/j. S0169260711002288.
media.2011.05.006, URL http://www.sciencedirect.com/science/article/pii/ [154] S.-N. Yu, C.-T. Chiang, C.-C. Hsieh, A three-object model for the similarity
S1361841511000569. searches of chest {CT} images, Comput. Med. Imaging Graph. 29 (8) (2005)
617–630, http://dx.doi.org/10.1016/j.compmedimag.2005.06.001, URL
[138] W. Hsu, S. Antani, L.R. Long, L. Neve, G.R. Thoma, SPIRS: A Web-based image
http://www.sciencedirect.com/science/article/pii/S0895611105000807.
retrieval system for large biomedical databases, Int. J. Med. Inf. 78, Sup-
ple (2009) S13 – S24, http://dx.doi.org/10.1016/j.ijmedinf.2008.09.006, URL [155] K. Yuan, Z. Tian, J. Zou, Y. Bai, Q. You, Brain {CT} image database build-
http://www.sciencedirect.com/science/article/pii/S1386505608001676. ing for computer-aided diagnosis using content-based image retrieval,
Inf. Process. Manag. 47 (2) (2011) 176–185, http://dx.doi.org/10.1016/
[139] J.I. Khan, D.Y.Y. Yun, Holographic image archive, Comput. Med. Imaging
j.ipm.2010.06.003, URL http://www.sciencedirect.com/science/article/pii/
Graph. 20 (4) (1996) 243–257, http://dx.doi.org/10.1016/S0895-
S030645731000052X.
6111(96)00017-1, URL http://www.sciencedirect.com/science/article/
[156] Y. Zheng, E. Daniel, A.A. Hunter III, R. Xiao, J. Gao, H. Li, M.G. Maguire,
pii/S0895611196000171.
D.H. Brainard, J.C. Gee, Landmark matching based retinal image alignment
[140] T.M. Lehmann, M.O. Güld, T. Deselaers, D. Keysers, H. Schubert, K.
by enforcing sparsity in correspondence matrix, Med. Image Anal. 18 (6)
Spitzer, H. Ney, B.B. Wein, Automatic categorization of medical images
(2014) 903–913, http://dx.doi.org/10.1016/j.media.2013.09.009, URL http:
for content-based retrieval and data mining, Comput. Med. Imaging Graph.
//www.sciencedirect.com/science/article/pii/S1361841513001461.
29 (2–3) (2005) 143–155, http://dx.doi.org/10.1016/j.compmedimag.
[157] P. Aggarwal, R. Vig, H.K. Sardana, Semantic and content-based medical image
2004.09.010, URL http://www.sciencedirect.com/science/article/pii/
retrieval for lung cancer diagnosis with the inclusion of expert knowledge
S0895611104001168.
and proven pathology, in: Image Information Processing (ICIIP), 2013 IEEE
[141] L.L.G. Oliveira, S.A. e Silva, L.H.V. Ribeiro, R.M. de Oliveira, C.J. Coelho, Second International Conference on, 2013, pp. 346–351, http://dx.doi.org/
A.L.S.S. Andrade, Computer-aided diagnosis in chest radiography for detec- 10.1109/ICIIP.2013.6707613.
tion of childhood pneumonia, Int. J. Med. Inf. 77 (8) (2008) 555–564, http://
[158] P.L. Stanchev, F. Fotouhi, MEDIMAGE - A multimedia database management
dx.doi.org/10.1016/j.ijmedinf.2007.10.010, URL http://www.sciencedirect.
system for alzheimer’s disease patients, in: Proceedings of the 5th Inter-
com/science/article/pii/S1386505607001803.
national Conference on Recent Advances in Visual Information Systems,
[142] V. Venkatraman, P.R. Chakravarthy, D. Kihara, {A}pplication of 3{D} VISUAL ’02, Springer-Verlag, London, UK, UK, 2002, pp. 187–193, URL http:
{Z}ernike descriptors to shape-based ligand similarity searching, J. Chem- //dl.acm.org/citation.cfm?id=647062.714610.
inform. 1 (2009) 19. [159] R. Gan, A.C. Chung, {M}ulti-dimensional mutual information based robust
[143] D. Pokrajac, V. Megalooikonomou, A. Lazarevic, D. Kontos, Z. Obradovic, image registration using maximum distance-gradient-magnitude, Inf. Pro-
Applying spatial distribution analysis techniques to classification of 3D cess. Med. Imaging 19 (2005) 210–221.
medical images, Artif. Intell. Med. 33 (3) (2005) 261–280, http://dx.doi.org/ [160] Y. Kawata, N. Niki, H. Ohmatsu, M. Kusumoto, R. Kakinuma, K. Mori, H.
10.1016/j.artmed.2004.07.001, URL http://www.sciencedirect.com/science/ Nishiyama, K. Eguchi, M. Kaneko, N. Moriyama, Three-dimensional CT image
article/pii/S0933365704001009. retrieval in a database of pulmonary nodules, in: Image Processing. 2002.
[144] A. Przelaskowski, Recovery of {CT} stroke hypodensity – An adaptive vari- Proceedings. 2002 International Conference on, vol. 3, 2002, pp. III–149–III–
ational approach, Comput. Med. Imaging Graph. (2015) http://dx.doi.org/ 152, http://dx.doi.org/10.1109/ICIP.2002.1038926.
10.1016/j.compmedimag.2015.03.006, URL http://www.sciencedirect.com/ [161] C.G. Zhao, H.Y. Cheng, Y.L. Huo, T.G. Zhuang, Liver CT-image retrieval based
science/article/pii/S089561111500066X. on Gabor texture, in: Engineering in Medicine and Biology Society, 2004.
[145] Y. Qian, R. Hui, X. Gao, 3D {CBIR} with sparse coding for image-guided IEMBS ’04. 26th Annual International Conference of the IEEE, vol. 1, 2004,
neurosurgery, Signal Process. 93 (6) (2013) 1673–1683, http://dx.doi.org/ pp. 1491–1494, http://dx.doi.org/10.1109/IEMBS.2004.1403458.
10.1016/j.sigpro.2012.10.020, URL http://www.sciencedirect.com/science/ [162] W. Tsang, A. Corboy, K. Lee, D. Raicu, J. Furst, Texture-based image retrieval
article/pii/S0165168412003829. for computerized tomography databases, in: Computer-Based Medical Sys-
[146] Y.-Y. Qin, J.T. Hsu, S. Yoshida, A.V. Faria, K.K. Oishi, P.G. Unschuld, G.W. Red- tems, 2005. Proceedings. 18th IEEE Symposium on, 2005, pp. 593–598, http:
grave, S.H. Ying, C.A. Ross, P.C.M. van Zijl, A.E. Hillis, M.S. Albert, C.G. Lyketsos, //dx.doi.org/10.1109/CBMS.2005.97.
M.I. Miller, S. Mori, Gross feature recognition of Anatomical Images based on [163] K.N. Manjunath, U.C. Niranjan, Linear models of cumulative distribution
Atlas grid (GAIA): Incorporating the local discrepancy between an atlas and function for content based medical image retrieval, in: Engineering in
a target image to capture the features of anatomic brain {MRI}, NeuroImage: Medicine and Biology Society, 2005. IEEE-EMBS 2005. 27th Annual Inter-
Clinical 3 (2013) 202–211, http://dx.doi.org/10.1016/j.nicl.2013.08.006, URL national Conference of the, 2005, pp. 6472–6475, http://dx.doi.org/10.1109/
http://www.sciencedirect.com/science/article/pii/S221315821300106X. IEMBS.2005.1615981.
[147] G.P. Robinson, H.D. Tagare, J.S. Duncan, C.C. Jaffe, Medical image col- [164] M.P. Heinrich, M. Jenkinson, M. Bhushan, T. Matin, F.V. Gleeson, S.M.
lection indexing: Shape-based retrieval using KD-trees, Comput. Med. Brady, J.A. Schnabel, MIND: Modality independent neighbourhood de-
Imaging Graph. 20 (4) (1996) 209–217, http://dx.doi.org/10.1016/S0895- scriptor for multi-modal deformable registration, Med. Image Anal. 16
6111(96)00014-6, URL http://www.sciencedirect.com/science/article/pii/ (7) (2012) 1423–1435, http://dx.doi.org/10.1016/j.media.2012.05.008, URL
S0895611196000146. http://www.sciencedirect.com/science/article/pii/S1361841512000643.
[165] J. Kalpathy-Cramer, W. Hersh, Effectiveness of global features for automatic [181] J. Aman, J. Yao, R.M. Summers, Reducing the false positive rate of computer
medical image classification and retrieval – The experiences of {OHSU} at aided detection for CT colonography using Content Based Image Retrieval,
ImageCLEFmed, Pattern Recognit. Lett. 29 (15) (2008) 2032–2038, http://dx. in: Biomedical Imaging: From Nano to Macro, 2009. ISBI ’09. IEEE Inter-
doi.org/10.1016/j.patrec.2008.05.013, URL http://www.sciencedirect.com/ national Symposium on, 2009, pp. 915–918, http://dx.doi.org/10.1109/ISBI.
science/article/pii/S016786550800189X. 2009.5193202.
[166] Y. Kawata, N. Niki, H. Ohmatsu, N. Moriyama, Example-based assisting [182] J.R. Ferreira Junior, M.C. Oliveira, P.M. de Azevedo-Marques, Integrating 3D
approach for pulmonary nodule classification in three-dimensional tho- image descriptors of margin sharpness and texture on a GPU-optimized
racic computed tomography images1, Acad. Radiol. 10 (12) (2003) 1402– similar pulmonary nodule retrieval engine, J. Supercomput. 73 (8) (2017)
1415, http://dx.doi.org/10.1016/S1076-6332(03)00507-5, URL http://www. 3451–3467, http://dx.doi.org/10.1007/s11227-016-1818-4, URL http://link.
sciencedirect.com/science/article/pii/S1076633203005075. springer.com/10.1007/s11227-016-1818-4.
[167] N. Kilic, O. Kursun, O.N. Ucan, Classification of the colonic polyps in CT- [183] T.H. Cormen, C. Stein, R.L. Rivest, C.E. Leiserson, Introduction to Algorithms,
colonography using region covariance as descriptor features of suspicious second ed., McGraw-Hill Higher Education, 2001.
regions, J. Med. Syst. 34 (2) (2010) 101–105, http://dx.doi.org/10.1007/ [184] V. Castelli, L.D. Bergman, Image Databases: Search and Retrieval of Digital
s10916-008-9221-1, URL http://dx.doi.org/10.1007/s10916-008-9221-1. Imagery, John Wiley & Sons, USA, 2004.
[168] D. Kontos, V. Megalooikonomou, J.C. Gee, Morphometric analysis of brain [185] H.H. Biscaro, H. Oliveira, L.C.C. Bergamasco, F.L.S. Nunes, A new descriptor for
images with reduced number of statistical tests: A study on the gender- retrieving 3D objects applied in congestive heart failure diagnosis, in: 2016
related differentiation of the corpus callosum, Artif. Intell. Med. 47 (1) 29th SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI),
(2009) 75–86, http://dx.doi.org/10.1016/j.artmed.2009.05.007, URL http:// IEEE, 2016, pp. 120–127, http://dx.doi.org/10.1109/SIBGRAPI.2016.025, URL
www.sciencedirect.com/science/article/pii/S0933365709000797. http://ieeexplore.ieee.org/document/7813024/.
[186] Y. Xu, L. Lin, H. Hu, D. Wang, Y. Liu, J. Wang, X. Han, Y.-W. Chen, A re-
[169] H. Müller, A. Rosset, J.-P. Vallée, F. Terrier, A. Geissbuhler, A reference
trieval system for 3D multi-phase contrast-enhanced CT images of focal
data set for the evaluation of medical image retrieval systems, Comput.
liver lesions based on combined bags of visual words and texture words, in:
Med. Imaging Graph. 28 (6) (2004) 295–305, http://dx.doi.org/10.1016/
2016 9th International Congress on Image and Signal Processing, BioMedical
j.compmedimag.2004.04.005, URL http://www.sciencedirect.com/science/
Engineering and Informatics (CISP-BMEI), IEEE, 2016, pp. 806–810, http:
article/pii/S0895611104000539.
//dx.doi.org/10.1109/CISP-BMEI.2016.7852820, URL http://ieeexplore.ieee.
[170] S. Murala, Q.M.J. Wu, {MRI} and {CT} image indexing and retrieval using
org/document/7852820/.
local mesh peak valley edge patterns, Signal Process., Image Commun. 29 (3) [187] C. Cortes, V. Vapnik, Support-vector networks, Mach. Learn. 20 (3) (1995)
(2014) 400–409, http://dx.doi.org/10.1016/j.image.2013.12.002, URL http: 273–297.
//www.sciencedirect.com/science/article/pii/S0923596513001872. [188] B. Scholkopf, A. Smola, Support vector machines, in: Encyclopedia of
[171] H. Pan, Q. Han, X. Xie, Z. Wei, J. Li, A similarity retrieval method in brain Biostatistics, John Wiley & Sons, Ltd., 2005, http://dx.doi.org/10.1002/
image sequence database, in: Proceedings of the 3rd International Con- 0470011815.b2a14038.
ference on Advanced Data Mining and Applications, ADMA ’07, Springer- [189] {A}lzheimer’s {D}isease {N}euroimaging {I}nitiative, 2015.
Verlag, Berlin, Heidelberg, 2007, pp. 352–364, http://dx.doi.org/10.1007/ [190] D. Sarwinda, A. Bustamam, Detection of Alzheimer’s disease using advanced
978-3-540-73871-8_33. local binary pattern from hippocampus and whole brain of MR images, in:
[172] T. Brosch, R. Tam, {M}anifold learning of brain {M}{R}{I}s by deep learning, 2016 International Joint Conference on Neural Networks (IJCNN), 2016, pp.
Med. Image Comput. Comput. Assist. Interv. 16 (Pt 2) (2013) 633–640. 5051–5056, http://dx.doi.org/10.1109/IJCNN.2016.7727865.
[173] X. Qian, H.D. Tagare, R.K. Fulbright, R. Long, S. Antani, Optimal embedding [191] {B}rain{W}eb: {S}imulated {B}rain {D}atabase, 2015.
for shape indexing in medical image databases, Med. Image Anal. 14 (3) [192] {DIARETDB1} - {S}tandard {D}iabetic {R}etinopathy {D}atabase, 2015.
(2010) 243–254, http://dx.doi.org/10.1016/j.media.2010.01.001, URL http: [193] Image{CLEF} - {T}he {CLEF} {C}ross {L}anguage {I}mage {R}etrieval {T}rack,
//www.sciencedirect.com/science/article/pii/S1361841510000101. 2015.
[174] M.M. Rahman, B.C. Desai, P. Bhattacharya, Medical image retrieval with [194] {IRMA} ({I}mage {R}etrieval in {M}edical {A}pplications), 2015.
probabilistic multi-class support vector machine classifiers and adaptive [195] {L}ung {I}mage {D}atabase {C}onsortium {(LIDC)}, 2015.
similarity fusion, Comput. Med. Imaging Graph. 32 (2) (2008) 95–108, [196] {NHANES} - {N}ational {H}ealth and {N}utrition {E}xamination {S}urvey,
http://dx.doi.org/10.1016/j.compmedimag.2007.10.001, URL http://www. 2015.
sciencedirect.com/science/article/pii/S0895611107001383. [197] S. Tomei, A. Reilhac, D. Visvikis, N. Boussion, C. Odet, F. Giammarile, C.
[175] M.X. Ribeiro, P.H. Bugatti, C. Traina Jr., P.M.A. Marques, N.A. Rosa, A.J.M. Lartizien, OncoPET DB: A freely distributed database of realistic simulated
Traina, Supporting content-based image retrieval and computer-aided diag- whole body 18F-FDG PET images for oncology, IEEE Trans. Nucl. Sci. 57 (1)
nosis systems with association rule-based techniques, Data Knowl. Eng. 68 (2010) 246–255, http://dx.doi.org/10.1109/TNS.2009.2034375.
(12) (2009) 1370–1382, http://dx.doi.org/10.1016/j.datak.2009.07.002, URL [198] K. Clark, B. Vendt, K. Smith, J. Freymann, J. Kirby, P. Koppel, S. Moore, S.
http://www.sciencedirect.com/science/article/pii/S0169023X09001025. Phillips, D. Maffitt, M. Pringle, L. Tarbox, F. Prior, The cancer imaging archive
(TCIA): maintaining and operating a public information repository, J. Digit.
[176] M.-R. Siadat, H. Soltanian -Zadeh, F. Fotouhi, K. Elisevich, Content-based
Imaging 26 (6) (2013) 1045–1057, http://dx.doi.org/10.1007/s10278-013-
image database system for epilepsy, Comput. Methods Programs Biomed.
9622-7, URL http://link.springer.com/10.1007/s10278-013-9622-7.
79 (3) (2005) 209–226, http://dx.doi.org/10.1016/j.cmpb.2005.03.012, URL
[199] M.P. Sarathi, M.A. Ansari, A schemefor patient study retrieval from 3D
http://www.sciencedirect.com/science/article/pii/S016926070500088X.
brain MR volumes, in: 2015 International Conference on Green Computing
[177] S.G. Armato, G. McLennan, L. Bidaut, M.F. McNitt-Gray, C.R. Meyer, A.P.
and Internet of Things, ICGCIoT, IEEE, 2015, pp. 214–218, http://dx.doi.org/
Reeves, B. Zhao, D.R. Aberle, C.I. Henschke, E.A. Hoffman, E.A. Kazerooni,
10.1109/ICGCIoT.2015.7380460, URL http://ieeexplore.ieee.org/document/
H. MacMahon, E.J.R. van Beek, D. Yankelevitz, A.M. Biancardi, P.H. Bland,
7380460/.
M.S. Brown, R.M. Engelmann, G.E. Laderach, D. Max, R.C. Pais, D.P.Y. Qing,
[200] H. Lu, W. Wang, Q. Liao, G. Zhang, Z. Zhou, in: W.W. Boonn, B.J. Liu
R.Y. Roberts, A.R. Smith, A. Starkey, P. Batra, P. Caligiuri, A. Farooqi, G.W.
(Eds.), Medical Image Retrieval System Using Multiple Features from 3D
Gladish, C.M. Jude, R.F. Munden, I. Petkovska, L.E. Quint, L.H. Schwartz, B.
ROIs, vol. 8319, 2012, pp. 1–8, http://dx.doi.org/10.1117/12.911806, URL
Sundaram, L.E. Dodd, C. Fenimore, D. Gur, N. Petrick, J. Freymann, J. Kirby,
http://proceedings.spiedigitallibrary.org/proceeding.aspx?doi=10.1117/12.
B. Hughes, A. Vande Casteele, S. Gupte, M. Sallam, M.D. Heath, M.H. Kuhn, E.
911806.
Dharaiya, R. Burns, D.S. Fryd, M. Salganicoff, V. Anand, U. Shreter, S. Vastagh, [201] M. Kociński, A. Klepaczko, A. Materka, M. Chekenya, A. Lundervold, 3D image
B.Y. Croft, L.P. Clarke, The Lung Image Database Consortium (LIDC) and Image texture analysis of simulated and real-world vascular trees, Comput. Meth-
Database Resource Initiative (IDRI): A completed reference database of lung ods Programs Biomed. 107 (2) (2012) 140–154, http://dx.doi.org/10.1016/
nodules on CT scans, Med. Phys. 38 (2) (2011) 915–931, http://dx.doi.org/ j.cmpb.2011.06.004, URL http://www.sciencedirect.com/science/article/pii/
10.1118/1.3528204, URL http://www.ncbi.nlm.nih.gov/pubmed/21452728, S0169260711001556.
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=PMC3041807. [202] Y. Yang, K.W.C. Foong, S.H. Ong, M. Yagi, K. Takada, An image-based method
[178] W. Zeng, R. Shi, Y. Wang, S.-T. Yau, X. Gu, Teichmüller shape descriptor and for quantification of lateral pterygoid muscle deformation, in: Bioinformat-
its application to Alzheimer’s disease study, Int. J. Comput. Vision 105 (2) ics Bioengineering (BIBE), 2012 IEEE 12th International Conference on, 2012,
(2013) 155–170, http://dx.doi.org/10.1007/s11263-012-0586-8. pp. 698–702, http://dx.doi.org/10.1109/BIBE.2012.6399752.
[179] N. Iyer, S. Jayanti, K. Lou, Y. Kalyanaraman, K. Ramani, Three-dimensional [203] N. Lal, N. Gupta, A. Sinhal, A review of image classification techniques in
shape searching: state-of-the-art review and future trends, Comput. content based image retrieval, Neera Lal et al,/(IJCSIT) Int. J. Comput. Sci. Inf.
Aided Des. 37 (5) (2005) 509–530, http://dx.doi.org/10.1016/j.cad. Technol. 3 (5) (2012).
2004.07.002, URL http://www.sciencedirect.com/science/article/pii/ [204] Z. Li, X. Zhang, H. Müller, S. Zhang, Large-scale retrieval for medical im-
S001044850400140X. age analytics: A comprehensive review, Med. Image Anal. 43 (Suppl. C)
[180] A. Foncubierta-Rodríguez, A. Vargas, A. Platon, P. Poletti, H. Müller, A. De- (2018) 66–84, http://dx.doi.org/10.1016/j.media.2017.09.007, URL http://
peursinge, Retrieval of 4D Dual Energy CT for Pulmonary Embolism Diagno- www.sciencedirect.com/science/article/pii/S136184151730138X.
sis, Springer, Berlin, Heidelberg, 2013, pp. 45–55, http://dx.doi.org/10.1007/ [205] IEEE, IEEE Xplore Digital Library, URL http://ieeexplore.ieee.org/Xplore/
978-3-642-36678-9_5. home.jsp.
[206] O. Kucuktunc, H. Ferhatosmanoglu, #x003BB;-diverse nearest neighbors [207] Y. Gao, Q. Dai, M. Wang, N. Zhang, 3D model retrieval using weighted
browsing for multidimensional data, IEEE Trans. Knowl. Data Eng. 25 (3) bipartite graph matching, Signal Process., Image Commun. 26 (1) (2011)
(2013) 481–493, http://dx.doi.org/10.1109/TKDE.2011.251. 39–47, http://dx.doi.org/10.1016/j.image.2010.10.006, URL http://www.
sciencedirect.com/science/article/pii/S092359651000113X.

Computer Science Review: Leila Cristina C. Bergamasco, Fátima L.S. Nunes

Uploaded by

Copyright:

Available Formats

Computer Science Review: Leila Cristina C. Bergamasco, Fátima L.S. Nunes

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Computer Science Review: Leila Cristina C. Bergamasco, Fátima L.S. Nunes

Uploaded by

Copyright:

Available Formats

Computer Science Review 31 (2019) 19–38

Contents lists available at ScienceDirect

Computer Science Review

Intelligent retrieval and classification in three-dimensional

• A systematic mapping covering intelligent retrieval for 3D medical images.

advance in CBIR techniques applied to medical images, with special

retrieval’’[Title/Abstract] AND 3D)

descriptor AND 3D AND retrieval 25

TITLE-ABSTR-KEY(3D medical ) and (volume retrieval) 84

TITLE-ABSTR-KEY(3D medical ) and (surface retrieval) 65

TITLE-ABSTR-KEY(descriptor) and (medical 3D) 428

‘‘similarity retrieval’’ AND 3D AND medical 69

‘‘similarity retrieval’’ and 3D and medical 94

ABSTRACT:(medical descriptor) AND ABSTRACT:(3D) 11

3D AND medical AND (‘‘image retrieval’’ or ‘‘model retrieval’’) 29

this type of studies were excluded by the first exclusion criterion

4. Descriptors Spherical Harmonics (SH) is a frequency-based space that rep-

Fig. 5. Spherical Harmonics demonstration with complex constants variations.

4.3. 3D Wavelet Transform 5.1. Graphs-based indexing methods

The 3D Wavelet Transform decomposes an image into different

5.2. K-d trees

A K-dimensional tree (k-d tree) is a complex type of indexing

specific image modalities (PET, MRI or CT). Some databases are

The scenario of medical images classification and retrieval em-

have to critically analyze this knowledge domain to identify the

10.2. Papers quality

You might also like