MRI-Facilitated Brain Tumor Detection Using Deep Learning and Image Preprocessing

This project focused on developing a hybrid biomedical imaging framework that integrates MATLAB-based preprocessing and Python-based deep learning models to detect brain tumors from MRI scans. The primary challenge addressed was the data scarcity problem common in medical imaging, where the limited dataset size often leads to underfitting and model instability.

Dataset and Preprocessing

The dataset—sourced from Kaggle’s Brain MRI Images for Tumor Detection—contained 253 images (155 tumor, 98 non-tumor) in mixed formats (.jpg, .jpeg, .png). To enhance diversity and combat overfitting, the dataset underwent batchwise preprocessing, expanding it to 604 images through contrast adjustment, wiener filtering, and anisotropic diffusion.

MATLAB-Based Image Processing

The MATLAB module performed tumor detection through a sequence of steps including anisotropic diffusion filtering for edge-preserving smoothing, thresholding and morphological operations for segmentation, and bounding box generation for tumor localization. It achieved approximately 92% detection accuracy on the test subset.

Deep Learning Models and Results

The following models were implemented and evaluated using the preprocessed dataset. Each model represents a stage of improvement in both accuracy and stability.