ǀĂŝůĂďůĞŽŶůŝŶĞĂƚǁǁǁ͘ƐĐŝĞŶĐĞĚŝƌĞĐƚ͘ĐŽŵ

 ǀĂŝůĂďůĞŽŶůŝŶĞĂƚǁǁǁ͘ƐĐŝĞŶĐĞĚŝƌĞĐƚ͘ĐŽŵ

^ĐŝĞŶĐĞŝƌĞĐƚ
^ĐŝĞŶĐĞŝƌĞĐƚ
Available online
Procedia at www.sciencedirect.com
Computer Science 00 (2023) 000–000
Procedia Computer Science 00 (2023) 000–000
www.elsevier.com/locate/procedia
www.elsevier.com/locate/procedia
ScienceDirect
Procedia Computer Science 233 (2024) 703–712

5th International Conference on Innovative Data Communication Technologies and Application (ICIDCA 2024)
5th International Conference on Innovative Data Communication Technologies and Application (ICIDCA 2024)
Improvement
Improvement of
of Brain
Brain Tumor
Tumor Categorization
Categorization using
using Deep
Deep
Learning: A Comprehensive Investigation and Comparative
Learning: A Comprehensive Investigation and Comparative
Analysis
Analysis
T. Lakshmi Prasanthi, N. Neelima
T. Lakshmi Prasanthi, N. Neelima
Department of Electronics and Communication Engineering, Amrita School of Engineering,Bengaluru, Amrita Vishwa Vidyapeetham,India.
Department of Electronics and Communication Engineering, Amrita School
[email protected], of Engineering,Bengaluru, Amrita Vishwa Vidyapeetham,India.
[email protected]
[email protected], [email protected]

Abstract
Abstract
A brain tumor is a critically severe health disorder that requires an accurate and timely diagnosis for effective treatment. Advances
A
in brain
medicaltumor is a critically
imaging and deepsevere health
learning disorder
methods thatshown
have requires an accurate
potential and timelythe
for enhancing diagnosis for effective
identification treatment. Advances
and categorization of brain
in medical imaging and deep learning methods have shown potential for enhancing the identification and
cancers throughout the years. In the present research, our study compares the accuracy of eight different deep learning models categorization of in
brain
the
cancers throughout
classification the years.
of brain In the
tumors present research,
employing brain MRIour study compares
data that involvethe accuracy of eight
Densenet121, different deep
EfficientNet B7,learning models in the
InceptionResNetV2,
classification of
Inception_V3, brain tumorsVGG16,
RestNet50V2, employing brainandMRI
VGG19, data that
Xception. involveimprove
To further Densenet121, EfficientNet
performance, B7, InceptionResNetV2,
we propose integrating a hybrid
deep learning technique. Efficient and timely diagnosis of brain tumors is critical for the treatment of patients,integrating
Inception_V3, RestNet50V2, VGG16, VGG19, and Xception. To further improve performance, we propose a hybrid
and our study aims
deep learning technique. Efficient and timely diagnosis of brain tumors is critical for the treatment of patients,
to achieve high recall, accuracy, and F1-score in this context. With a precision of 96.63%, our innovative convolutional neural and our study aims
to achieve
network high technique
(CNN) recall, accuracy,
achieved and F1-score results
outstanding in this incontext. With diagnosis.
brain tumor a precisionAlso,
of 96.63%,
our study our innovativethe
investigates convolutional neural
unique capabilities
network
of certain(CNN)
models,technique achievedand
such as VGG19 outstanding results in brain
InceptionResNetV2, andtumor diagnosis. Also,
their possibilities our study
for better investigates
glioma the unique
tumor detection capabilities
efficiency. Our
of certain models, such as VGG19 and InceptionResNetV2, and their possibilities for better glioma tumor detection
results, in particular, provide insight into possible uses of deep learning frameworks, including the integration of hybrid techniques, efficiency. Our
results, in particular, provide insight into possible uses of deep learning frameworks, including
in medical imaging, offering an innovative approach for increased brain tumor detection and identification. the integration of hybrid techniques,
in medical imaging, offering an innovative approach for increased brain tumor detection and identification.
© 2024
© 2024 The
The Authors.
Authors. Published
Published by
by Elsevier
ELSEVIER B.V.B.V.
© 2024
This The
is an
an Authors.
open accessPublished by ELSEVIER
article under
under the CC B.V.
CC BY-NC-ND
BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
(https://creativecommons.org/licenses/by-nc-nd/4.0)
This
This is open
is an open access
access article
article under the
the scientific
CC BY-NC-ND license
license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review
Peer-review under
under responsibility
responsibility of
of the
the scientific committee
committee ofthe
of the 5thInternational
5th International Conference
onon Innovative Data
Peer-review
Communicationunder responsibility
Technologies of the scientific
and Application committee of the 5th InternationalConference
Conference Innovative
on Data
Innovative Communication
Data
Technologies and Application
Communication Technologies and Application
Keywords: Brain Tumor, Deep Learning Algorithms, Medical Imaging, Image Classification, Neural Network Models;
Keywords: Brain Tumor, Deep Learning Algorithms, Medical Imaging, Image Classification, Neural Network Models;

1. Introduction
1. Introduction
Cancer is one of the most widespread illnesses on the globe, with an estimated 1.8 million new diagnoses and 600,000
Cancer is one of the most widespread illnesses on the globe, with an estimated 1.8 million new diagnoses and 600,000
deaths in the US in 2020. Cancer is characterized by the unregulated development of abnormal cells within the body. It
deaths in thechanges
arises from US in 2020. Cancer
in cellular is characterized
function or defectsby themake
that unregulated development
it more difficult for a of
cellabnormal cells
to undergo within thecell
controlled body. It
death
arises from changes in cellular function or defects that make it more difficult for a cell to undergo controlled cell death
[1]. This affects many tissues and organs and promotes the growth of malignancies. The growth can push on brain
[1]. This
tissue, affectsinmany
resulting tissues
serious and organs
problems. even ifand
the promotes the growth
tumor is harmless. of tumor
Brain malignancies.
cause aboutThe 2.17%
growthofcan push on
all cancer brain
deaths,
tissue, resulting in serious problems. even if the tumor is harmless. Brain tumor cause about 2.17% of all cancer deaths,

1877-0509 © 2024 The Authors. Published by ELSEVIER B.V.

1877-0509 © 2024
This is an open Thearticle
access Authors. Published
under by ELSEVIER
the CC BY-NC-ND B.V.(https://creativecommons.org/licenses/by-nc-nd/4.0)
license
This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the scientific committee of the 5th International Conference on Innovative Data Communication Technologies
Peer-review under responsibility of the scientific committee of the 5th International Conference on Innovative Data Communication Technologies
and Application
and Application
1877-0509 © 2024 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0)
Peer-review under responsibility of the scientific committee of the 5th International Conference on Innovative Data
Communication Technologies and Application
10.1016/j.procs.2024.03.259
704 T. Lakshmi Prasanthi et al. / Procedia Computer Science 233 (2024) 703–712
2 T. Lakshmi Prasanthi / Procedia Computer Science 00 (2019)
000–000
and the five-year survival rate for glioblastoma is weak, at roughly 5.6%. The incidence of brain tumor and the worrying
statistics have driven ongoing study on the issue, with doctors and scientists seeking strategies to avoid tumor, create
more effective medications, develop better diagnostic tools, and develop better methodologies to assess and categorize
tumor. This study incorporates novel approaches for investigating brain architecture as well as the creation of AI
technologies [3].
In recent years, the merging of AI technologies and healthcare imaging has shown significant potential for enhancing
the precision and efficacy of diagnosing a variety of medical disorders. The detection and classification of brain tumor
in magnetic resonance imaging (MRI) images is a critical focus. The efficiency of neural networks, the advantages of
transfer learning for healthcare imaging, the impact of data augmentation on the efficiency of models, and the possible
benefits of the cross-transformer network will all be thoroughly discussed.

2. Literature review

The realm of brain tumor analysis via MRI imaging has been extensively explored in recent studies. Kumar et al.
[4] conducted a meticulous investigation into segmentation techniques, revealing discrepancies in accuracy and
efficiency across methods. Advocating for hybrid segmentation involves combining the strengths of different methods
and adapting methodologies to accommodate diverse tumor types and imaging scenarios. Liu et al. [5] introduced
transformer-based approaches to medical image analysis, improving diagnostic precision, though challenges inadapting
transformers and managing computational intensity were acknowledged. ZainEldin H et al. [6] proposed a novel
technique using deep learning and optimization, showing promise in detection and classification, but dataset diversity
and generalization challenges arose. Amin, J. et al.'s [7] survey highlighted machine learning methods' promise in brain
tumor detection and classification, suggesting larger, diverse datasets and advanced algorithms to address data
imbalance and generalization issues. Anagun et al. [8] evolved a brain tumor detection method that utilizesCNN with
promising accuracy, while emphasizing dataset diversity, class imbalance, and interpretability. Integrating insights from
these studies reveals the evolving landscape of brain tumor analysis, calling for hybrid methods, domain- specific
strategies, diverse datasets, optimization refinement, and explainability techniques to enhance accuracy and patient
care. Studies by Badza and Barjaktarovic [9], Khan et al. [11], and Hassan Ali Khan et al. [13] collectively highlight
the limitations posed by small datasets, emphasizing data augmentation as a pivotal tool to augment sample diversity
and combat overfitting. A consistent theme across Srinivasan et al. [10] and Khan et al. [11] is the imperativefor dataset
diversity and the mitigation of biases, emphasizing the need for larger, inclusive datasets and techniques to ensure
equitable model performance. The optimization of CNN architectures and hyperparameters, as highlighted by Badza
and Barjaktarovic [9] and Hassan Ali Khan et al. [10], emerges as a central avenue for enhancing model accuracyand
generalizability. Furthermore, transfer learning, advocated by Srinivasan et al. [10] and Khan et al. [11], offers a potent
means to leverage pre-existing knowledge and improve adaptability when confronting different tumor types andimaging
modalities. The challenge of interpretability and explainability, articulated by Masood et al. [12], underscoresthe need
for tailored solutions that bridge the gap between complex model outputs and clinical understanding. In synthesis, an
integrative approach that leverages data augmentation, optimizes architectures, promotes diversity, ensures fairness,
explores transfer learning, and enhances interpretability holds the potential to propel the accuracy andclinical relevance
of DL models in brain tumor categorization. Collective efforts towards enriched datasets and interpretable AI
methodologies are pivotal in driving advancements in this critical medical domain.
The realm of brain tumor identification and classification has been significantly propelled by the integration of DL
and ML approaches, as evidenced by a series of noteworthy studies. Biratu et al. [14] conducted an encompassing
survey of prevailing algorithms in this domain, unveiling challenges centered on precision and resilience. To surmount
these obstacles, the authors recommended the amalgamation of advanced machine learning methods and the utilization
of more expansive and diverse datasets, a consensus echoed by subsequent research. Mahmud et al. [15] pioneered
deep learning networks for brain tumor detection, achieving heightened accuracy, though constrained datasets and
potential overfitting surfaced as hurdles. This study underscored the need for partnerships with medical institutions to
access larger datasets and the application of data augmentation and transfer learning to enhance generalization. Tiwari
et al. [16] harnessed CNN for multiclass brain tumor detection, yielding encouraging accuracy. However, challenges
in accommodating tumor characteristic variations and image quality fluctuations were acknowledged, with solutions
 T. Lakshmi Prasanthi et al. / Procedia Computer Science 233 (2024) 703–712 705
T. Lakshmi Prasanthi / Procedia Computer Science 00 (2019) 3
000–000
3. Methodology

This study presents a method for categorized brain tumor using multiple deep learning frameworks such as
Densenet121, EfficientNet B7, InceptionResNetV2, Inception_V3, RestNet50V2, VGG16, VGG19, Xception, and a
custom CNN. There are three categorization modules to examine: "BTD”,"MRI-D," and "Augmented dataset." In both
the training and validation sets, the dataset is rigorously balanced among classes. The models display high feature
extraction skills after being pretrained on ImageNet. Accuracy, F1 score, sensitivity, specificity, and precision are
typical measures used to evaluate efficiency [8].

3.1. Dataset

Three datasets have been employed in this research: The Brain Tumor Dataset (BTD), the MRI-D dataset, and an
augmented dataset.
• BTD Dataset
The BTD dataset is sourced from Kaggle Table 1 shows the split of images in the training and validation categories
for both the "No Tumor" and "Tumor" groups, as well as the overall number of images within each folder. Sample
image of BTD Dataset and MRI-Dataset is shown in figure 1.

Table 1. Tumor and No tumor BTD Dataset.

Label No Tumor Tumor Total Images
Training Folder 1450 1450 2900
Validation Folder 1450 1450 2900

Fig. 1. Sample image of BTD Dataset and MRI-Dataset.

• MRI-Dataset

The number of validation and training images for each of the three classes is shown in Table 2. Each class has
1000 training and 1000 validation images, for an aggregate of 2000 images.

Table 2. Glioma, meningioma and Pituitary MRI-Dataset.

Label Glioma Meningioma Pituitary Total Images
Training Folder 1000 1000 1000 3000
Validation Folder 1000 1000 1000 3000

• Augmented dataset
Combining the MRI-D and BTD datasets, the augmented dataset is formed. The arrangement of images among
different folders in the brain tumor dataset is as follows in table 3: The folder names like “Glioma", "Meningioma”,
“No Tumor", and "Pituitary" comprised 1000 images each. The "Training Folder" contains 1000 images that are equally
spread out among the three tumor and one no tumor types. The "Validation Folder" contains 4000 images, which
replicates the arrangement of the training data, increasing the entire number of images in the dataset by over 4000 The
"Testing Folder" contains 300 images, which replicates the arrangement of the testing data, these further data for each
class are used for making confusion matrixes for all eight deep learning models.

Table 3. Augmented dataset.

Label Glioma Meningioma No_Tumor Pituitary Total Images
Training Folder 1000 1000 1000 1000 4000
Validation Folder 1000 1000 1000 1000 4000
Testing Folder 300 300 300 300 1200
706 T. Lakshmi Prasanthi et al. / Procedia Computer Science 233 (2024) 703–712
4 T. Lakshmi Prasanthi / Procedia Computer Science 00 (2019)
000–000
3.2. CNN Architecture

Convolutional neural networks (CNNs) are essential for image-based activities, especially medical imaging
applications like classifying brain tumors. To comprehend and classify intricate patterns in medical images, the
architecture depicted in the figure is intended to extract hierarchical characteristics from 3-channel MRI scans. CNN
Architecture is shown in figure 2.

Fig. 2. CNN Architecture.

3.3. DL techniques for brain tumor categorization

The presented methodology uses an advanced CNN-based method for classifying brain tumors, starting with the
acquisition of brain MRIs, which are the main diagnostic method for tumors [11]. MRI images are scaled and cropped
in order to provide consistent input. Then, using transfer learning rather than building a CNN from begin, data is
carefully divided into separate training, validation, and future test sets [12]. Major pre-trained models, such as DenseNet
121, EfficientNet B7, and InceptionV3, which were trained on large-scale datasets like ImageNet, are used to extract
features from MRI scans by identifying important patterns associated with tumor. The structure is expanded by adding
layers after feature extraction. A probability distribution across the different tumor classifications is provided by the
concluding layer, which has a Softmax activation and enables precise classification of tumor types such gliomas,
meningiomas, and pituitary [13]. A variety of measures are used in the evaluation of algorithmic performance, including
as precision, recall, accuracy, and F1-score after prediction. Deep Learning-Based Brain Tumor Classification Process
is shown in figure 3.

Fig. 3. Deep Learning-Based Brain Tumor Classification Process.

 T. Lakshmi Prasanthi et al. / Procedia Computer Science 233 (2024) 703–712 707
T. Lakshmi Prasanthi / Procedia Computer Science 00 (2019) 5
000–000
3.4. Performance evaluation

The hyperparameters for the augmented dataset's brain tumor classification are listed in Table 4. For effective model
training, a batch size of 16 was used, processing 16 data in each iteration. The algorithm was able to iteratively optimize
its parameters using the increased brain tumor dataset over the period of a 10-epoch training process. The output layer's
activation function, Softmax, generated probability distributions for different tumor types. The loss measure, which
was appropriate for the multi-class classification challenge, was categorical crossentropy, and the Adam optimizer
dynamically modified learning rates to improve convergence. In the case of brain tumor classification, these uniform
hyperparameters among various algorithms promoted a common training framework that allowed for insightful
comparisons.
Table 4: Hyperparameters for Augmented dataset
Batch size 16
Epoch 10
Activation Function Softmax
Optimizer Adam
Loss Function Categorical crossentropy

Our employ a number of fundamental measures [15], to assess the performance of our brain tumor categorization
method. These parameters include accuracy, F1 score, recall and precision.

4. Purpose

The goal of this study is to improve medical imaging, particularly the use of MRI data for brain tumor categorization.
A thorough assessment of eight deep learning models with a focus on accuracy, recall, precision, and F1-score is the
key objective. The detection of glioma tumors is critical for prompt and efficient treatment, and this is the main
emphasis of the study. By presenting a novel Convolutional Neural Network (CNN) technique, the study aims to set a
new benchmark for high-performance brain tumor detection. Understanding the performance of models like VGG19
and InceptionResNetV2 is aimed at expanding the knowledge base of deep learning frameworks in medical imaging.
Improving the detection of brain tumors is the ultimate aim, with potential applications for better patient outcomes and
medical treatments.

5.Major Research Findings

5.1. BTD Dataset

a) Evaluation of Pre-trained DL Algorithms with MRI-D Dataset

Figure 4 depicts how pre-trained models such as DenseNet121, InceptionResNetV2, and VGG16 perform in binary
tumor categorization as a result of exposure to varied ImageNet data. Due to a lack of exposure, models such as
EfficientNetB7 and InceptionV3, particularly without pre-training, perform poorly. These findings highlight the critical
significance of pre-training in the acquisition of useful picture attributes. Even when not pre-trained, EfficientNetB7
and InceptionV3 demonstrate potential, indicating their linked capabilities. The graph emphasizes the need for pre-
training for higher tumor classification accuracy, with several model selections based on the dataset and training
circumstances.
708 T. Lakshmi Prasanthi et al. / Procedia Computer Science 233 (2024) 703–712
6 T. Lakshmi Prasanthi / Procedia Computer Science 00 (2019)
000–000

Fig. 4. Evaluation of Pre-trained DL Algorithms for Binary Tumor Categorization.

4.1. MRI Dataset

a) DL Framework Effectiveness for Brain Tumor Identification

Xception with ImageNet weights is the most effective approach shown in Figure 5, continuously outperforming on
the basis of mean precision throughout every case. Its exceptional and constant performance outperforms all other
designs, making it the best option for brain tumor categorization. VGG16 and VGG19, both initiated with ImageNet
weights, show robust and reliable efficiency, frequently acquiring high mean accuracies. This constancy highlights their
aptitude for identifying brain cancers. When tested across situations, EfficientNetB7 proved to be the least accurate model,
continuously demonstrating the lowest mean precision. Image rotation improves accuracy in certain circumstances,
such as InceptionV3 and InceptionResNetV2, but decreases efficiency in others, such as DenseNet121 and
EfficientNetB7.

Fig. 5. Accuracy of Different Models Under Different Training Conditions.

4.2. Augmented Dataset

a) DL Framework Comparison for Brain Tumor Identification Using Augmented Dataset

InceptionResNetV2 consistently achieves the highest mean efficiency across all situations, making it the framework
with the highest median precision performance. This implies that InceptionResNetV2 is well-suited to categorizing
brain tumor pictures in the larger dataset. ResNet50V2, VGG16, and VGG19 operate as well, with consistently high
mean precision across all classes. EfficientNetB7 seems to be the least accurate model across all classes, with the lowest
mean accuracy on a consistent basis. The graph depicts the variation in reliability across every model and circumstance.
 T. Lakshmi Prasanthi et al. / Procedia Computer Science 233 (2024) 703–712 709
T. Lakshmi Prasanthi / Procedia Computer Science 00 (2019) 7
000–000
Models with minimal variability, such as VGG16 and VGG19, indicate more consistent effectiveness across all tumor
classifications. Model Accuracy with Augmented Dataset is shown in figure 6.

Fig. 6. Model Accuracy with Augmented Dataset.

b) Assessing the Effectiveness of DL Approaches for Classifying Brain Tumor

The InceptionResNetV2 (ImageNet) approach shows the best performance, with a fascinating precision of 0.990 when
pretrained using ImageNet weights. VGG16, trained on ImageNet, performs well and consistently, with a precision of
0.987. The precision of VGG19 pretrained on ImageNet is appropriate at 0.978, while the accuracy of ResNet50V2
pretrained on ImageNet is excellent at 0.982. InceptionV3 achieves a high precision of 0.943 when trained with data
enhancement (rotation), whereas Xception, trained using ImageNet, performs as well with a precision of 0.978.
DenseNet121, which was trained on ImageNet, performs moderately well, with a precision of 0.978. Models such as
ResNet50V2 with rotation-based data enhancement and Xception with rotation-based data enhancement show decent
but reduced accuracies. Despite any pre-training, InceptionV3 has the lowest precision of 0.443, while EfficientNetB7
(none) has the least efficiency of 0.257. The top algorithms are InceptionResNetV2, VGG16, and VGG19 pretrained
on ImageNet, whereas EfficientNetB7 and InceptionV3 without pre-training are the most desirable possibilities. Model
Performance in Brain Tumor Classification is shown in figure 7.

Fig. 7. Model Performance in Brain Tumor Classification.

5. Practical Implications

The innovative CNN model outperformed traditional approaches in the classification of brain tumors, achieving
exceptional metrics, including 96.63% accuracy, 85.33% recall, 92.33% precision, and a 97.00% F1-score. The
outstanding accuracy of the CNN model can be attributed to several key factors. Firstly, the model integrates robust
mechanisms for detecting features, specifically designed to capture intricate patterns and features from brain MRI
images associated with different tumor classes. The architecture, particularly the convolutional layers, facilitates the
extraction of relevant and discriminative features crucial for precise classification.Furthermore, the CNN model
demonstrates strong generalization capabilities through transfer learning from pre-trained models such as DenseNet121
710 T. Lakshmi Prasanthi et al. / Procedia Computer Science 233 (2024) 703–712
8 T. Lakshmi Prasanthi / Procedia Computer Science 00 (2019)
000–000
and EfficientNet B7. The selection of model architectures is based on their proven efficacy in image classification
tasks, contributing to the exceptional accuracy observed. Moreover, specialized architectures like InceptionResNetV2
and VGG19 exhibit a strong proficiency in identifying glioma tumors, highlighting the versatility of the approach.
These findings suggest a comprehensive strategy that balances model architecture, feature extraction, and
generalization, ultimately resulting in a highly accurate brain tumor classification system.

6. Originality/Value

6.1. Comparative Analysis of Brain Tumor Classification Models from augmented dataset

The unique CNN model outperformed all other brain tumor categorization approaches in our complete evaluation. As
analyzed with the other algorithms, it had the greatest accuracy (96.63%), recall (85.33%), precision (92.33%), and F1-
score (97.00%). This outstanding success may be attributed to the model's capacity to identify features from brain MRI
images, its durability in dealing with image noise and unpredictability, and its amazing generalization capability. Other
models performed admirably, with DenseNet121 and EfficientNet B7 exhibiting high accuracy rates, giving them viable
options for brain tumor categorization. The accuracy of InceptionResNetV2 and VGG19 for glioma tumor detection
was high. VGG19 and InceptionResNetV2 excelled in identifying glioma tumor with high precision. Although the
Xception and ResNet50V2 models performed well, the Custom CNN, DenseNet121, and EfficientNet B7 models
improved them marginally. Comparative Analysis of Brain Tumor Classification Models is shown in figure 8.
Performance Comparison of Brain Tumor Classification Models is shown in table 5.

Table 5. Performance Comparison of Brain Tumor Classification Models.

Model Name Metric Glioma Meningioma No_Tumor Pituitary

Accuracy 0.9229 0.9229 0.9229 0.9229
Densenet121 Recall 0.7033 0.9867 0.9067 0.9752
Precision 0.9769 0.6311 0.9963 0.7867
F1-Score 0.8178 0.7698 0.9494 0.8708
EfficientNet B7 Accuracy 0.6713 0.6713 0.6713 0.6713
Recall 0.8900 0.4800 0.0000 0.0000
Precision 0.3944 0.2753 0.0000 0.0000
F1-Score 0.5466 0.3499 0.0000 0.0000
Accuracy 0.9479 0.9479 0.9479 0.9479
InceptionResNetV2
Recall 0.7633 0.9667 0.9233 0.9300
Precision 0.9662 0.7379 0.9928 0.9588
F1-Score 0.8529 0.8369 0.9568 0.9442
Inception_V3 Accuracy 0.9246 0.9246 0.9246 0.9246
Recall 0.7400 0.8867 0.8133 0.9567
Precision 0.9487 0.6982 1.0000 0.8416
F1-Score 0.8315 0.7812 0.8971 0.8955
RestNet50V2 Accuracy 0.9408 0.9408 0.9408 0.9408
Recall 0.6833 0.9400 0.9500 0.9533
Precision 0.9809 0.7212 0.9661 0.9377
F1-Score 0.8055 0.8162 0.9580 0.9455
VGG16 Accuracy 0.9567 0.9567 0.9567 0.9567
Recall 0.7967 0.9533 0.9733 0.9300
Precision 0.9598 0.7793 0.9832 0.9721
F1-Score 0.8707 0.8576 0.9782 0.9506
VGG19 Accuracy 0.9508 0.9508 0.9508 0.9508
Recall 0.8333 0.8567 0.9267 0.9900
Precision 0.9058 0.8185 0.9929 0.9000
F1-Score 0.8681 0.8371 0.9586 0.9429
 T. Lakshmi Prasanthi et al. / Procedia Computer Science 233 (2024) 703–712 711
T. Lakshmi Prasanthi / Procedia Computer Science 00 (2019) 9
000–000
Xception Accuracy 0.9442 0.9442 0.9442 0.9442
Recall 0.7233 0.9367 0.9500 0.9433
Precision 0.9775 0.7454 0.9628 0.9279
F1-Score 0.8314 0.8301 0.9564 0.9355
Custom_CNN Accuracy 0.9663 0.9663 0.9663 0.9663
Recall 0.8533 0.9233 0.9700 0.9833
Precision 0.9808 0.8445 0.9898 0.9306
F1-Score 0.9127 0.8822 0.9798 0.9562

Fig. 8. Comparative Analysis of Brain Tumor Classification Models.

712 T. Lakshmi Prasanthi et al. / Procedia Computer Science 233 (2024) 703–712
10 T. Lakshmi Prasanthi / Procedia Computer Science 00 (2019)
000–000
7. Conclusion & Future scope

In this research on brain tumor categorization, our study examined the efficiency of a wide range of different models
other than the standard Convolutional Neural Network (CNN). Along with the CNN, extensively tested eight other
models, consisting of Densenet121, EfficientNet B7, InceptionResNetV2, Inception_V3, RestNet50V2, VGG16,
VGG19, and Xception. Despite our exclusive CNN model standing out as having the best performance, with a fantastic
accurate rate of 96.63% and great recall, precision, and F1-score, the other models all demonstrated notable capabilities.
For example, DenseNet121 and EfficientNet B7 have outstanding precision rates, resulting in viable choices for brain
tumor categorization. VGG19 and InceptionResNetV2 displayed outstanding accuracy in detecting glioma tumor.
Although algorithms such as Xception and RestNet50V2 fared effectively, our exclusive CNN, DenseNet121, and
EfficientNet B7 models improved a bit, highlighting the possibility of continuing tuning. This study not only presents
a cutting-edge CNN model but also provides vital insights into the various features of deep learning frameworks, with
the potential to improve rapid brain tumor detection and patient treatment in the future. Future research should strive
to modify and implement such models in real-world clinical settings to achieve measurable results.

References

[1] Kalaiselvi T, Padmapriya S T, Sriramakrishnan P, Somasundaram K., "Advancements of MRI-based brain Tumor Segmentation from
Traditional to Recent Trends- A Review", Current Medical Imaging, Bentham Science, Vol. 18, No.12, pp. 1261-1275 , 2022. DOI:
10.2174/1573405617666211215111937
[2] Shivalila.H, Neelima.N, Deepa.k, Tolga Ozer, “An evolutionary model for sleep quality analytics using fuzzy system”, Proceedings of
the institution of mechanical engineers Part H: Journal of Engineering in Medicine, (SCI indexed) IF-1.8, 2023.
[3] Bulusu, S., Sai Surya Siva Prasad, R., Telluri, P., Neelima, N, ”Methods for epileptic seizure prediction using EEG signals:
A survey”, Lecture Notes in Networks and Systems, vol:130, pp: 101– 115, 2020.
[4] Madan.k, Anusha.K, Neelima.N, “Research on different classifiers for early detection of lung nodules”, - IJRTE ,vol:8, Issue-2S3,
pp:1037- 1040, 2019Liu, Z., Lv, Q., Yang, Z., Li, Y., Lee, C. H., & Shen, L. (2023). Recent progress in transformer-based medical image
analysis. *Computers in Biology and Medicine, 164*, 107268. https://doi.org/10.1016/j.compbiomed.2023.107268.
[5] ZainEldin H, Gamel SA, El-Kenawy EM, Alharbi AH, Khafaga DS, Ibrahim A, Talaat FM. Brain Tumor Detection and Classification
Using Deep Learning and Sine-Cosine Fitness Grey Wolf Optimization. Bioengineering (Basel). 2022 Dec 22;10(1):18. doi:
10.3390/bioengineering10010018. PMID: 36671591; PMCID: PMC9854739.
[6] Amin, J., Sharif, M., Haldorai, A. et al. Brain tumor detection and classification using machine learning: a comprehensive survey.
Complex Intell. Syst. 8, 3161–3183 (2022). https://doi.org/10.1007/s40747-021-00563-y.
[7] Anagun, Y. Smart brain tumor diagnosis system utilizing deep convolutional neural networks. Multimed Tools Appl (2023).
https://doi.org/10.1007/s11042-023-15422-w.
[8] Badža MM, Barjaktarović MČ. Classification of Brain Tumor from MRI Images Using a Convolutional Neural Network. Applied
Sciences. 2020; 10(6):1999. https://doi.org/10.3390/app10061999
[9] Srinivasan S, Bai PSM, Mathivanan SK, Muthukumaran V, Babu JC, Vilcekova L. Grade Classification of Tumor from Brain Magnetic
Resonance Images Using a Deep Learning Technique. Diagnostics (Basel). 2023 Mar 17;13(6):1153. doi: 10.3390/diagnostics13061153.
PMID: 36980463; PMCID: PMC10046932.
[10] Khan, Muhammad & Ashraf, Imran & Alhaisoni, Majed & Damaševičius, Robertas & Scherer, Rafal & Rehman, Amjad & Ahmad, Syed
& Bukhari, Syed Ahmad Chan. (2020). Multimodal Brain Tumor Classification Using Deep Learning and Robust Feature Selection: A
Machine Learning Application for Radiologists. Diagnostics. 10. 565. 10.3390/diagnostics10080565.
[11] Masood M, Nazir T, Nawaz M, Mehmood A, Rashid J, Kwon H-Y, Mahmood T, Hussain A. A Novel Deep Learning Method for
Recognition and Classification of Brain Tumor from MRI Images. Diagnostics. 2021; 11(5):744.
https://doi.org/10.3390/diagnostics11050744
[12] Hassan Ali Khan, Wu Jue, Muhammad Mushtaq, Muhammad Umer Mushtaq. Brain tumor classification in MRI image using
convolutional neural network [J]. Mathematical Biosciences and Engineering, 2020, 17(5): 6203-6216. doi: 10.3934/mbe.
[13] Biratu, Erena Siyoum, 2020328.Friedhelm Schwenker, Yehualashet Megersa Ayano, and Taye Girma Debelee. 2021. "A Survey of Brain
Tumor Segmentation and Classification Algorithms" Journal of Imaging 7, no. 9: 179. https://doi.org/10.3390/jimaging7090179.
[14] Mahmud, Md Ishtyaq, Muntasir Mamun, and Ahmed Abdelgawad. 2023. "A Deep Analysis of Brain Tumor Detection from MR Images
Using Deep Learning Networks" Algorithms 16, no. 4: 176. https://doi.org/10.3390/a16040176.
[15] Tiwari P, Pant B, Elarabawy MM, Abd-Elnaby M, Mohd N, Dhiman G, Sharma S. CNN Based Multiclass Brain Tumor Detection Using
Medical Imaging. Comput Intell Neurosci. 2022 Jun 21; 2022:1830010. doi: 10.1155/2022/1830010. PMID: 35774437; PMCID:
PMC9239800.
[16] Rasool M, Ismail NA, Boulila W, Ammar A, Samma H, Yafooz WMS, Emara AM. A Hybrid Deep Learning Model for Brain Tumour
Classification. Entropy (Basel). 2022 Jun 8; 24(6):799. doi: 10.3390/e24060799. PMID: 35741521; PMCID: PMC9222774.