Deep Learning Model for Brain Tumor Detection

Accurate and reliable brain tumor detection is essential in medical imaging, yet challenges such as limited and imbalanced datasets hinder model performance. This project proposes a comprehensive approach to enhance detection accuracy by integrating data augmentation, synthetic data generation, and advanced machine learning techniques. Initially, we employed a CNN on a dataset derived from a recent research paper, achieving an accuracy of 90%. Leveraging transfer of learning with an already trained model, we further improved the accuracy to 94%. To address dataset imbalance, we augmented the dataset to create 1,500 images per class, increasing the total from 3,000+ to 4,500 images. We then utilized GANs to generate 500 additional synthetic images per class, resulting in a dataset of 6,000 images. By applying an optimized CNN model with techniques like batch normalization, this enhanced dataset an accuracy reaches of 98.22%.