(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 14, No. 3, 2023

Improved Multiclass Brain Tumor Detection using

Convolutional Neural Networks and Magnetic
Resonance Imaging
Mohamed Amine Mahjoubi1, Soufiane Hamida2, Oussama El Gannour3, Bouchaib Cherradi4, Ahmed El Abbassi5,
Abdelhadi Raihani6
EEIS Laboratory, ENSET of Mohammedia, Hassan II University of Casablanca, Mohammedia 28830, Morocco 1, 2, 3, 4, 6
GENIUS Laboratory, SupMTI of Rabat, Rabat, Morocco 2
STIE Team, CRMEF Casablanca-Settat, Provincial Section of El Jadida, El Jadida 24000, Morocco 4
ERTTI Laboratory, FST of Errachidia, My Ismail University, Errachidia, Morocco 5

Abstract—Recently, Deep learning algorithms, particularly Since DL is a subset of machine learning, it does not use any
Convolutional Neural Networks (CNNs), have been applied manually created features [8]. In several disciplines, the use of
extensively for image recognition and classification tasks, with DL and ML for segmentation, detection, prediction,
successful results in the field of medicine, such as in medical classification and early diagnosis using medical data has been
image analysis. Radiologists have a hard time categorizing this promoted [9]–[25]. ML and DL as fields of Artificial
lethal illness since brain tumors include a variety of tumor cells. Intelligence (AI) find their applications in many others field
Lately, methods based on computer-aided diagnostics claimed to such as handwritten recognition and natural language
employ magnetic resonance imaging to help with the diagnosis of processing [26]–[29].
brain cancers (MRI). Convolutional Neural Networks (CNNs)
are often used in medical image analysis, including the detection Deep Learning (DL) is a subfield of Machine Learning
of brain cancers. This effort was motivated by the difficulty that (ML) that uses artificial neural networks with multiple layers to
physicians have in appropriately detecting brain tumors, model complex patterns and relationships in data [30], [31].
particularly when they are in the early stages of brain bleeding. Image classification is a task in computer vision where a model
This proposed model categorized the brain image into four is trained to identify and categorize objects in images [32],
distinct classes: (Normal, Glioma, Meningioma, and Pituitary). [33]. In the context of brain tumors, convolutional neural
The proposed CNN networks reach 95% of recall, 95.44% networks (CNNs) are often used for image classification [34].
accuracy and 95.36% of F1-score.
A CNN is a type of neural network specifically designed for
Keywords—Deep learning; convolutional neural networks;
image processing and analysis. It uses convolutional layers to
brain tumor; classification; magnetic resonance imaging detect local patterns and features in images, allowing it to learn
and classify objects within an image [35]. In the case of brain
I. INTRODUCTION tumor classification, a CNN is trained on a large dataset of
medical images of brains, where each image is labeled as
A brain tumor develops as a result of the overproduction
containing a tumor or not. The CNN then uses the features it
and proliferation of cells in the skull. The body's command
has learned to classify new, unseen images as containing a
center, the brain, can be burdened by tumors, which can also be
tumor or not. This can aid for the early diagnosis and treatment
harmful to a person's health [1]. It has been reported in the
of brain tumors.
study that brain tumors account for between 85% and 90% of
all major Central Nervous System (CNS) tumors [2]. CNN was initially utilized in 1980 [36], [37]. In essence, it
Radiologists have extensively used the medical imaging is a disguised Multilayer Perceptron (MLP) network. CNN's
technique for tumor detection [3], [4]. Among the current computational capacity is based on a model of the human
medicalities, MRI is the method of choice for brain tumors brain. Humans use an object's visual appearance to detect and
because of its astronomical nature. Radiologists routinely identify it. Tens of thousands of photos of the same item are
manually detect brain cancers. Depending on the radiologist's used to educate our kids how to distinguish objects. This aids a
level of training and experience, the tumor grading process can youngster in recognizing or foreseeing things that they have
take a while. The interpretation is expensive and wrong. The never encountered before. Similar in operation, CNN is well
associated challenges are attributed to specific traits, such as recognized for processing images.
the substantial variation in form, dimensions, and magnitude
for the same tumor type. Additionally, various diseases have In this paper, we proposed CNN-based model for medical
similar appearances [5], [6]. A successful Computer Aided image analysis, which is a sophisticated algorithm that utilizes
Diagnosis (CAD) system requires the development of feature DL techniques to categorize brain images into four distinct
extraction [7]. This is a challenging process that necessitates classes: Normal, Glioma, Meningioma, and Pituitary. These
prior knowledge of the domain problem because the accuracy four classes encompass the most common types of brain
of the classification depends on the correctly extracted features. tumors and are critical in providing accurate and precise
diagnosis and treatment. The model's impressive performance

406 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 3, 2023

metrics, including a recall rate of 95%, an accuracy of 95.44%, fully automated since they required a human to manually
and an F1-score of 95.36%, demonstrate the model's ability to identify tumor locations.
detect brain tumors with high precision and accuracy. The
recall rate, which is a measure of the model's ability to The authors in [40] described a technique for improving
correctly identify all positive cases, is at an impressive 95%, classification performance. To put the suggested method to the
indicating that the model has a low false-negative rate. test on a big dataset, the authors employ three feature
Additionally, the accuracy score, which measures the model's extraction methods: intensity histogram, gray level co-
ability to classify the images correctly, is also at a remarkable occurrence matrix (GLCM), and bag-of-words model (BoW).
95.44%. Finally, the F1-score, which is a combined measure of Using the enlarged tumor region as the ROI enhances the
the model's precision and recall, is at a remarkable 95.36%, intensity histogram, GLCM, and BoW model accuracies by
indicating the model's overall performance in categorizing 82.31% vs. 71.39%, 84.75% vs. 78.18%, and 88.19% vs.
brain images. These high-performance metrics are critical in 83.54%, respectively. Ring partitioning can improve accuracy
medical image analysis as they help physicians diagnose and by up to 87.54%, 89.72%, and 91.28% in addition to increasing
treat brain tumors accurately and efficiently. The proposed region. These experimental findings demonstrate that the
model is an essential tool for radiologists and physicians, as it proposed strategy for classifying brain cancers in T1-weighted
reduces the subjectivity involved in manually interpreting CE-MRI is both possible and effective.
medical images and provides a reliable and objective method In [41], the authors suggested a method for classifying
for diagnosis. brain tumors utilizing a set of deep characteristics and ML
The remainder of this article is structured as follows: classifiers. In the proposed framework, they get deep
Section II explores the related works to the search for brain characteristics by brain magnetic resonance (MR) pictures
tumor detection. In Section III, the proposed CNN-based using the notion of transfer learning and numerous pre-trained
model is presented, including its architecture, training deep convolutional neural networks. The top three deep
methodology, and evaluation metrics. Section IV provides an features for multiple ML classifiers are chosen and
concatenated into a deep feature set, which is then fed through
overview of the materials and methods utilized in the study.
Section V presents the experimental results, and in Section VI, multiple ML classifiers to predict the final output.
the study's findings and conclusions are discussed. The authors in [42] proposed a method to improve the
segmentation of brain tumors in magnetic resonance imaging
II. RELATED WORKS (MRI) by incorporating an additional classification network.
DL and AI play a crucial role in MRI The segmentation is performed using a convolutional neural
image processing through techniques such as segmentation, network (CNN) which is trained on a dataset of MRI images
recognition, and categorization. They are also employed in with corresponding tumor masks. The CNN is then combined
the classification and detection of brain cancer. Numerous with a classification network that is trained to distinguish
studies have been conducted on the identification and between different types of tumors. The output of the
segmentation of brain MRI images. A review of classification network is used to refine the segmentation results
international literature was conducted to assess the use of by identifying and removing false positives. The authors
DL in identifying and categorizing brain tumors. evaluate their method on two publicly available datasets and
compare it to other state-of-the-art methods. They report that
In [38], the authors used newly designed CapsNets to allow their method achieves higher accuracy and Dice similarity
CapsNet to access neighboring tissues while remaining focused coefficient (DSC) scores than the other methods. They also
on the core target. As a result, a modified CapsNet architecture perform ablation studies to analyze the contribution of the
for brain tumor classification is presented, with coarse tumor classification network and find that it significantly improves
boundaries incorporated as extra inputs into its pipeline to the segmentation results.
increase CapsNet's focus. The proposed method outperforms
its competitors significantly. In [43], the authors provide an overview of the challenges
in brain tumor segmentation and discuss how deep learning
The authors in [39] used a convolutional neural network to techniques have been applied to address these challenges. The
perform multimodal brain tumor categorization for early survey covers various types of deep learning models, including
diagnosis (CNN). With an accuracy of 92.66 %, the CNN convolutional neural networks (CNNs), recurrent neural
model can categorize brain cancers into 5 types (Normal, networks (RNNs), and generative adversarial networks
Glioma, Meningioma, Pituitary, and Metastatic). The grid (GANs). The authors also discuss the advantages and
search optimization approach is used to automatically define limitations of these methods and provide insights into future
critical hyperparameters in CNN models. The suggested CNN directions for research.
model is compared to popular cutting-edge CNN models such
as AlexNet, Inceptionv3, ResNet-50, VGG-16, and GoogleNet. III. THE PROPOSED CNN-BASED MODEL
Using huge publicly available clinical datasets, satisfactory
classification results are produced. This methodology has a The structure of a CNN-based model can be customized
number of drawbacks that may be listed as follows, despite the and optimized by selecting specific hyperparameters. These
fact that the recommended methods for classifying brain hyperparameters play a crucial role in determining the overall
tumors differ. The accuracy provided by current techniques is architecture of the model, including the number of
unsatisfactory due to the significance of MRI classification in convolutional layers, the activation functions used in each
the medical field. Some classification algorithms could not be layer, and the number of hidden units per layer. The number of

407 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 3, 2023

convolutional layers determines the depth of the neural

network and its ability to extract high-level features from the
input image. The more convolutional layers the model has, the
more complex and sophisticated features it can extract, leading
to better performance. However, adding too many layers can
also result in overfitting, where the model becomes too
specialized to the training data and performs poorly on new
data.
The choice of activation functions also affects the model's
Fig. 2. The proposed CNN model architecture.
ability to extract features accurately. Activation functions
introduce non-linearities to the model, allowing it to model
Brain tumor detection is a critical problem in medical
complex relationships between the input and output. Common
imaging, and a Convolutional Neural Network (CNN) can be
activation functions used in CNNs include ReLU, sigmoid, and
used as a detection system. A CNN is a type of DL algorithm
tanh, among others. The choice of activation function can
that is designed to analyze image data and can be trained to
significantly impact the model's performance, and selecting the
recognize specific features in the images. In the case of brain
right activation function is critical in optimizing the model's
tumor detection, a CNN is trained using a large dataset of
accuracy. After extensive experimentation, we arrived at the
medical images that includes both normal and abnormal scans.
optimal selection of hyperparameters for the multi-layered
The CNN is then used to identify and locate areas of the brain
model depicted in Fig. 1. This is chosen in: Five layers of
that may contain a tumor. This is done by processing the
convolution2D that differ in the number of filters, so that it
images and identifying certain patterns and features that are
doubles as you go deeper, starting with the first layer, which
characteristic of brain tumors. The output of the CNN can be
contains 32 filters, and ending with the last layer, which
used to assist medical professionals in the diagnosis of brain
contains 512 filters, all to extract many features; five layers of
tumors and to guide further testing and treatment. The use of a
max pooling in order to extract important information from the
CNN as a brain tumor detection system has the potential to
previous convolution2D layer; flattening layer, to render the
improve the accuracy and speed of diagnosis, which can be
information in one dimension; then a Dense layer with 128
crucial in treating brain tumors effectively.
units; finally, the Dance layer with 4 units, due to the number
of final classes was used in which the SoftMax function was IV. MATERIALS AND METHODS
used because it is the most used function in multi-class models
(in our case 4 classes). In all these layers, the ReLU activation A. Dataset Collection
function was used because it is the most advanced compared to The "Brain Tumor MRI Dataset" on Kaggle is a collection
the other functions. Fig. 2 displays the layout of our proposed of magnetic resonance imaging (MRI) scans of the brain. The
CNN architecture. dataset includes MRI scans of both healthy individuals and
individuals with brain tumors, which have been annotated by
medical experts. The annotations indicate the presence,
location, and type of brain tumor in the MRI scans. In Figure 3,
sample images from the dataset are depicted.
In fact, a publicly available Kaggle database is utilized, as
is described in this section, which also provides some
information on the database. The following three datasets were
combined to generate this one: Figshare1, SARTAJ dataset2,
Br35H3.

Fig. 3. Some sample images of the dataset.

1
https://figshare.com/articles/dataset/brain_tumor_dataset/1512427
2
https://www.kaggle.com/datasets/sartajbhuvaji/brain-tumor-
classification
Fig. 1. Flowchart of a System for Detecting Brain Tumors. 3
https://www.kaggle.com/datasets/ahmedhamada0/brain-tumor-detection

408 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 3, 2023

The 7022 MRI scans of the human brain in this dataset are solved and can have a significant impact on the performance
divided into four classes. The dataset can be used for a variety of the CNN. It's important to carefully choose the activation
of research purposes, including developing and evaluating function in a CNN, as the wrong choice can lead to slow
computer vision algorithms for brain tumor detection and training and convergence, or even prevent the network from
classification, as well as for training ML models for medical
learning the desired patterns in the data [5].
diagnosis and treatment planning. However, it is important to
note that the generalizability of the models developed using 3) Pooling layer: It comes after the convolution layer.
this dataset should be carefully evaluated, as it is based on a This layer's task is to shrink the feature map, which implies
limited sample of MRI scans. fewer computations and parameters are required to operate the
network. Therefore, it can be claimed that a summary of the
B. Background on CNN features is the output of this layer. There are several
DL models use a hierarchical framework to learn high- techniques to pool data; in this instance, max pooling was
level abstractions from input images [44]. Due to the utilized. The feature map's most objects that are covered by
availability of large-scale annotated datasets and the fact that the filters are selected using max pooling [5].
CNN has demonstrated to be the most effective DL method for
4) Classification layer: In our CNN architecture, the
assessing medical pictures. The well-known CNN models [45]
ImageNet, AlexNet, VGG16, GoogLeNet, Inception-V3 and classification layer comes last. This extensively used classifier
ResNet101 have made significant strides in image recognition. which is a fully - connected feed-forward network. The
However, the area of medical imaging lacks a comparable neurons in the layers with complete connections are all
annotated dataset. Medical image classification using CNN is connected to the neurons in the layer below. By merging the
frequently done using one of two methods [46]. The first is traits of the preceding layers, this layer recognizes the input
called "learning from the ground up," while the second is image and predicts classes based on it. A total number of
called "Transfer Learning." The network layers that make up output classes is based on the number of classes in the source
the CNN include a convolution layer, an activation layer, a dataset. In this paper, the classification layer uses the
Maxpooling layer, and a classification layer. Each of these "SoftMax" activation function to classify the features created
levels is explained as follows [47], [48].
from the input images in the previous layer into separate
1) Convolution layer: The convolutional layer comes first groups depending on the training data [5].
[7]. This layer is in charge of identifying an input word's C. Confusion Matrix and Evaluation Metrics
attributes. This step merely combines the entry neuron by a
A confusion matrix is a table used to evaluate the
filter depending on the input and requirement to produce the
performance of a classification model by summarizing the
feature map [49]. A neural activation function is used to predicted and actual class labels of a dataset. It displays the
introduce nonlinearity. The animal visual cortex served as a number of true positives (TP), false positives (FP), true
model for CNN computing. It decodes visual data and has a negatives (TN), and false negatives (FN) for each class. In a
fine sensitivity to the input's smallest subregions [50]. The binary classification problem, the confusion matrix is a 2x2
convolutional layer's main elements are its receptive field, table with two rows and two columns representing the actual
stride, dilation, and padding [51]. Fig. 4 displays the and predicted classes. The first row represents the actual
application of the convolutional layer. negative and the second row represents the actual positive. The
first column represents the predicted negative and the second
column represents the predicted positive. For a multi-class
classification problem, the confusion matrix is an n x n table,
where n is the number of classes. The rows represent the actual
classes and the columns represent the predicted classes. Each
cell in the matrix represents the number of instances that
belong to a particular actual class and a particular predicted
class [52]. Based on the entries in the confusion matrix, several
Fig. 4. Convolution layer application. evaluation metrics can be computed to assess the performance
of the classifier. Some commonly used evaluation metrics
2) Activation layer: The activation function is a crucial include:
component of a Convolutional Neural Network (CNN) model P N
ur y (1)
as it defines the output of a neuron in response to a given P N FP FN
input. The activation function determines the range of values P
Pr ision (2)
that the neuron can output and allows the model to introduce P FP

non-linearity into the decision boundary, making it possible to P

ll (3)
learn complex relationships between the input and output P FN

variables. Common activation functions used in CNNs include ll Pr ision

F -s or (4)
ll Pr ision
the Rectified Linear Unit (ReLU), the Sigmoid function, and
N
the Hyperbolic Tangent (Tanh) function. The choice of Pr ision (5)
N FP
activation function depends on the specific problem being

409 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 3, 2023

The results of a confusion matrix can be used to calculate This table describes the architecture of a CNN for image
various performance metrics, such as accuracy, precision, classification. The table lists each layer in the network, its type
recall, and F1 score. A high number of true positives and true (e.g., Conv2D, MaxPooling2D, Dense, Flatten), the output
negatives indicate a high accuracy of the model, while a high shape of the layer, and the number of parameters in that layer.
number of false positives and false negatives indicate low The network starts with an input layer that takes in an image of
accuracy. Precision indicates the proportion of positive shape (224,224,3), which means the image has a height and
predictions that are actually correct, while recall indicates the width of 224 pixels and 3 color channels (representing RGB
proportion of actual positive instances that were correctly values). The next layer is a Conv2D layer with 32 filters, which
predicted. The F1 score is the harmonic mean of precision and means it will learn 32 different feature maps from the input
recall, and is a good overall indicator of a model's image. The MaxPooling2D layer performs down-sampling by
performance. In summary, a confusion matrix provides detailed taking the maximum value in a region of the feature map and
information about the performance of a model, while reduces its size. This process is repeated several times with
evaluation metrics provide a concise, single value increasing number of filters in the Conv2D layer and reducing
representation of the model's performance [53]. the size of the feature map in the MaxPooling2D layer. Finally,
the feature maps are flattened into a 1D vector and passed
V. EXPERIMENTAL RESULTS through two dense (fully connected) layers to output the final
A. Algorithms Best Parameters prediction. The numbers of parameters in each layer, along
with the total number of trainable and non-trainable parameters
The main objective of this project is to build a system are also listed. The total number of parameters in the network
capable of classifying medical images and giving correct is 7,789,484, and all of them are trainable. To train the model,
decisions with a large percentage. All this depends on the we need training data and for validation also we need test data.
quality of the existing data, the convolutional neural network, Once we are satisfied with the test result of the model, we can
the number of its layers and coefficients, and the layers used to use it to make predictions on new data. In addition, for the
process the data needed to train the network. properties and parameters of the model starting with the
Our study proposed a CNN architecture model; we number of epochs =50 (epochs are the number of typical
retrieved the input 224×224 MRI image data with RGB color repetitions of the training).
channels having a batch size of 32 by our CNN model. To configure our training model, we need to call the
Initially, we added five convolutional layers; in addition, there compile method with the loss function we want to use. The
are five max-pooling layers, one flattening layer, and activation type of optimization and the metrics our model should evaluate
(ReLU). The model develops the capacity to produce during training and testing we will use: The Adam optimizer is
hierarchical qualities automatically by use of a succession of due to the fact that, as mentioned earlier, it is one of the most
hidden layers. A final class label is determined by using a widely used optimizers because it is the fastest method and
Softmax function on the outputs of this layer. In this proposed also converges quickly to correct for learning rate latency and
model, we have an output layer that generates a high contrast. The categorical cross entropy loss function is
quadridimensional vector representing four different used when there are multiple label classes. Metrics is a
classifications of cerebral tumors. Table I displays the function used to evaluate the performance of your model.
summary of our model CNN proposed architecture.
The best parameters for algorithms in CNNS involve a
TABLE I. THE SUMMARY OF MODEL DESCRIPTION variety of elements, such as learning size, batch size, optimiser,
loss function, activation functions, and epochs. Different
Layer (type) Output (shape) Parameters parameters have to be tuned based on the network architecture,
Input layer (224, 224, 3) 0 dataset, and task. Table II presented the best parameters of our
Conv2D (224, 224, 32) 896 proposed model.
MaxPooling2D (112,112,32) 0 TABLE II. THE BEST HYPER-PARAMETERS USED FOR THE TL MODELS IN
Conv2D (224,224,32) 18496 TRAINING PHASE

MaxPooling2D (112,112,64) 0 Learning Batch Loss

Network Optimizer Epochs
rate Size Function
Conv2D (56,56,128) 73856 Proposed Categorical
1.00e10-4 32 Adam 50
MaxPooling2D (28,28,128) 0 model cross entropy
Conv2D (28,56,256) 295168 B. Training Results
MaxPooling2D (14,14,256) 0 Our experiment was based on a dataset of brain MRI
Conv2D (14,14,512) 1180160 images. We trained a CNN model to detect whether the MRI
MaxPooling2D (7,7,512) 0
image contains a tumor (glioma, meningioma and pituitary
tumor) or not. Finally, we compared the diagnostic and
Flatten (None, 25088) 0 computational results of our model with the related works. On
Dense (None, 128) 3211392 our training data set, our proposed model has a 95.44%
Dense (None, 4) 516 accuracy rate. The graph depicts the accuracy and loss during
the construction and validation phases of our proposed CNN
model are presented in Fig. 5 and Fig. 6.

410 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 3, 2023

The graphs show the performance of the model in terms of tumor, and pituitary. The ROC curve, as shown in Fig. 8,
accuracy and loss as it is being developed and tested. Accuracy displays the true positive rate, also known as recall, against the
refers to the measure of how well the model is able to correctly false positive rate. The proposed model demonstrated excellent
predict the outcomes or classes of the input data. It is usually performance on the brain MRI dataset, accurately classifying a
expressed as a percentage, where higher values indicate better high number of samples in each class. For instance, 148 out of
performance. Loss, on the other hand, is a measure of how 156 samples in the glioma class and 193 out of 195 samples in
much error there is between the predicted outputs of the model the no tumor class were correctly classified. Overall, the results
and the actual outputs. Lower values of loss indicate that the demonstrate the model's effectiveness in accurately detecting
model is more accurate in its predictions. brain cancers, particularly in the early stages of brain bleeding.
However, there are also some misclassified samples, as shown
in the entries in the off-diagonal cells of the matrix. For
example, five samples from the glioma class were misclassified
as meningioma, and 3 samples from the pituitary class were
misclassified as no tumor.
Based on this confusion matrix, it is possible to compute
various evaluation metrics such as accuracy, precision, recall,
and F1-score presented in Table III to quantify the performance
of the proposed model. These metrics can provide a more
comprehensive understanding of the model's performance and
help identify areas for improvement.

Fig. 5. Accuracy of the proposed model.

Fig. 7. The confusion matrix of the proposed model.

Fig. 6. Loss of the proposed model.

These graphs show the accuracy and loss of the CNN

model at different stages of its development. The construction
phase likely refers to the phase where the model is being
trained on a dataset to learn the patterns and relationships in the
data. The validation phase refers to the phase where the model
is being tested on a separate set of data to evaluate its
performance and generalizability.
C. Testing Results
The proposed model achieved an impressive accuracy of
95.44% in the detection of brain cancers. Table 4 presents the
detection performance results of the model with an 80% train
and 20% test split, showing the confusion matrix of the
classifier model, which displays the number of correctly and
incorrectly classified samples in each class. The confusion
matrix, depicted in Figure 7, reveals the model's strong
performance across all four classes: glioma, meningioma, no Fig. 8. ROC curves findings for the proposed model.

411 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 3, 2023

TABLE IV. CLASSIFICATION REPORT OF THE PROPOSED MODEL optimized for classifying cerebral tumors based on MRI
images, and it may not be effective in other medical image
Classes Precision Recall F1-score Support
analysis tasks. Moreover, like any deep learning model, the
Glioma 0.91 0.95 0.93 156 proposed CNN model requires a significant amount of data to
Meningioma 0.94 0.89 0.95 173 be trained effectively. While the model achieved high accuracy
Notumor 0.96 0.99 0.98 195
rates on the specific dataset used, it may not perform as well on
smaller or lower-quality datasets. Additionally, the model may
Pituitary 0.99 0.98 0.99 178 be computationally expensive, which may limit its practical
application in certain settings. Table IV shows the comparison
This table represents the evaluation metrics of a classifier
of the proposed model's performance with other methods based
model that has been trained on a medical imaging dataset with
on classification precision. Our proposed model achieved the
four different classes: Glioma, Meningioma, No tumor, and
highest accuracy of 95.44% compared to previous work in this
Pituitary. The metrics shown are precision, recall, F1-score,
field, which shows the superior quality of the model.
and support. Precision is the fraction of true positive
predictions among all positive predictions, recall is the fraction TABLE V. COMPARISON WITH PREVIOUS WORK
of true positive predictions among all actual positive cases, F1-
score is the harmonic mean of precision and recall, and support Authors Method/ Brain tumor data Accuracy (%)
is the number of samples in each class. The results show that [42] Segmentation using a CNN 89.99
the model performs well in all classes with F1-scores ranging
from 0.93 to 0.99, indicating that the model has good precision [38] Capsule networks/MRI images 90.89
and recall values. [54] CNN + Augmenting/ MRI images 91.28

D. Discussion [39] CNN Multi-classifying / MRI images 92.66

The proposed CNN model achieved the highest accuracy [41] Deep feature + ML classifiers/ MRI images 93.72
rate among other deep neural models for classifying cerebral [43] CNN, RNN and GANs 95
tumors into four classes on MRI images. This model has the This work CNN Multi-classifying/MRI images 95.44
potential to aid medical professionals in diagnosing brain
tumors accurately and efficiently. The architecture of the VI. CONCLUSIONS AND PERSPECTIVES
model is designed for optimal performance on medical image
datasets, making it a promising tool in the field of medical To summarise, the proposed CNN model for classifying
image analysis. Comparison with other studies that used cerebral tumors based on MRI images is a promising tool for
various techniques, such as capsule networks, data aiding medical professionals in diagnosing brain tumors
augmentation and partitioning, and feature sets with ML accurately and efficiently. The high accuracy rate achieved by
classifiers, showed that DCNNs are the most effective the proposed model compared to other deep neural models
technique for classifying brain tumors based on MRI scans. indicates the effectiveness of DCNNs in classifying brain
The accuracy of the models ranged from 90.89% to 95.44%, tumors based on MRI scans. The proposed model's architecture
indicating the effectiveness of deep learning techniques in this is designed for optimal performance on medical image
domain. However, it is essential to note that the datasets, making it a promising tool in the field of medical
generalizability of these models to other datasets and clinical image analysis. However, it is essential to recognize the
scenarios should be evaluated with care. As with any ML limitations of the proposed model, such as the need to carefully
model, its effectiveness is dependent on the quality and size of evaluate its generalizability to other datasets and clinical
the dataset used to train and validate it. Thus, the performance scenarios, as well as the potential computational cost and data
of the proposed model should be assessed on other datasets, requirements. Future research should focus on developing
and it should be tested under different clinical scenarios before more robust and accurate models by exploring alternative
being deployed in real-world settings. The proposed method is approaches and using larger and more diverse datasets.
also applicable for different MRI classifications, and it can be Additionally, the proposed model's applicability for
combined with transfer learning techniques to achieve more different MRI classifications and potential incorporation of
accurate classifications in the future. Transfer learning can help other types of medical imaging data, such as CT scans, PET
reduce the amount of data needed to train a model, and it can scans, or MRS data, offer exciting possibilities for improving
improve the accuracy of the model by leveraging the the accuracy and effectiveness of medical image analysis. The
knowledge learned from other datasets. development of more sophisticated models and techniques will
While the proposed CNN model for classifying cerebral lead to more accurate and efficient diagnosis and treatment of
tumors based on MRI images appears to be highly effective, cerebral tumors, ultimately benefiting patients and medical
there are several limitations to consider. First and foremost, the professionals alike.
model's accuracy and effectiveness may be limited to the Our future research will focus on developing more robust
specific dataset used to train and validate it. To ensure its and accurate models by exploring alternative approaches and
generalizability, it is crucial to evaluate the model on other using larger and more diverse datasets. It may also be worth
datasets and clinical scenarios. Another limitation of the investigating the potential for incorporating other types of
proposed model is that it may not be suitable for classifying medical imaging data.
other types of tumors or medical conditions. It is important to
recognize that the model was specifically designed and

412 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 3, 2023

REFERENCES Patients with Pneumonia using X-r y Lung Im g s,” dv. s i. t hnol.
eng. syst. j., vol. 5, no. 5, pp. 167–175, 2020, doi: 10.25046/aj050522.
[1] F. J. Díaz-Pernas, M. Martínez-Zarzuela, M. Antón-Rodríguez, and D. [17] H. Mouj hid, B. Ch rr di, nd L. B h tti, “Convolutional Neural
González-Ort g , “ D p L rning pproach for Brain Tumor Networks for Multimodal Brain MRI Images Segmentation: A
Classification and Segmentation Using a Multiscale Convolutional Comp r tiv Study,” in Sm rt ppli tions nd D t n lysis, vol.
N ur l N twork,” H lth r , vol. 9, no. , p. 53, F b. 0 , doi: 1207, M. Hamlich, L. Bellatreche, A. Mondal, and C. Ordonez, Eds.
10.3390/healthcare9020153. Cham: Springer International Publishing, 2020, pp. 329–338. doi:
[2] M. L. Bondy t l., “Br in tumor pid miology: Cons nsus from th 10.1007/978-3-030-45183-7_25.
Brain Tumor Epid miology Consortium,” C n r, vol. 3, no. S7, pp. [18] H. Moujahid, B. Cherradi, M. Al-S r m, nd L. B h tti, “Di gnosis of
1953–1968, Oct. 2008, doi: 10.1002/cncr.23741. COVID-19 Disease Using Convolutional Neural Network Models Based
[3] Y. Zh ng, . Li, C. P ng, nd M. W ng, “Improv Gliobl stom r nsf r L rning,” in Innov tiv Syst ms for Int llig nt H lth
Multiforme Prognosis Prediction by Using Feature Selection and Informatics, vol. 72, F. Saeed, F. Mohammed, and A. Al-Nahari, Eds.
Multipl K rn l L rning,” IEEE/ACM Trans. Comput. Biol. and Cham: Springer International Publishing, 2021, pp. 148–159. doi:
Bioinf., vol. 13, no. 5, pp. 825–835, Sep. 2016, doi: 10.1007/978-3-030-70713-2_16.
10.1109/TCBB.2016.2551745. [19] O. rr d , . ih ni, O. Bou tt n , nd B. Ch rr di, “Fuzzy
[4] . L. Si g l, K. D. Mill r, nd . J m l, “C n r st tisti s, 0 5: cardiovascular diagnosis system using clinical dat ,” in 0 8 4th
C n r St tisti s, 0 5,” C : C n r Journ l for Clini i ns, vol. 65, International Conference on Optimization and Applications (ICOA),
no. 1, pp. 5–29, Jan. 2015, doi: 10.3322/caac.21254. Mohammedia, Apr. 2018, pp. 1–4. doi: 10.1109/ICOA.2018.8370549.
[5] P. iw ri t l., “CNN B s d Multi l ss Br in umor D t tion Using [20] S. Hamida, O. E. Gannour, B. Cherradi, H. Ouajji, and A. Raihani,
M di l Im ging,” Comput tion l Int llig n nd N uros i n , vol. “Optimiz tion of M hin L rning lgorithms Hyper-Parameters for
2022, pp. 1–8, Jun. 2022, doi: 10.1155/2022/1830010. Improving the Prediction of Patients Infected with COVID- 9,” in 0 0
[6] Department of Computer Applications, National Institute of Technology, IEEE 2nd International Conference on Electronics, Control,
iru hir p lli, milN du, Indi , K vith . S, nd P. J. . lphons , “ Optimization and Computer Science (ICECOCS), Kenitra, Morocco,
Hybrid Cryptosyst m to Enh n S urity in Io H lth C r Syst m,” Dec. 2020, pp. 1–6. doi: 10.1109/ICECOCS50124.2020.9314373.
IJWMT, vol. 9, no. 1, pp. 1–10, Jan. 2019, doi: [21] O. Asmae, R. Abdelhadi, C. Bouchaib, S. Sara, and K. Tajeddine,
10.5815/ijwmt.2019.01.01. “P rkinson’s Dis s Id ntifi tion using KNN nd NN lgorithms
[7] E.-S. A. El-Dahshan, H. M. Mohsen, K. Revett, and A.-B. M. Salem, b s d on Voi Disord r,” in 0 0 st Int rn tion l Conf r n on
“Comput r-aided diagnosis of human brain tumor through MRI: A Innovative Research in Applied Science, Engineering and Technology
surv y nd n w lgorithm,” Exp rt Syst ms with ppli tions, vol. 4 , (IRASET), Meknes, Morocco, Apr. 2020, pp. 1–6. doi:
no. 11, pp. 5526–5545, Sep. 2014, doi: 10.1016/j.eswa.2014.01.021. 10.1109/IRASET48871.2020.9092228.
[8] P. iw ri t l., “CNN B s d Multi l ss Br in umor D t tion Using [22] O. El Gannour, S. Hamida, B. Cherradi, A. Raihani, and H. Moujahid,
M di l Im ging,” Comput tion l Int llig n nd N uros i n , vol. “P rform n Ev lu tion of r nsf r L rning hniqu for utom ti
2022, pp. 1–8, Jun. 2022, doi: 10.1155/2022/1830010. Detection of Patients with COVID-19 on X- y Im g s,” in 0 0 IEEE
2nd International Conference on Electronics, Control, Optimization and
[9] B. Cherradi, O. Terrada, A. Ouhmida, S. Hamida, A. Raihani, and O.
Computer Science (ICECOCS), Kenitra, Morocco, Dec. 2020, pp. 1–6.
Bou tt n , “Comput r-Aided Diagnosis System for Early Prediction of
doi: 10.1109/ICECOCS50124.2020.9314458.
Atherosclerosis using Machine Learning and K-fold cross-v lid tion,” in
2021 International Congress of Advanced Technology and Engineering [23] O. Daanouni, B. Cherr di, nd . miri, “Pr di ting di b t s dis s s
(ICOTEN), Taiz, Yemen, Jul. 2021, pp. 1–9. doi: using mix d d t nd sup rvis d m hin l rning lgorithms,” in
10.1109/ICOTEN52080.2021.9493524. Proceedings of the 4th International Conference on Smart City
Applications, Casablanca Morocco, Oct. 2019, pp. 1–6. doi:
[10] K. Usm n nd K. jpoot, “Br in tumor l ssifi tion from multi-
10.1145/3368756.3369072.
mod lity M I using w v l ts nd m hin l rning,” P tt rn n l
Applic, vol. 20, no. 3, pp. 871–881, Aug. 2017, doi: 10.1007/s10044- [24] O. Terrada, B. Cherradi, S. Hamida, A. Raihani, H. Moujahid, and O.
017-0597-8. Bouattane, “Pr di tion of P ti nts with H rt Dis s using rtifi i l
N ur l N twork nd d ptiv Boosting t hniqu s,” in 0 0 3rd
[11] S. Hamida, O. El Gannour, B. Cherradi, A. Raihani, H. Moujahid, and
International Conference on Advanced Communication Technologies
H. Ou jji, “ Nov l COVID-19 Diagnosis Support System Using the
and Networking (CommNet), Marrakech, Morocco, Sep. 2020, pp. 1–6.
Stacking Approach and Transfer Learning Technique on Chest X-Ray
doi: 10.1109/CommNet49926.2020.9199620.
Im g s,” Journ l of H lth r Engin ring, vol. 2021, pp. 1–17, Nov.
2021, doi: 10.1155/2021/9437538. [25] L. Hu , Y. Gu, X. Gu, J. Xu , nd . Ni, “ Nov l Br in M I Im g
Segmentation Method Using an Improved Multi-View Fuzzy c-Means
[12] O. rr d , B. Ch rr di, . ih ni, nd O. Bou tt n , “ th ros l rosis
Clust ring lgorithm,” Front. N uros i., vol. 5, p. 66 674, M r. 0 ,
dis s pr di tion using Sup rvis d M hin L rning hniqu s,” in
doi: 10.3389/fnins.2021.662674.
2020 1st International Conference on Innovative Research in Applied
Science, Engineering and Technology (IRASET), Meknes, Morocco, [26] S. Hamida, B. Cherradi, O. Terrada, A. Raihani, H. Ouajji, and S.
Apr. 2020, pp. 1–5. doi: 10.1109/IRASET48871.2020.9092082. L ghm ti, “ Nov l F tur Extr tion Syst m for Cursiv Word
Vocabulary Recognition using Local Features Descriptors and Gabor
[13] D. Lamrani, B. Cherradi, O. E. Gannour, M. A. Bouqentar, and L.
Filt r,” in 0 0 3rd Int rn tion l Conf r n on Advanced
B h tti, “Br in umor D t tion using M I Im g s nd Convolutional
Communication Technologies and Networking (CommNet), Marrakech,
N ur l N twork,” IJ CS , vol. 3, no. 7, 0 , doi:
Morocco, Sep. 2020, pp. 1–7. doi:
10.14569/IJACSA.2022.0130755.
10.1109/CommNet49926.2020.9199642.
[14] S. Laghmati, B. Cherradi, A. Tmiri, O. Daanouni, and S. Hamida,
[27] S. H mid , B. Ch rr di, nd H. Ou jji, “H ndwritt n r bi Words
“Cl ssifi tion of P ti nts with Br st C n r using N ighbourhood
Recognition System Based on HOG and Gabor Filter D s riptors,” in
Component Analysis and Supervised Machin L rning hniqu s,” in
2020 1st International Conference on Innovative Research in Applied
2020 3rd International Conference on Advanced Communication
Science, Engineering and Technology (IRASET), Meknes, Morocco,
Technologies and Networking (CommNet), Marrakech, Morocco, Sep.
Apr. 2020, pp. 1–4. doi: 10.1109/IRASET48871.2020.9092067.
2020, pp. 1–6. doi: 10.1109/CommNet49926.2020.9199633.
[28] S. Hamida, B. Cherradi, O. El Gannour, O. Terrada, A. Raihani, and H.
[15] O. El G nnour t l., “Con t n tion of Pr -Trained Convolutional
Ou jji, “N w D t b s of Fr n h Comput r S i n Words H ndwritt n
Neural Networks for Enhanced COVID-19 Screening Using Transfer
Vo bul ry,” in 0 Int rn tion l Congr ss of dv n d hnology
L rning hniqu ,” El troni s, vol. , no. , p. 03, D . 0 , doi:
and Engineering (ICOTEN), Taiz, Yemen, Jul. 2021, pp. 1–5. doi:
10.3390/electronics11010103.
10.1109/ICOTEN52080.2021.9493438.
[16] H. Moujahid, B. Cherradi, O. E. Gannour, L. Bahatti, O. Terrada, and S.
[29] S. Hamida, O. El Gannour, B. Cherradi, H. Ouajji, and A. Raihani,
H mid , “Convolution l N ur l N twork B s d Cl ssifi tion of
“H ndwritt n omput r s i n words vo bul ry r ognition using

413 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 14, No. 3, 2023

on t n t d onvolution l n ur l n tworks,” Multim d ools ppl, [41] J. K ng, Z. Ull h, nd J. Gw k, “M I-Based Brain Tumor Classification
Nov. 2022, doi: 10.1007/s11042-022-14105-2. Using Ensemble of Deep F tur s nd M hin L rning Cl ssifi rs,”
[30] S. Hamida, B. Ch rr di, . ih ni, nd H. Ou jji, “P rform n Sensors, vol. 21, no. 6, p. 2222, Mar. 2021, doi: 10.3390/s21062222.
Evaluation of Machine Learning Algorithms in Handwritten Digits [42] H. . Nguy n, . . L , . V. Nguy n, nd N. . Nguy n, “Enh n ing
ognition,” in 0 9 st Int rn tion l Conf r n on Sm rt Syst ms MRI Brain Tumor Segmentation with an Additional Classification
and Data Science (ICSSD), Rabat, Morocco, Oct. 2019, pp. 1–6. doi: N twork,” 0 0, doi: 10.48550/ARXIV.2009.12111.
10.1109/ICSSD47982.2019.9003052. [43] Z. Liu t l., “D p l rning b s d br in tumor s gm nt tion: surv y,”
[31] O. El Gannour, B. Cherradi, S. Hamida, M. Jebbari, and A. Raihani, Complex Intell. Syst., vol. 9, no. 1, pp. 1001–1026, Feb. 2023, doi:
“S r ning M di l F M sk for Coron virus Pr v ntion using D p 10.1007/s40747-022-00815-5.
L rning nd utoML,” in 0 nd Int rn tion l Conf r n on [44] S. Sh rm t l., “D p L rning Mod l for utom tic Classification
Innovative Research in Applied Science, Engineering and Technology nd Pr di tion of Br in umor,” Journ l of S nsors, vol. 0 , pp. –11,
(IRASET), Meknes, Morocco, Mar. 2022, pp. 1–7. doi: Apr. 2022, doi: 10.1155/2022/3065656.
10.1109/IRASET52964.2022.9737903.
[45] S. Jun j , . Jun j , G. Dhim n, S. B hl, nd S. K utish, “ n ppro h
[32] K. . Moh m d, E. Els m hy, nd . S l m, “COVID-19 Disease for Thoracic Syndrome Classification with Convolutional Neural
Detection based on X-Ray Image Classification using CNN with GEV N tworks,” Comput M th M thods M d, vol. 0 , p. 3900 54, S p.
Activation Fun tion,” IJ CS , vol. 3, no. 9, 0 , doi: 2021, doi: 10.1155/2021/3900254.
10.14569/IJACSA.2022.01309103.
[46] S. Jun j , . Jun j , G. Dhim n, S. B hl, nd S. K utish, “ n ppro h
[33] K. Ej z t l., “S gm nt tion M thod for P thologi l Br in umor nd for Thoracic Syndrome Classification with Convolutional Neural
ur t D t tion using M I,” ij s , vol. 9, no. 8, 0 8, doi: N tworks,” Comput tional and Mathematical Methods in Medicine, vol.
10.14569/IJACSA.2018.090851. 2021, pp. 1–10, Sep. 2021, doi: 10.1155/2021/3900254.
[34] D. K. Sahoo, S. Mishr , nd M. N. Moh nty, “Br in umor [47] C. Monga, D. Gupta, D. Prasad, S. Juneja, G. Muhammad, and Z. Ali,
Segmentation and Classification from MRI Images using Improved “Sust in bl N twork by Enh n ing ttribut -Based Selection
FLICM Segmentation and SCA Weight Optimized Wavelet-ELM Mechanism Using Lagrange Int rpol tion,” Sust in bility, vol. 4, no.
Mod l,” IJ CS , vol. 3, no. 7, 0 , doi: 10, Art. no. 10, Jan. 2022, doi: 10.3390/su14106082.
10.14569/IJACSA.2022.0130753.
[48] M. J obson nd C. H dg oth, “L v ls of 5,6-dihydrouridine in relaxed
[35] H. Surrisyad and W. -, “ Fast Military Object Recognition using nd hlor mph ni ol tr nsf r ribonu l i id,” Bio h mistry, vol. 9, no.
Extr m L rning ppro h on CNN,” IJ CS , vol. , no. , 0 0, 12, pp. 2513–2519, Jun. 1970, doi: 10.1021/bi00814a018.
doi: 10.14569/IJACSA.2020.0111227.
[49] S. Kanwal, J. Rashid, J. Kim, S. Juneja, G. Dhiman, and A. Hussain,
[36] P. M. M h, I. Sk ln , M. Eri , nd . John, “Influ n of Int rn t of “Mitig ting th Co xist n hniqu in Wir l ss Body r
things on human psychology (internet of thoughts) for education, N tworks By Using Sup rfr m Int rl ving,” IE E Journ l of
h lth r , nd busin ss s,” E I Endors d r ns Mob Com ppl, vol. Research, pp. 1–15, Mar. 2022, doi: 10.1080/03772063.2022.2043788.
7, no. 2, p. e1, Aug. 2022, doi: 10.4108/eetmca.v7i2.2627.
[50] K. Kour t l., “Controlling gronomi V ri bl s of S ffron Crop Using
[37] . Dh nkh r, S. Jun j , . Jun j , nd V. B li, “K rn l P r m t r Io for Sust in bl gri ultur ,” Sust in bility, vol. 4, no. 9, rt. no.
Tuning to Tweak the Performance of Classifiers for Identification of 9, Jan. 2022, doi: 10.3390/su14095607.
H rt Dis s s:,” Int rn tion l Journ l of E-Health and Medical
Communications, vol. 12, no. 4, pp. 1–16, Jul. 2021, doi: [51] R. Qian, S. Sengan, and S. Jun j , “English l ngu g t hing b s d on
10.4018/IJEHMC.20210701.oa1. big data analytics in augmentative and alternative communication
syst m,” Int J Sp h hnol, vol. 5, no. , pp. 409–420, Jun. 2022,
[38] P. fsh r, K. N. Pl t niotis, nd . Moh mm di, “C psul N tworks for doi: 10.1007/s10772-022-09960-1.
Brain Tumor Classification Based on MRI Images and Coarse Tumor
Bound ri s,” in IC SSP 0 9 - 2019 IEEE International Conference on [52] O. El Gannour, S. Hamida, S. Saleh, Y. Lamalem, B. Cherradi, and A.
Acoustics, Speech and Signal Processing (ICASSP), Brighton, United ih ni, “COVID-19 Detection on X-Ray Images using a Combining
Kingdom, May 2019, pp. 1368–1372. doi: Mechanism of Pre-tr in d CNNs,” Int rn tion l Journ l of dv n d
10.1109/ICASSP.2019.8683759. Computer Science and Applications, pp. 564–570, 2022.
[39] E. Irm k, “Multi-Classification of Brain Tumor MRI Images Using [53] H. D li nis, “Ev lu tion M tri s nd Ev lu tion,” in Clini l xt
D p Convolution l N ur l N twork with Fully Optimiz d Fr m work,” Mining, Cham: Springer International Publishing, 2018, pp. 45–53. doi:
10.1007/978-3-319-78503-5_6.
Iran J Sci Technol Trans Electr Eng, vol. 45, no. 3, pp. 1015–1036, Sep.
2021, doi: 10.1007/s40998-021-00426-9. [54] J. Ch ng t l., “Enh n d P rform n of Br in umor Cl ssifi tion
[40] Y. Yang et al., “Gliom Gr ding on Conv ntion l M Im g s: D p via Tumor Region Augm nt tion nd P rtition,” PLoS ONE, vol. 0, no.
10, p. e0140381, Oct. 2015, doi: 10.1371/journal.pone.0140381.
L rning Study With r nsf r L rning,” Front N uros i, vol. , p.
804, 2018, doi: 10.3389/fnins.2018.00804.

414 | P a g e
www.ijacsa.thesai.org