Electroencephalogram (EEG) Signal

Analysis for Diagnosis of Major

Depressive Disorder (MDD): A Review

Shalini Mahato and Sanchita Paul

Abstract Depression or Major Depressive Disorder (MDD) is a psychiatric dis-

order. It is the major contributor to overall global burden of disease. Any deteri-
oration in brain functioning is reflected in Electroencephalogram (EEG) signals.
EEG signals are highly complex, nonstationary and nonlinear. So, it is very difficult
to analyze visually and identify changes in the waveform in order to classify MDD
and normal subjects. Thus, computerized quantitative EEG is used for the analysis
of signals. Support Vector Machine (SVM) using band power feature reported an
accuracy of 98.33% and using Kernel Eigen-Filter-Bank Common Spatial Patterns
(KEFB-CSP) gave an accuracy of 91.67% [1, 2]. Logistic Regression (LR) using
band power feature reported an accuracy of 98.33%, using 4 nonlinear features
combination provided an accuracy of 90%, using wavelet transform provided an
accuracy of 87.5% and using only alpha power gave an accuracy of 73.3% [1, 3, 4].
Naïve Bayesian (NB) using band power feature provided an accuracy of 96.8% [1].
Artificial Neural Network using Relative Wavelet Energy (RWE) reported an
accuracy of 98.11%, using power spectrum feature gave accuracy of 84% and using
Lep–Ziv complexity accuracy of 60–80% was reported [5, 6, 7]. Linear
Discriminant Analysis (LDA) reported an accuracy of 91.2% using SASI (Spectral
Asymmetry Index) and DFA (Detrended Fluctuation Analysis) [8]. Decision Tree
provided an accuracy of 80% using EEG band power as feature [9]. The study
reveals that, in general, high classification accuracy is achieved by SVM, LR and
ANN and highest classification accuracy of 98.33% is achieved by SVM. Highest
accuracy is achieved by SVM because it is more robust and computationally more
efficient due to maximal margin gap between separating hyper planes and kernel
trick. The study gives some ideas which could be helpful for guiding and improving
future researches. Since linear and nonlinear method for feature extraction are both
equally efficient. So any of the linear/nonlinear feature can be used. For feature

S. Mahato (&)
S. Paul

Department of Computer Science and Engineering,
Birla Institute of Technology, Mesra, Ranchi 835215, India
e-mail: [email protected]
S. Paul
e-mail: [email protected]

© Springer Nature Singapore Pte Ltd. 2019 323

V. Nath and J. K. Mandal (eds.), Nanoelectronics, Circuits and Communication
Systems, Lecture Notes in Electrical Engineering 511,
https://doi.org/10.1007/978-981-13-0776-8_30
324 S. Mahato and S. Paul

selection and reduction, Genetic Algorithm (GA), Principal Components Analysis

(PCA) and Linear Discriminant Analysis (LDA) can be used. Since SVM, LR and
ANN give high classification accuracy and any of them can be used or any hybrid
technique like GA-SVM, GA-ANN can be used. This chapter compares various
EEG signal analysis techniques, compares their accuracy and methodology used
and finally recommends the most suitable technique based on the accuracy for
detection of depression. The chapter also summarizes some of the key finding
related to EEG features based on present state of art. These results could be helpful
for guiding and improving the future research in this area.

Keywords Major Depressive Disorder (MDD) Electroencephalogram (EEG)




Biomarker Quantitative EEG Band power Spectral Asymmetry Index (SASI)


Relative Wavelet Energy (RWE) Higuchi’s Fractal Dimension (HFD)

1 Introduction

According to World Health Organization (WHO), depression is the major cause of

disability globally with respect to Years Lived with Disability (YLDs). Around
322 million people are suffering from depression globally. Females (5.1%) are
more affected by depression as compared to males (3.6%). Rate of increase of
number of people suffering from depression is 18.4% from 2005 to 2015 [10]. It is
predicted by WHO that depression will be the leading cause of disease burden
globally by 2030 [9].
A person suffering from MDD is not interested in usual activities and pleasure
along with four additional symptoms of depression for at least continuous 2 weeks
[11]. A person affected from depression suffers from setback at occupational, social,
and personal life. At worse, a person suffering from depression can even commit
suicide. Only less than 50% of the depression patients get treatment because of
incorrect diagnosis, insufficient resources, lack of trained medical care providers
and societal disgrace associated with mental disorder. At times, it happens that a
person suffering from depression is not detected and unaffected person is wrongly
identified and prescribed antidepressant.
The major reason behind wrong diagnosis is because of absence of any accepted
biomarkers for MDD or any other psychiatric disorder. Diagnosis of depression is
done on the basis of accepted depression classification criteria like Diagnostic and
Statistical Manual of Mental Disorders-Fourth Edition (DSM-IV), Beck Depression
Inventory (BDI), and Hamilton Depression Rating Scale (HAM-D), which depends
on some questionnaire asked and observed behavior during the interview session
between MDD patient and health practitioner.
No objective criteria exist for the assessment of depression presently. Analysis
on the basis of objective criteria would be helpful in detection of depression in early
stages so that it can be prevented from becoming severe and reaching untreatable
state.
Electroencephalogram (EEG) Signal Analysis for Diagnosis … 325

Change in bioelectrical activity of the brain is expected whenever there is any

deterioration in brain functioning. This change can be analyzed by brain screening,
which can be done with the help of EEG. EEG is cost-effective and easily available.
It also provides high temporal resolution as the time scale of EEG is in milliseconds
which is same as the time frame of neural activity.
Since EEG signals are highly complex, nonstationary and nonlinear, manually
interpreting the change in the signal for classifying normal and depression subjects
is a complex task. So computer-aided signal processing is needed to automatically
classify depression and normal subjects. Extensive research work has been done in
this field for distinguishing between MDD and normal subjects.
The study helps in identifying certain key ideas for future work. Both linear and
non linear method for feature extraction can be used as they are equally efficient.
The study reveals that highest classification was achieved using band power and
RWE. So either of them can be used or a combination of both can be used. GA,
PCA, LDA all are efficient for feature selection and reduction, so any of them can
be used. Since high classification accuracy is achieved by SVM, LR and ANN any
of them can be used or any hybrid technique like GA-SVM, GA-ANN can be used.

2 Quantitative EEG

First human EEG was recorded in 1920s by Hans Berger, a German neuroanatomist
and psychiatrist. With the development of digital computer technology, it was
possible to access and quantify correctly the exact values of many more EEG
variable compared to that obtained by visual inspection of EEG signals. This lead to
the development of Quantitative EEG (QEEG) It provides additional measurement
or topographic display of voltage or frequency of EEG signal. It mathematically
processes EEG signal to format or domain such that specific waveform component
is highlighted, which gives information or numerical values related to EEG data
which can be used for comparison or subsequent evaluation. QEEG technique
which deals with topographic display of voltage or frequency over entire scalp is
called EEG Brain Mapping. QEEG involves factors like amplitude of signal, fre-
quency, waveform, hemispheric reactivity.

3 Methodology

EEG signal consists of information about brain activity from different regions of
brain with the help of electrodes. These EEG signals may be contaminated with
various kinds of artifacts like eye blinks, movements, muscular activity, and
electrical noise from power line. These artifacts need to be removed with the help of
software or by visual detection by experts.
326 S. Mahato and S. Paul

After the removal of artifacts, feature extraction is carried out from the signals
which are used to differentiate between depression and normal subjects.
Basically two types of methods are used for feature extraction: linear and
nonlinear. In linear analysis, frequency and time domain approaches are used.
Mathematical theory of dynamical nonlinear systems forms the basis of nonlinear
EEG analysis.
Feature extracted with linear methods are inter-hemispheric asymmetry, band
power, EEG measurements (amplitude, frequency and power), Wavelet Transform
(WT), Fast Fourier Transform (FFT), Auto Regressive (AR) model, and so on.
Feature extracted with nonlinear methods are such as Higuchi’s Fractal Dimension
(HFD), Approximate Entropy (ApEn), Lyapunov Exponent (LE), Correlation
Dimension (CD), Detrended Fluctuation Analysis (DFA), and so on.
Since a large number of features are extracted from each of the channels, using
all the features for classification would lead to high computational cost. Some of the
researchers prefer to decrease this cost by using some of significant features and not
all features. This would substantially reduce the computational cost. Some of the
feature selection and reduction techniques used are Genetic Algorithm (GA),
Principal Components Analysis (PCA), and Linear Discriminant Analysis (LDA).
Further, the classification is done with the help of different classifier like LR,
SVM, NB, Back Propagation-Artificial Neural Network (BP ANN), LDA,
self-organizing competitive network, k-means clustering, k-Nearest Neighbor
(KNN), Kernel Principal Component Analysis (KPCA), Decision Tree (DT),
Multilayered Perceptron Neural Network (MLPNN), Probablistic Neural Network
(PNN) (Fig. 1).

Fig. 1 Stages of EEG signal

analysis for detection of
depression
Electroencephalogram (EEG) Signal Analysis for Diagnosis … 327

4 Literature Review

Classification accuracy depends to a great extent on the EEG feature used and
classification techniques used. Extensive research has been done using different
EEG features and different classifiers to improve classification accuracy between
depression patients and normal subjects. Some of the findings related to EEG
features based on present state of the art are discussed in the table.

5 Discussion

Table 1 depicts highest classification accuracy of 98.4%, which was achieved by

SVM classifier using all signal power and alpha asymmetry features. LR’s classi-
fication accuracy was found to be higher than LDA and KNN [3]. True detection
rate of HFD (94%) was reported to be higher than SASI (88%) [12]. CD had shown
the highest classification accuracy of (83.3%) among DFA, CD, HFD, LE along
with LR classifier [3]. It was reported that combination of DFA, HFD, and CD with
LR classifier provided high classification accuracy of (90%) (Tables 2 and 3).
It can be observed from the above analysis that there are a number of conflicting
results, which are mainly due to the following reasons:
(i) Lack of standardized common dataset on MDD
(ii) Different conditions during EEG recording (EO, EC, falling sleep, emotional
challenges, etc.) and different environment (heat, temperature, noise, illu-
mination, etc.)
(iii) Different channels are used for signal analysis
(iv) EEG recording length and length of epoch are different
(v) Different procedures for artifacts removal
(vi) Different or incomparable measures used.

6 Conclusion

From the above discussion, it is evident that EEG signals are efficient and reliable in
discriminating between MDD patient and normal subjects. But the major barrier in
finding standardized biomarkers is the absence of standardized common dataset,
techniques, and methodology used. Thus, these issues need to be solved in order to
establish clinically useful and authentic biomarker.
In this review chapter, several papers relating to EEG signal analysis and
depression has been summarized which would be beneficial in guiding and
improving the future research work in this area such that higher accuracy, sensi-
tivity and specificity can be achieved. This would help in the development of
Table 1 Summary of findings related to EEG signal’s features
328

Feature Description Findings

Band power EEG signal can be split into frequency bands: delta Alpha power: Exceptionally high alpha activity was found in the resting state in the
(linear (0.5–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), and posterior region of the brain during depression [13, 2] but conflicting results were
method) beta (13–30 Hz). Band power is the average power also found as per [4] which reported lower alpha signal power in all parts of the
contained in a particular frequency band brain during resting state. Low left frontal alpha activity during emotional challenge
(approach and withdrawal) in MDD patients [2]. Alpha power attains the highest
classification accuracy for discriminating between normal subjects and MDD as
compared to other bands [3, 14]. During emotional challenge (approach and
withdrawal), EEG alpha asymmetry was a stronger indicator of depression than in
resting state alpha asymmetry [2]. Alpha asymmetry was found to be decreased
remarkably in MDD especially in women when neuro-feedback treatment was done
[15]. Occipital alpha activity was found to be increased in depression patient. Right
parieto-temporal region showed comparatively higher alpha activity for depression
patients [13]
Beta power: Found to be higher in depression patients in central, temporal and
parietal region of brain [13, 8]. Beta power positively correlates with stress
symptoms. For depression patients beta values increased only for left hemisphere
[13]
Theta power: It is found to be less in all parts of brain in depression patient [8]
which was contradictory to the finding of increase of theta activity in occipital and
parietal region [13]. For depression patients, there was increase in theta values in
both left and right hemisphere [13]
Delta power: It is slightly more in temporal, central and parietal region of the brain
in depression [9]. But significantly higher temporal delta activity was found in
Acute Unstable Depression Syndrome (AUDS) patients than MDD patients.
Compared to MDD, Acute Unstable Depressive Syndrome (AUDS) had higher delta
activity in right hemisphere [16]
Classification accuracy higher than alpha band power was achieved, when all the
bands were analyzed together [3, 14]. Increase in the current source density maximal
(continued)
S. Mahato and S. Paul
Table 1 (continued)
Feature Description Findings
in the right hemisphere and highest in the alpha (55.6%) then theta (30.6%) band in
parietal and frontal region in depression patients [17]
SASI (linear SASI is calculated as the difference between SASI values were positive for depression patients and negative for normal subjects.
method) especially selected higher and lower frequency It is positively correlated with Hamilton Depression Scale (HAM-D) scores [7, 18].
band further from which central band is eliminated Pz channel provided maximum difference in SASI in differentiating normal subjects
and MDD patients. Using SASI, no statistically significant difference in MDD and
normal subjects with respect to hemispheric asymmetry was found. SASI classifying
accuracy were highest in parietal brain area [12]
HFD Higuchi’s algorithm is used to find the fractal Value of HFD increases with depression in all EEG channel. Statistically significant
(nonlinear dimension in time domain directly. Fractal difference between MDD patients and normal subjects for HFD in temporal, parietal
method) dimension determines the degree of irregularity in and occipital channel [19]. Using HFD, no statistically significant difference was
discrete time sequence found in MDD and normal subjects with respect to hemispheric asymmetry. HFD
classifying accuracy highest in parietal brain area [12]
RWE RWE is the relative energy related with different Normal subjects reported higher value of RWE in gamma, beta, alpha, and theta
(nonlinear frequency band. It reveals the degree of bands as compared to depression patients. Depression patients reported higher RWE
method) disorderness in different frequency bands value in delta band [20]. Higher Wavelet Entropy (WE) (mean WE = 1.327) was
found in depression patients than those of normal subjects (mean WE = 1.266). For
depression patients, resting state WE (mean WE = 1.349) was found to be higher
Electroencephalogram (EEG) Signal Analysis for Diagnosis …

than in mental arithmetic state (mean WE = 1.243) [21]. Using RWE, it was found
that right hemisphere has higher energy distribution in depressed patients than
normal subjects [22]
ApEn ApEn is the complexity measure which quantifies ApEn was much lower in MDD patients as compared to normal subjects [23]
(nonlinear regularity and complexity in time series data. ApEn
method) value is positively correlated with unpredictability
and randomness of the signal
DFA DFA analyzes the deviation of the signal with In channels O2 and Pz, the difference between MDD patients and normal subjects
(nonlinear respect to local signal’s tendency in data window were found to be a maximum for DFA. DFA values are significantly lower for
method) depressive group [18]
329
Table 2 Summary of previous work on EEG signal analysis for detection of depression
330

S. No. Ref. Methodology Features used Analysis Accuracy

method
1. [8] Power of band and alpha inter-hemispheric EEG band power, LR, SVM, 98.33% (LR), 96.8% (NB), 98.4% (SVM)
asymmetry used as feature. Z-score Inter-hemispheric NB
standardization was done. Most significant feature asymmetry
selected using Receiver Operating Characteristics
(ROC). LR, SVM and NB used as classifier
2. [4] Power spectrum used as feature and ANN is used Power spectrum, ANN 84%
as classifier FFT
3. [12] Linear SASI and nonlinear HFD used separately SASI, HFD t-test 85% (for both SASI and HFD)
as feature and t-test is used for analysis
4. [22] High-frequency noise is removed using Total RWE, WE ANN 98.11%
Variation Filtering (TVF). Classification of
frequency bands was done using 6-level
multi-resolution decomposition method for
Discrete Wavelet Transform (DWT). RWE used
for feature extraction in different decomposition
level. ANN used as classifier
5. [3] 4 nonlinear feature and 4 band power used for EEG band power, 4 LDA, LR, 90% (LR + 4 nonlinear features), 80% (KNN + 4
feature extraction. Feature selection done using nonlinear features KNN nonlinear features), 86.6% (LDA + 4 nonlinear
GA. KNN, LDA, and LR used as classifier using (DFA, HFD, CD features), 70% (KNN + alpha band power),
both set of features separately and LE) 73.3% (LR and LDA + alpha band power)
6. [14] Band power is used for feature extraction. LDA EEG band power DT 80%
and GA for feature reduction and selection. DT
used as classifier
7. [24] Cross-correlation is used for feature extraction for Cross-correlation SVM 95.5%
which 5 features extracted (Peak Value (PV),
instant at which peak occurs (IP), Equivalent
Width (EW), Centroid (C) and Mean Square
(continued)
S. Mahato and S. Paul
Table 2 (continued)
S. No. Ref. Methodology Features used Analysis Accuracy
method
Abscissa (MSA)) and 5 D feature vector is
formed. SVM is for classifying epilepsy patients
and normal subjects
8. [6] The author has proposed a new method: Kernel Kernel SVM 91.67% (KEFB-CSP + SVM)
Eigen-Filter-Bank Common Spatial Patterns Eigen-Filter-Bank
(KEFB-CSP) for feature extraction. KPCA is used Common Spatial
for feature reduction. LDA, KNN and SVM used Patterns
as classifier (KEFB-CSP)
9. [20] WT used for feature extraction. Most significant Wavelet transform LR 87.5%
features are selected using ROC. LR used as
classifier to predict the response of
antidepressant’s treatment outcome in MDD
patients
10. [18] Linear SASI and nonlinear DFA used for feature DFA, SASI LDA, 76.5% (SASI), 70.6% (DFA), 91.2%
extraction in Pz channel. LDA and Mann– Mann– (SASI + DFA)
Whitney statistical test was used as the classifier. Whitney
Performance of SASI and DFA are compared and statistical
Electroencephalogram (EEG) Signal Analysis for Diagnosis …

combination of SASI and DFA Also used as test

feature
11. [5] Classification of schizophrenic, depression and Lep–Ziv complexity BP ANN 60–80%
normal subjects in 4 different states (eyes closed
and 3 cognitive tasks) is done using Lep–Ziv
complexity for feature extraction and BP ANN as
classifier
331
Table 3 Summary of previous works related to EEG
332

S. No. Ref. Methodology Features used Analysis method Accuracy

1 [25] DWT used for splitting signal into Set of statistical features (Each SVM + LDA, 100% (SVM + LDA), 99.5%
sub-band. Set of statistical features were sub-band’s standard deviation, SVM + PCA, (SVM + ICA), 98.75%
used as feature. PCA, ICA and LDA absolute values mean, wavelet SVM + ICA (SVM + PCA), 98% (SVM)
used for feature reduction. SVM used average power, adjacent sub-band’s
for classifying individual as healthy or ratio of absolute mean)
having epileptic seizure
2 [26] Four feature extraction methods are (AR) model parameters, band power, SVM + bidirectional 99.38% (SVM + bidirectional
used: (AR) model parameters, band fractal dimension and wavelet energy search, search), 84.62%
power, fractal dimension and wavelet LDA + bidirectional (LDA + bidirectional search),
energy. Bidirectional search and plus-L search, SVM + plus-L 99.54% (SVM + plus-L minus-R),
minus-R techniques are used to select minus-R, LDA + plus-L 88.32% (SVM + plus-L minus-R)
most informative channel. GA used for minus-R
feature selection. LDA and SVM used
for classifying schizophrenic and normal
subjects
3 [1] Band power was used as feature. PCA Band power BP ANN, self-organizing 60–80% (BP ANN), 40–60%
used for dimension reduction. BP ANN, competitive network (Self-organizing competitive
self-organizing competitive network for network)
classifying subject to normal or
schizophrenic
4 [27] DWT is used for splitting into sub Statistical feature (Mean, Median, k-means clustering Highest 83.3% (alpha band
bands. Statistical features and frequency Mode, Standard deviation, feature + k-means clustering)
domain features are used. GA used for Maximum, Minimum) Frequency
feature reduction. k-means clustering domain features (Band power, FD
used for identifying abnormal brain’s and Energy)
EEG signal
5 [28] DWT is used for splitting into sub Statistical feature such as variance, SVM 98.75%
bands. Statistical feature of each energy, maximum and minimum
sub-band used as input to SVM sample value in power spectral
classifier which detects seizure signal density
(continued)
S. Mahato and S. Paul
Table 3 (continued)
S. No. Ref. Methodology Features used Analysis method Accuracy
6 [29] ICA with 5 different methods (Linear Power spectrum, variance and mean Multilayered perceptron 88.75% (MLPNN), 83.75% (PNN),
trends, Spectral pattern, Extreme values, of the signal neural network 82.2% (SVM)
Kurtosis, Data Improbability) was used (MLPNN), Probablistic
for artifact removal. DWT done for neural network (PNN),
decomposition of signal. Power SVM
spectrum, variance and mean of the
signal used for feature extraction.
MLPNN, PNN and SVM used to
classify cursor movements
7 [23] Entropy and Standard deviation used as Entropy, Standard deviation MLPNN, SVM (Radial 100% (MLPNN and SVM)
input feature to NB, ANN and SVM Basis Function)
(using linear, quadratic polynomial, and
Gaussian radial basis function) for
classifying epileptic patients
8 [30] Feature extracted with power spectral Power spectral density ANN 98.8%
density and input to ANN for epileptic
seizure detection
9 [31] Approximate Entropy (ApEn) and Auto Approximate Entropy (ApEn) and t-test, 80%
Mutual Information (AMI) was used for Auto Mutual Information (AMI)
Electroencephalogram (EEG) Signal Analysis for Diagnosis …

feature extraction-test is used for

identifying Alzheimer’s disease patients
10 [32] Autoregressive (AR) parameter are used Autoregressive coefficients ANN 88.89%
as input features to ANN for identifying
psychiatric disorder(obsessive
compulsive disorder and schizophrenia
333
334 S. Mahato and S. Paul

computer-aided system for identifying depression patients. This system could be of

great help in identifying depression patients in early stages preventing it from
becoming severe and untreatable which in terms would improve quality of life of
the patient and substantially decrease the cost associated with medical care.

References

1. Li Y, Fan F (2005) Classification of schizophrenia and depression by EEG with ANNs. In:
27th annual international conference of engineering in medicine and biology society,
IEEE-EMB, 2005, pp 1–6
2. Stewart JL, Coan JA, Towers DN, Allen JJB (2014) Resting and task-elicited prefrontal EEG
alpha asymmetry in depression: support for the capability model. Psychophysiology, 446–455
3. Hosseinifarda B, Moradia MH, Rostami R (2013) Classifying depression patients and normal
subjects using machine learning techniques and nonlinear features from EEG signal. Comput
Methods Programs Biomed 109:339–345
4. Mantri S, Agrawal P, Patil D, Wadhai V (2015) Non invasive EEG signal processing
framework for real time depression analysis. In: SAI intelligent systems conference,
pp 518–521
5. Fan F, Li Y, Qiu Y, Zhu Y (2005) Use of ANN and complexity measures in cognitive EEG
discrimination. In: Engineering in medicine and biology society, 27th annual international
conference, IEEE-EMBS, 2005, pp 1–6
6. Liao S, Wu C, Huang H, Cheng W, Liu Y (2017) Major depression detection from EEG
signals using kernel eigen-filter-bank common spatial patterns. Sensors, pp 1–14
7. Hinrikus H, Suhhova A, Bachmann M, Aadamsoo K, Vohma U, Lass J, Tuulik V (2009)
Electroencephalographic spectral asymmetry index for detection of depression. Med Biomed
Eng Comput, pp 1291–1299
8. Mumtaz W, Xia L, Ali SSA, Yasin MMAM, Hussain M, Malik AS (2017)
Electroencephalogram (EEG)-based computer-aided technique to diagnose major depressive
disorder (MDD). Biomed Signal Process Control 31:108–115
9. World Health Organization (2017) Depression and other common mental disorders global
health estimates. WHO Document Production Services, Geneva, Switzerland
10. World Health Organization (2011) Global burden of mental disorders and the need for a
comprehensive, coordinated response from health and social sectors at the country level.
Report by the Secretariat. EB 130/9
11. American Psychiatric Association (1994) Diagnostic and statistical manual of mental
disorders, 4th edn. American Psychiatric Association Washington, DC, pp 339–345
12. Bachmann M, Lass J, Suhhova A, Hinrikus H (2013) Spectral asymmetry and Higuchi’s
Fractal dimension measures of depression electroencephalogram. Comput Math Methods
Med 2013:1–8
13. Grin-Yatsenko VA, Baas I, Ponomarev VA, Kropotov JD (2010) Independent component
approach to the analysis of EEG recordings at early stages of depressive disorders. Clin
Neurophysiol 121:281–289
14. Mohammadi M et al (2015) Data mining EEG signals in depression for their diagnostic value.
BMC Med Inf Decis Mak, pp 108–123
15. Bruder GE, Stewart JW, Hellerstein D, Alvarenga JE, Alschuler D, McGratha PJ (2012)
Abnormal functional brain asymmetry in depression: evidence of biologic commonality
between major depression and dysthymia. Psychiatry Res, pp 250–254
Electroencephalogram (EEG) Signal Analysis for Diagnosis … 335

16. Bjork MH, Sand T, Bråthen G, Linaker OM, Morken G, Nilsen BM, Vaaler AE (2008)
Quantitative EEG findings in patients with acute, brief depression combined with other
fluctuating psychiatric symptoms: a controlled study from an acute psychiatric department.
BMC Psychiatry, 2008, pp 1–6
17. Ricardo-Garcell J (2009) EEG sources in a group of patients with major depressive disorders.
Int J Psychophysiol 71:70–74
18. Bachmann M, Lass J, Hinrikus H (2017) Single channel EEG analysis for detection of
depression. Biomed Signal Process Control 31:391–397
19. Subasi A (2007) EEG signal classification using wavelet feature extraction and a mixture of
expert model. Expert Syst Appl 32:1084–1093
20. Mumtaz W, Xia L, Yasin MAM, Ali SSA, Malik AS (2016) A wavelet-based technique to
predict treatment outcome for major depressive disorder. PLoS ONE, pp 1–6
21. Li Y, Li Y, Tong S, Tang Y, Zhu Y (2007) More normal EEGs of depression patients during
mental arithmetic than rest. In: Joint meeting of the 6th international symposium on
noninvasive functional source imaging of the brain and heart and the international conference
on functional biomedical imaging, 2007, pp 165–168
22. Puthankattil SD, Joseph PK (2012) Classification of EEG signals in normal and depression
conditions by ANN using RWE and signal entropy. J Mech Med Biol 12:1240019–1240032
23. Sood M, Bhooshan SV (2014) Automatic processing of EEG signals for Seizure detection
using soft computing techniques. In: IEEE international conference on recent advances and
innovations in engineering, 2014, pp 1–6
24. Chandaka S, Chatterjee A, Munshi S (2009) Cross-correlation aided support vector machine
classifier for classification of EEG signals. Expert Syst Appl 36:1329–1336
25. Subasi A, Gursoy MI (2010) EEG signal classification using PCA, ICA, LDA and support
vector machines. Expert Syst Appl 37:8569–8666
26. Sabetia M, Boostani R, Katebi SD, Price GW (2007) Selection of relevant features for EEG
signal classification of schizophrenic patients. Biomed Signal Process Control 2:122–134
27. Kalaivani M, Kalaivani V, Devi AV (2014) Analysis of EEG signal for the detection of brain
abnormalities. In: IJCA proceedings on international conference on simulations in computing
nexus pp 1–6
28. Kumar RSS, Jose JP (2011) Seizure detection in EEG using time frequency analysis and
SVM. In: International conference on emerging trends in electrical and computer technology
(ICETECT), IEEE, pp 1–6
29. Kousarrizi MRN (2009) Feature extraction and classification of EEG signals using wavelet
transform, SVM and artificial neural networks for brain computer interfaces. In: International
joint conference on bioinformatics, systems biology and intelligent computing, IEEE, 2009,
pp 352–355
30. Tzallas AT, Tsipouras MG, Fotiadis DI (2007) The use of time-frequency distributions for
epileptic seizure detection in EEG recordings. In: Proceedings of the 29th annual international
conference of the IEEE EMBS, 2007, pp 3–6
31. Abásolo D, Hornero R, Escudero J, Gomez C, Garcia M, Lopez M (2006) Approximate
entropy and auto mutual information analysis of the electroencephalogram in Alzheimer’s
disease patients. In: IET 3rd international conference on advances in medical, signal and
information processing, IEEE, 2006, pp 1–6
32. Tsoi C, So DSC, Sergejew A (1993) Classification of electroencephalogram using artificial
neural networks. In: 7th NIPS conference conference: advances in neural information
processing systems 6, 1993, pp 1151–1158