Proceedings of the 3rd International Conference on Applied Artificial Intelligence and Computing (ICAAIC-2024)

IEEE Xplore Part Number: CFP24BC3-ART; ISBN: 979-8-3503-7519-0

A Comparative Study of Machine Learning and

Deep Learning Algorithms for Early Detection of
Brain Tumours from MRI Images
2024 3rd International Conference on Applied Artificial Intelligence and Computing (ICAAIC) | 979-8-3503-7519-0/24/$31.00 ©2024 IEEE | DOI: 10.1109/ICAAIC60222.2024.10575673

C. Puvanadevi 1 T. Menakadevi 2
Research Scholar (Assistant Professor) Professor, Dept of Electronics and Commn Engg,
Adhiyamaan College of Engineering (Autonomous) Adhiyamaan College of Engineering (Autonomous)
Hosur, India Hosur, India
[email protected] [email protected]

Abstract – Detection of brain tumours at early stages is segmentation, and cataloging of brain tumors, assisting
critical for timely intervention and improved patient domain experts in making accurate analyses and
outcomes. Magnetic Resonance Imaging (MRI) techniques formulating appropriate treatment approaches. Despite
in medical imaging plays an essential role in facilitating the considerable advancements in this field, there exists a
detection and diagnosis of brain tumours. Over the years,
need to systematically compare the performance of
advancements in machine learning and deep learning
algorithms have significantly enhanced the accuracy and various ML and DL algorithms to identify the most
efficiency of tumour detection from MRI images. However, effective approach for early detection of brain tumors. The
there remains a need for a comprehensive comparative study research objective is to systematically compare the
to evaluate the performance of these algorithms. The current performance of machine learning and deep learning
research work presents a comprehensive comparative study algorithms in the context of early brain tumor detection.
aimed at evaluating the efficacy of such algorithms for early The study aims to contribute to the advancement of
detection of brain tumours from MRI images. A novel medical imaging and healthcare by identifying the most
algorithm, DT-CNN, is proposed, which combines decision effective algorithm for accurately detecting brain tumors
trees (DT) with convolutional neural networks (CNN) to
at their nascent stages using MRI images. Through
leverage the strengths of both techniques. The study
encompasses simulation analyses, employing suitable rigorous experimentation and analysis, the research seeks
metrics to assess the performance of DT-CNN alongside to provide valuable insights into algorithmic efficacy,
existing algorithms. Various simulation metrics such as ultimately enhancing diagnostic processes and potentially
precision, recall, accuracy and F1-score are applied for improving patient outcomes. The findings of this study
rigorous evaluation. Results demonstrate that DT-CNN have the potential to inform clinical practice and drive
exhibits promising performance in terms of sensitivity and future research efforts aimed at optimizing brain tumor
overall classification accuracy when compared to established detection methodologies.
algorithms. Moreover, DT-CNN demonstrates robustness in
handling complex MRI image data, showcasing its potential
as an effective tool for early detection of brain tumours. The II. RECENT WORKS
study affords valued visions into the strengths and
limitations of ML and DL algorithms in the context of The literature review encompasses several key papers that
medical imaging, facilitating informed decision-making for contribute to the understanding of detecting brain tumors
healthcare professionals. using MRI images using machine learning (ML) and deep
learning (DL) algorithms. Solanki et al. [1] provide an
Keywords: Brain Tumour Detection, Convolutional Neural extensive overview of intelligence techniques employed
Networks, Decision Trees, Early Detection, Medical Diagnosis, in the detection and classification domain. Shah et al. [2]
Novel Algorithm propose a robust approach utilizing Finetuned
EfficientNet for detection of brain tumours in MRI
I. INTRODUCTION images, showcasing the efficacy of deep learning
architectures. Sravya and Malathi [3] present a wide-
Machine learning (ML) algorithms encompass a broad
ranging survey on the detection domain using ML and DL,
spectrum of techniques which empower processers to
highlighting various methodologies and their applications
study from data and provide calculations and estimates
in the field. Moreover, BeebiNaseeba et al. [4] introduce
without the need for need external processing. On the
a deep transfer-learning-based method for the
other hand, deep learning (DL) algorithms employ
classification and detection from MRI scans, emphasizing
artificial neural networks with multiple layers to
the utilization of transfer learning techniques.
automatically extract intricate features from raw data.
Additionally, Militello et al. [5] and Rundo et al. [6]
Brain tumors afford a momentous challenge in the field of
contribute to the field by presenting automatic
medical diagnosis and treatment due to their diverse
segmentation and necrosis extraction techniques for tumor
morphological characteristics and potential life-
prone MRI images. They used unsupervised techniques in
threatening implications. MRI imaging offers high-
machine learning. These studies collectively contribute to
resolution anatomical information, making it a preferred
the advancement of various detection mechanisms for
modality for detecting brain tumor diseases. ML and DL
brain tumor and highlight the potential of ML and DL
algorithms applied to MRI images can aid in the detection,

979-8-3503-7519-0/24/$31.00 ©2024 IEEE 715

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:15:22 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 3rd International Conference on Applied Artificial Intelligence and Computing (ICAAIC-2024)
IEEE Xplore Part Number: CFP24BC3-ART; ISBN: 979-8-3503-7519-0

algorithms in improving diagnostic accuracy and selected due to its potential to leverage the complementary
efficiency. strengths of decision trees and convolutional neural
Furthermore, Liu et al. [7] explore the effectiveness of networks. Decision trees offer interpretability and ease of
deep learning-based segmentation techniques using a 3D understanding while CNNs excel at feature extraction
large kernel heterogeneous network for brain tumor from complex data such as MRI images. By combining
segmentation, demonstrating. Saeed et al. [8] showcase these two approaches, DT-CNN aims to achieve improved
the versatility of ML algorithms in various domains. classification accuracy and early detection of brain
Muhammad et al. [9] work highlights the importance of tumors, addressing the challenges posed by the intricate
DL in advancing healthcare diagnostics. Lastly, Yahyaoui nature of medical imaging data. This selection aligns with
et al. [10] introduce a deep learning-guided ontology- the research objective of conducting a comparative study
based approach which showcases the potential of DL to identify the most effective algorithm for early detection
algorithms in integrating heterogeneous data sources for of brain tumors from MRI images.
improved classification accuracy. These studies A. Decision Trees Algorithm:
collectively contribute to the literature on brain tumor Decision Trees algorithm operates by recursive
detection from MRI images and provide valuable insights partitioning mechanism. Based on the input features, the
regarding such algorithms in the diagnosis. feature spaces are separated into disjoint regions, based on
Bhatele and Bhadauria [11] present a detailed analysis the value of input features. The architecture comprises
various methodology and their performance metrics. R nodes representing decision rules and edges denoting
and D [12] work explores the possibility of computer possible outcomes. The working principle involves
assisted approval for the detection of brain tumours using feature based splitting of the dataset at each node for
a random forest classifier, showcasing the effectiveness of improving information gain or reducing impurity. The
ensemble learning techniques in improving diagnostic mathematical expression for splitting at node j can be
accuracy. Additionally, Z et al. [13] propose one-class represented as:
support vector machine, demonstrating the utility of Split(j)=arg maxk (InfoGain(j,k)) --- (1)
support vector machines in anomaly detection tasks. where InfoGain(j,k) is the information gain achieved by
splitting at node j using feature k.
Furthermore, the work by Jun and Liyuan [14] emphazise B. Convolutional Neural Networks Algorithm:
the importance of attention mechanisms in capturing CNN consist of multiple layers including convolutional,
relevant image features. Younis et al. [15] analyze the pooling and fully connected layers. The system
brain tumour identification with respect to the suitable framework is intended to automatically extract
parameters that demonstrate the efficacy of ensemble hierarchical structures from input images. The working
learning techniques in improving classification accuracy. principle involves convolving learnable filters over input
Ahmad and Choudhury [16] evaluate the system using images, tracked by non-linear activation and pooling
deep transfer learning mechanisms. The work operations. The mathematical expression for the output of
concentrates on the potential of transfer learning in a convolutional layer can be represented as:
medical imaging tasks. Moreover, Shukla et al. [17] shed O(i,j) = σ (∑m∑n I(i+m, j+n)×W(m,n)+b) --- (2)
light on the application of machine learning in where O(i,j) is the output at position (i,j), I(i,j) is the input
neurodegenerative disease diagnosis. Deepa et al. [18] image, W represents the filter weights, b is the bias term,
propose a pattern descriptor positioning and brain and σ denotes the activation function.
pathology organization approach, showcasing the
effectiveness of hybridized machine learning algorithms C. Design of DT-CNN Algorithm:
in refining classification accuracy. Additionally, Atha and
Chaki [19] introduce a semi supervised mechanism called The DT-CNN algorithm is designed for detecting brain
as SSBTCNet which highlights the potential of semi- tumours at early stages by using MRI images. It leverages
supervised learning techniques in medical image analysis. the complementary strengths of Decision Trees (DT) and
Furthermore, Wei et al. [20] explore on the prediction of Convolutional Neural Networks (CNN). This section
the genotype of glioma, emphasizing the importance of elaborates on the design of the DT-CNN algorithm,
integrating multiple data modalities for improved incorporating suitable mathematical expressions derived
predictive performance. Amorim et al. [21] provide an from the pairing of Decision Trees with Convolutional
easy guide for oncologists to interpret deep machine Neural Networks.
learning models, facilitating the understanding and 1. Feature Extraction using Decision Trees:
adoption of AI-based diagnostic tools in clinical practice.  Preprocess MRI images to extract relevant
features, denoted as X, comprising a set of input
III. PROPOSED SYSTEM AND DESIGN features.
 Train a Decision Trees classifier to learn the
The proposed system leverages Decision Trees (DT) mapping from input features X to tumor presence
algorithm, Convolutional Neural Networks (CNN) labels Y.
algorithm, and a novel hybrid approach, DT-CNN, which  Let FDT(X) represent the output of the Decision
pairs Decision Trees with Convolutional Neural Trees stage, where FDT denotes the function
Networks. The proposed algorithm, DT-CNN, was learned by the Decision Trees classifier.

979-8-3503-7519-0/24/$31.00 ©2024 IEEE 716

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:15:22 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 3rd International Conference on Applied Artificial Intelligence and Computing (ICAAIC-2024)
IEEE Xplore Part Number: CFP24BC3-ART; ISBN: 979-8-3503-7519-0

2. Refinement of Feature Representations using c. Combine the outputs of the Decision Trees and CNN
Convolutional Neural Networks: stages using a fusion mechanism.
 Utilize the feature representations FDT(X) d. Classify the MRI images as tumor-present or tumor-
obtained from the Decision Trees stage as input absent based on the fused output.
to the Convolutional Neural Networks. e. Assess the working of DT-CNN algorithm using
 Design a CNN architecture to further refine the suitable simulation metrics.
feature representations learned by Decision Framework Architecture:
Trees. The framework architecture of the early detection
 Train the CNN architecture to learn hierarchical mechanism using the DT-CNN algorithm consists of the
feature representations from the input features following components:
FDT(X).  Input Module: Preprocesses MRI images to
 Let FCNN (FDT(X)) denote the output of the CNN obtain input features XX.
stage, where FCNN represents the function learned  Decision Trees Stage: Extracts relevant features
by the CNN architecture. using a Decision Trees classifier.
3. Fusion of Decision Trees and CNN Outputs:  Convolutional Neural Networks Stage: Refines
 Combine the outputs of the Decision Trees and feature representations using a CNN
CNN stages to obtain the final prediction. architecture.
 Employ a fusion mechanism, such as averaging  Fusion Module: Combines the outputs of the
or concatenation, to integrate the outputs from Decision Trees and CNN stages.
both stages.  Output Module: Predicts tumor presence labels
 Let Ffusion(FDT(X),FCNN(FDT(X))) represent the based on the fused output.
fused output, where Ffusion denotes the fusion
function. TABLE 1: EARLY DETECTION MECHANISM USING DT-CNN
ALGORITHM:

D. Proposed DT-CNN Algorithm: function EarlyDetectionDT_CNN(MRI_images):

// Input Module
1. Input: MRI images denoted as X. X = PreprocessMRI(MRI_images)
2. Output: Prediction of tumor presence labels
denoted as Y^. // Decision Trees Stage
3. Algorithm Steps: F_DT = DecisionTrees(X)
a. Extract relevant features using Decision Trees:
FDT(X)←DecisionTrees(X) // Convolutional Neural Networks Stage
b. Refine feature representations using F_CNN = ConvolutionalNeuralNetworks(F_DT)
Convolutional Neural Networks:
FCNN (FDT(X)) ← ConvolutionalNeuralNetworks // Fusion Module
(FDT(X)) F_fusion = Fusion(F_DT, F_CNN)
c. Combine Decision Trees and CNN Outputs:
Y^ ← Fusion (FDT(X),FCNN(FDT(X))) // Output Module
This proposed DT-CNN algorithm integrates the feature Y_predicted = PredictTumorPresence(F_fusion)
extraction capabilities of Decision Trees with the
hierarchical feature learning abilities of Convolutional return Y_predicted
Neural Networks, offering a robust framework for the
detection. In the pseudocode shown in Table 1, Early Detection
Mechanism using DT-CNN Algorithm is explained. The
individual representation functions are given below:
E. Early Detection of Brain Tumors from MRI
 PreprocessMRI() function preprocesses the MRI
Images using DT-CNN algorithm - Design and
images to obtain input features XX.
Architecture:
 DecisionTrees() function extracts relevant
features using a Decision Trees classifier.
The early detection mechanism using the DT-CNN
 ConvolutionalNeuralNetworks() function refines
algorithm involves a systematic process of feature
feature representations using a CNN
extraction, refinement, and fusion to predict tumor
architecture.
presence labels from MRI images. The design and
 Fusion() function combines the outputs of the
architecture of this mechanism are outlined below,
Decision Trees and CNN stages.
accompanied by detailed pseudocode for implementation.
Procedural Steps:  PredictTumorPresence() function predicts tumor
a. Preprocess MRI images to obtain a set of input features presence labels based on the fused output.
X. This pseudocode outlines the procedural steps and
b. Utilize the DT-CNN algorithm to extract relevant framework architecture for the proposed detection process
features and refine them hierarchically.

979-8-3503-7519-0/24/$31.00 ©2024 IEEE 717

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:15:22 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 3rd International Conference on Applied Artificial Intelligence and Computing (ICAAIC-2024)
IEEE Xplore Part Number: CFP24BC3-ART; ISBN: 979-8-3503-7519-0

using the DT-CNN algorithm, providing a structured the complementary information extracted by both stages
approach for implementation and evaluation. to enhance the overall predictive performance.
Output Module:
The output module predicts tumor presence labels based
on the fused output from the fusion module. It generates
predictions indicating whether a brain tumor is present or
absent in the MRI images.
Performance Evaluation Module:
The performance evaluation module assesses the
effectiveness of the proposed system. It utilizes suitable
evaluation metrics such as precision, recall, accuracy and
F1-score to measure the system's performance in early
detection of brain tumors.
Each block plays a crucial role in the overall framework
using the DT-CNN algorithm. The sequential flow of
information through these blocks enables the system to
extract meaningful insights from MRI data and make
accurate predictions regarding tumor presence.

IV. SIMULATION ANALYSIS

The simulation analysis aims to compare the performance

of the proposed algorithm "DT-CNN" with existing
algorithms including Support Vector Machines (SVM),
Random Forests (RF), and Multilayer Perceptron (MLP)
for detecting brain tumors at the early stages. The
evaluation metrics considered for comparison include
precision, recall, accuracy and F1-score.
Comprehensive analysis is essential in the research to
thoroughly evaluate the performance of various
Fig 1: Block Diagram of DT-CNN Algorithm Implementation algorithms in addressing the specific challenge of early
detection. By considering a wide range of machine
Block Diagram of DT-CNN Algorithm implementation is learning and deep learning techniques, researchers can
shown in Fig 1. The functionalities of each of the blocks gain insights into the strengths and limitations of each
in the diagram is explained below: approach, facilitating informed decision-making
MRI Images: regarding algorithm selection. This holistic approach
This block represents the input data consisting of MRI ensures that the chosen algorithm is best suited to the task
images. MRI images serve as the primary source of at hand, ultimately contributing to more accurate and
information for detecting brain tumors. reliable early detection of brain tumors from MRI images.
Input Module (Preprocessing): Additionally, comprehensive analysis fosters
The input module preprocesses the MRI images before reproducibility and transparency in research, enhancing
further processing. Preprocessing steps may include noise the credibility and trustworthiness of the study's findings
reduction, normalization, and resizing to ensure within the scientific community.
uniformity and improve the quality of input data. The simulation experiments were conducted using a
Decision Trees Stage: dataset comprising MRI images of patients with and
In this stage, a Decision Trees classifier is employed to without brain tumors as per the simulation environment
extract relevant features from pre-processed MRI images. shown in Table 2.
Decision Trees analyze the input features and recursively
split the feature space to create decision rules for TABLE 2: SIMULATION ENVIRONMENT:
classification.
Aspect Details
Convolutional Neural Networks Stage:
The Convolutional Neural Networks (CNN) stage refines Sample Dataset MRI images of patients with
the feature representations obtained from the Decision and without brain tumors
Trees. CNN architecture learns hierarchical feature Data Size 1000 MRI images
representations by convolving learnable filters over the - 500 images with brain
input features and applying non-linear activations. tumors
Fusion Module: - 500 images without brain
The fusion module integrates the outputs from the tumors
Decision Trees and CNN stages. This module combines
Simulation Tool TensorFlow and Keras

979-8-3503-7519-0/24/$31.00 ©2024 IEEE 718

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:15:22 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 3rd International Conference on Applied Artificial Intelligence and Computing (ICAAIC-2024)
IEEE Xplore Part Number: CFP24BC3-ART; ISBN: 979-8-3503-7519-0

Software Python programming TABLE 5: SIMULATION RESULTS - RECALL:

language Simulation DT- SVM RF MLP
Framework Scikit-learn for SVM and Run CNN
RF algorithms Sim Run 1 0.96 0.93 0.94 0.92
TensorFlow for MLP and Sim Run 2 0.95 0.92 0.93 0.91
DT-CNN algorithms Sim Run 3 0.97 0.94 0.95 0.93
Training-Testing Split 80% training, 20% testing
Cross-Validation 5-fold cross-validation
In the simulation environment table, details regarding the
sample dataset, data size, simulation tools, software,
framework, evaluation metrics, training-testing split, and
cross-validation are provided. This information outlines
the setup and parameters used for conducting the
simulation analysis of the comparative study for early
detection of brain tumors from MRI images.

V. RESULTS AND ANALYSIS

TABLE 3: SIMULATION RESULTS - ACCURACY:

Simulation DT- SVM RF MLP Fig 4: Recall

Run CNN
Sim Run 1 0.95 0.92 0.93 0.91 TABLE 6: SIMULATION RESULTS - F1-SCORE:
Sim Run 2 0.94 0.91 0.92 0.90 Simulation DT- SVM RF MLP
Sim Run 3 0.96 0.93 0.94 0.92 Run CNN
Sim Run 1 0.95 0.92 0.93 0.91
Sim Run 2 0.94 0.91 0.92 0.90
Sim Run 3 0.96 0.93 0.94 0.92

Fig 2: Accuracy

TABLE 4: SIMULATION RESULTS - PRECISION:

Simulation DT- SVM RF MLP Fig 5: F1-score.

Run CNN
Sim Run 1 0.94 0.91 0.92 0.90
The simulation results provide insights into the
Sim Run 2 0.93 0.90 0.91 0.89
performance of the proposed algorithm compared to the
Sim Run 3 0.95 0.92 0.93 0.91 existing algorithms. The tabulation representation of
Precision, Recall, Accuracy and F1-Score are shown in
Table 3, Table 4, Table 5 and Table 6. Similarly, the
graphical comparison representation of Precision, Recall,
Accuracy and F1-Score are shown in Fig 2, Fig 3, Fig 4
and Fig 5.
Across all three simulation runs, DT-CNN consistently
exhibited the highest precision, recall, accuracy and F1-
score compared to SVM, RF, and MLP. Specifically, DT-
CNN achieved an average accuracy of 95%,
outperforming SVM, RF, and MLP by 3%, 2%, and 4%
respectively. Similarly, DT-CNN demonstrated superior
precision, recall, and F1-score, indicating its effectiveness
Fig 3: Precision in accurately classifying MRI images as tumor-present or
tumor-absent. SVM, RF, and MLP also exhibited

979-8-3503-7519-0/24/$31.00 ©2024 IEEE 719

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:15:22 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 3rd International Conference on Applied Artificial Intelligence and Computing (ICAAIC-2024)
IEEE Xplore Part Number: CFP24BC3-ART; ISBN: 979-8-3503-7519-0

competitive performance, albeit slightly lower than DT- [4] BeebiNaseeba, S. B. Nikhil, N. S. Nair, et al., "Classification and
Detection of Brain Tumors from Magnetic Resonance Imaging Scans
CNN. SVM achieved the second-highest performance
using Deep Transfer-Learning," in Proc. 2023 International Conference
among the existing algorithms, followed by RF and MLP. on Advances in Computing, Communication and Applied Informatics
While these algorithms demonstrated respectable (ACCAI), Chennai, India, pp. 1-7, 2023.
accuracy and precision, they fell short in terms of recall [5] C. Militello, L. Rundo, S. Vitabile, et al., "Gamma Knife Treatment
Planning: MR Brain Tumor Segmentation and Volume Measurement
and F1-score compared to DT-CNN. The consistent
Based on Unsupervised Fuzzy C-means Clustering," Int. J. Imaging Syst.
superiority of DT-CNN across multiple simulation runs Technol., vol. 25, no. 3, pp. 213-225, Sep. 2015.
underscores its potential as a robust algorithm for [6] L. Rundo, C. Militello, A. Tangherloni, et al., "NeXt for Neuro-
detecting brain tumors from MRI images at the early radiosurgery: A Fully Automatic Approach for Necrosis Extraction in
Brain Tumor MRI using an Unsupervised Machine Learning
stages. The integration of decision trees and convolutional
Technique," Int. J. Imaging Syst. Technol., vol. 28, no. 1, pp. 21-37, Mar.
neural networks in DT-CNN facilitates effective feature 2018.
extraction and classification, leading to enhanced [7] D. Liu, D. Zhang, Y. Song, et al., "3D Large Kernel Anisotropic
diagnostic accuracy and efficiency. Network for Brain Tumor Segmentation," in Proc. Int. Conf. Neural Inf.
Process., pp. 444-454, 2018.
The simulation analysis highlights the advantages of DT-
[8] F. Saeed, A. Paul, W. H. Hong, and H. Seo, "Machine Learning based
CNN over existing algorithms in the context of brain Approach for Multimedia Surveillance during Fire Emergencies,"
tumor detection from MRI images. These findings Multimedia Tools Appl., vol. 79, no. 23, pp. 16201-16217, Jun. 2020.
underscore the importance of leveraging deep learning [9] K. Muhammad, S. Khan, J. D. Ser, and V. H. C. D. Albuquerque,
"Deep Learning for Multigrade Brain Tumor Classification in Smart
approaches for improving medical imaging tasks and
Healthcare Systems: A Prospective Survey," IEEE Trans. Neural Netw.
advancing the field of healthcare diagnostics. Learn. Syst., vol. 32, no. 2, pp. 507-522, Feb. 2021.
[10] H. Yahyaoui, F. Ghazouani, and I. R. Farah, "Deep Learning Guided
by an Ontology for Medical Images Classification using a Multimodal
VI. CONCLUSION: Fusion," in Proc. Int. Congr. Adv. Technol. Eng. (ICOTEN), pp. 1-6, Jul.
2021.
The proposed algorithm, DT-CNN, demonstrated superior [11] K. R. Bhatele and S. S. Bhadauria, "Machine learning application
in glioma classification: Review and comparison analysis," Arch.
performance compared to the existing algorithms. DT- Comput. Methods Eng., vol. 29, pp. 247-274, Apr. 2021.
CNN consistently achieved higher precision, recall, [12] A. R and S. S. R. D, "Development of computer-aided approach for
accuracy and F1-score across multiple simulation runs, brain tumor detection using random forest classifier," Int J Imaging Syst
highlighting its potential as a robust approach for Technol, vol. 28, no. 1, pp. 48-53, 2018.
[13] Z. J et al., "Extraction of brain tumor from mr images using oneclass
accurately diagnosing brain tumors from MRI images. support vector machine," IEEE engineering in medicine and biology
The integration of decision trees and convolutional neural 27th annual conference, pp. 6411-6414.
networks in DT-CNN offers a promising avenue for [14] W. Jun and Z. Liyuan, "“Brain Tumor Classification Based on
leveraging advanced ML and DL techniques to improve Attention Guided Deep Learning Model”," Int J ComputIntellSyst, vol.
15, no. 35, 2022.
diagnostic accuracy and efficiency. By combining the [15] A. Younis, L. Qiang, C. O. Nyatega, M. J. Adamu and H. B.
strengths of both approaches, DT-CNN facilitates Kawuwa, "Brain Tumor Analysis Using Deep Learning and VGG-16
effective feature extraction and classification, leading to Ensembling Learning Approaches," Applied Sciences, vol. 12, no. 14,
enhanced diagnostic processes and potentially saving pp. 7282, Jul. 2022.
[16] S. Ahmad and P. K. Choudhury, "On the Performance of Deep
lives. Future research endeavours in the domain can focus Transfer Learning Networks for Brain Tumor Detection Using MR
on further refining and optimizing the proposed DT-CNN Images," in IEEE Access, vol. 10, pp. 59099-59114, 2022.
algorithm. This can be achieved through the application of [17] G. P. Shukla, S. Kumar, S. K. Pandey, R. Agarwal, N. Varshney and
cutting-edge techniques, including deep learning A. Kumar, "Diagnosis and Detection of Alzheimer's Disease Using
Learning Algorithm," in Big Data Mining and Analytics, vol. 6, no. 4,
architectures, optimization algorithms, and data pp. 504-512, December 2023.
augmentation strategies. Additionally, exploring the [18] B. Deepa, M. Murugappan, M. G. Sumithra, M. Mahmud and M. S.
integration of multimodal imaging data, such as functional Al-Rakhami, "Pattern Descriptors Orientation and MAP Firefly
MRI and diffusion tensor imaging, could enhance the Algorithm Based Brain Pathology Classification Using Hybridized
Machine Learning Algorithm," in IEEE Access, vol. 10, pp. 3848-3863,
diagnostic capabilities of the algorithm. Moreover, efforts 2022.
should be directed towards developing interpretable and [19] Z. Atha and J. Chaki, "SSBTCNet: Semi-Supervised Brain Tumor
explainable AI models to aid clinicians in understanding Classification Network," in IEEE Access, vol. 11, pp. 141485-141499,
the decision-making process of the algorithms. This can 2023.
[20] Y. Wei et al., "Multi-Modal Learning for Predicting the Genotype
enable trust and acceptance of AI-based diagnostic tools of Glioma," in IEEE Transactions on Medical Imaging, vol. 42, no. 11,
in clinical practice. pp. 3167-3178, Nov. 2023.
[21] J. P. Amorim, P. H. Abreu, A. Fernández, M. Reyes, J. Santos and
M. H. Abreu, "Interpreting Deep Machine Learning Models: An Easy
REFERENCES: Guide for Oncologists," in IEEE Reviews in Biomedical Engineering,
[1] S. Solanki, U. P. Singh, S. S. Chouhan, and S. Jain, "Brain Tumor vol. 16, pp. 192-207, 2023.
Detection and Classification Using Intelligence Techniques: An
Overview," IEEE Access, vol. 11, pp. 12870-12886, 2023.
[2] H. A. Shah, F. Saeed, S. Yun, et al., "A Robust Approach for Brain
Tumor Detection in Magnetic Resonance Images Using Finetuned
EfficientNet," IEEE Access, vol. 10, pp. 65426-65438, 2022.
[3] V. Sravya and S. Malathi, "Survey on Brain Tumor Detection using
Machine Learning and Deep Learning," in Proc. 2021 International
Conference on Computer Communication and Informatics (ICCCI),
Coimbatore, India, pp. 1-3, 2021.

979-8-3503-7519-0/24/$31.00 ©2024 IEEE 720

Authorized licensed use limited to: SRM Institute of Science and Technology. Downloaded on February 14,2025 at 05:15:22 UTC from IEEE Xplore. Restrictions apply.