Page 79 - 2024F

72 UEC Int’l Mini-Conference No.53

BTDCGANet: An Advanced Deep Transfer Learning and Generative Adversarial Network-
Based Data Augmentation Approach for Brain Tumor Classification from Magnetic
Resonance Imaging (MRI) Scans
Shabnur Anonna Akhy*, Hayaru Shouno and Tahir Hussain
UEC Exchange Study Program (JUSST Program)
Department of Informatics
The University of Electro-Communication Tokyo, Japan
a2495007@gl.cc.uec.ac.jp
1.Introduction

Diagnosing brain tumors using magnetic resonance imaging (MRI) is a complex task due to variations in tumor size, shape, and spatial distribution [1] [2]. Traditional
classification models often face challenges such as imbalanced datasets, low precision, and limited interpretability, which can result in biased learning and reduced accuracy [3].
This work addresses these challenges by integrating image processing, transfer learning (TL), and Generative Adversarial Networks (GANs) to enhance MRI-based tumor
classification. EfficientNetB0 is used for feature extraction, while DCGANs generate synthetic data to mitigate class imbalance and improve model robustness. Additionally,
explainable AI techniques such as LIME enhance transparency [4] in model predictions. The primary goal of this study is to explore a comparative analysis between GANs and
diffusion models to determine the most effective approach for early and reliable brain tumor classification.
2. Background Study
TABLE 1. Number of MRI Images in Training and Testing Datasets
References Approaches Applied Model Utilized Dataset Accuracy (%) Explainability
Oksuz et al., Machine SVM and KNN Figshare BT 97.25 No
(2022) [20] Learning classifiers dataset - 2017
Ullah N. et al., Transfer Deep BTDNet Kaggle BT 98.96 Yes
(2024) [25] Learning Detection MRI-
2021
A. Nag et al., Transfer TumorGANet Kaggle BT MRI 99.53 Yes FIGURE 3. Image pre-processing
(2024) [27] Learning (GANs + VGG-16) Dataset-2021
2.1 Contribution 4.1 Working Principle of GAN
GANs vs. Diffusion Models – Conducts a comparative study to determine the most
effective method for early and reliable brain tumor detection.
Enhanced MRI-Based Brain Tumor Classification – Utilizes EfficientNetB0 for
feature extraction and DCGANs for data augmentation.
3. Dataset
Utilized 7023 MRI images, categorized into glioma, meningioma, pituitary tumors, and
no tumor cases. Dataset includes 5712 training and 1311 testing samples,
preprocessed for consistency.
TABLE 2. Number of MRI Images in Training and
Testing Datasets
Name of Tumor Training Testing Total
FIGURE 4. Working principle of Generative adversarial Network[2].
Glioma Tumor 1321 300 1621
5.Results
Meningioma 1339 306 1645
Tumor
Pituitary tumor 1457 300 1757 For predicted glioma, pituitary, meningioma, and no tumor analysis twelve transfer
learning methods.
No tumor 1595 405 2000
Conducts a comparative study to determine the most effective method for early
TOTAL 5712 1311 7023 and reliable brain tumor detection.
FIGURE 1. Sample images from each class
4. Methods
Expected Gan and Diffusion Based Method for Imbalanced Multi-Class Brain
tumor Classification

FIGURE 5. LIME based misclassification result analysis [2]
6. Future Work

Expanding the dataset with diverse MRI scans will enhance model robustness.
Real-time implementation and clinical trials are planned to validate BTDCGANet
efficiency in practical medical settings.
FIGURE 2. Work flow diagram
References
[1] AL-SHEHARI, T.A.H.E.R., KADRIE, M., AL-RAZGAN, M.U.N.A. and ALFAKIH, T., 2024. TumorGANet: A Transfer Learning and Generative Adversarial Network-Based Data Augmentation Model for Brain Tumor Classification.
[2] Nag, A., Mondal, H., Hassan, M.M., Al-Shehari, T., Kadrie, M., Al-Razgan, M., Alfakih, T., Biswas, S. and Bairagi, A.K., 2024. TumorGANet: A transfer learning and generative adversarial network-based data augmentation model for
brain tumor classification. IEEE access.
[3] Skandarani, Y., Jodoin, P.M. and Lalande, A., 2023. Gans for medical image synthesis: An empirical study. Journal of Imaging, 9(3), p.69.
[4] Tanaka, F.H.K.D.S. and Aranha, C., 2019. Data augmentation using GANs. arXiv preprint arXiv:1904.09135.