Brain Tumor Image Classification Using Deep Learning

Abstract

Brain tumors represent a severe medical condition that can lead to significant neurological damage or death if not diagnosed early. Traditional diagnostic methods, though effective, often involve time-consuming processes and require expert interpretation. In this project, we propose a deep learning-based approach to automate brain tumor detection using magnetic resonance imaging (MRI) scans. The system leverages the VGG19 convolutional neural network architecture, pre-trained on ImageNet and further fine-tuned for binary classification of MRI images into “Tumor” and “No Tumor” categories. The model architecture is enhanced with additional dense layers to improve its ability to learn complex patterns specific to medical imaging.The classification model is deployed within a Flask-powered web application, enabling users to upload brain scan images and receive classification results in real-time. The image is preprocessed, resized to a fixed dimension, and passed through the trained model for prediction. The application aims to provide an accessible tool for preliminary tumor screening, particularly in scenarios where immediate medical expertise may not be available. This integration of deep learning and web development demonstrates the potential for AI-driven tools to supplement healthcare workflows, reduce diagnostic delays, and support clinical decision-making.It serves as a foundational step toward building scalable, intelligent medical diagnostic tools that prioritize both performance and accessibility.

Introduction

1. Introduction

• Brain tumors are life-threatening and require early, accurate diagnosis.

• Manual MRI analysis is time-consuming, expert-dependent, and often delayed in resource-poor regions.

• AI and Deep Learning, especially Convolutional Neural Networks (CNNs), offer a faster and more accurate alternative.

2. Objective

• Build a web-based system to classify brain MRI scans into “Tumor” or “No Tumor”.

• Use VGG19 CNN architecture with transfer learning for high accuracy and efficiency.

• Integrate with a Flask web application for real-time usage in clinical or low-resource settings.

3. Literature Highlights

• CNNs outperform traditional ML by learning directly from images (Mohsen et al.).

• VGG19 + fine-tuning offers strong performance on medical data (Hossain et al.).

• Previous work focused on segmentation; this system focuses on binary classification.

4. Limitations of Existing Systems

• Traditional ML required manual feature extraction (e.g., texture, brightness).

• Most deep learning systems are not user-friendly or web-integrated.

• High-end models need powerful hardware, not ideal for remote or rural clinics.

5. Proposed Solution

• Deep Learning Model: VGG19 (pre-trained on ImageNet) + custom dense layers.

• Web App: Built using Flask, allows real-time image upload and classification.

• Lightweight and scalable, usable even on basic hardware.

6. System Architecture

Workflow includes:

1. Data Collection: MRI scans from public datasets.

2. Image Preprocessing: Resize (240×240), RGB conversion, normalization.

3. Model Adaptation: Transfer learning on VGG19 with Softmax and ReLU activations.

4. Training: Binary classification using categorical cross-entropy loss.

5. Web Integration: Image upload, real-time processing, result display.

7. Model Performance

• Training Accuracy improved from 78.2% to 93.2% in 5 epochs.

• Validation Accuracy reached 91.45%.

• Training Loss decreased steadily to 0.065, indicating effective learning and low overfitting.

8. Key Features

? Accurate brain tumor detection from MRI
? Uses pre-trained VGG19 for efficiency
? Real-time classification via user-friendly Flask web app
? Lightweight – runs on standard computers
? No manual feature extraction required

Conclusion

The brain tumor classification project demonstrates how deep learning can effectively be applied in the healthcare domain to support early and accurate diagnosis. By employing a pre-trained VGG19 architecture and customizing it for binary classification, the system achieves notable accuracy with minimal training overhead. The use of transfer learning allowed for efficient feature extraction from complex MRI images, while the added dense layers enabled the model to adapt specifically to the task of tumor detection.

References

[1] Simonyan, K., & Zisserman, A. (2015). Very Deep Convolutional Networks for Large-Scale Image Recognition. arXiv preprint arXiv:1409.1556. [2] Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., et al. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115–118. [3] Chakrabarty, A., & Shuvo, S. B. (2020). Brain tumor detection using convolutional neural networks (CNNs). 2020 International Conference on Innovations in Science, Engineering and Technology (ICISET), IEEE. [4] Chollet, F. (2015). Keras: Deep Learning Library for Theano and TensorFlow. https://keras.io [5] Abadi, M., Barham, P., Chen, J., et al. (2016). TensorFlow: A System for Large-Scale Machine Learning. In 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI 16), 265–283. [6] OpenCV Team. (2019). Open Source Computer Vision Library. https://opencv.orgFlask Documentation. (2024). Flask Web Framework Documentation.

Copyright

Copyright © 2025 Rakesh Chalamala, Dr. M. Dhanalakshmi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.