Retinoblastoma Diagnosis using Transfer Learning and Explainable AI

This project explores the application of transfer learning and explainable AI techniques for the diagnosis of retinoblastoma, a rare eye cancer primarily affecting children. It implements and compares several pre-trained CNN architectures, including InceptionV3, ResNet50, VGG16, VGG19, and Xception, to find the most accurate model for early detection. The preprocessing pipeline integrates techniques like resizing, Gaussian blur, normalization, and Dual Tree Complex Wavelet Transform (DTCWT) to enhance model performance. The study also utilizes LIME (Local Interpretable Model-agnostic Explanations) to interpret model decisions and map key anatomical features in fundus images associated with retinoblastoma. InceptionV3 achieved the highest accuracy across all stages, while the use of LIME provided greater transparency and interpretability in the diagnostic process. The research has been accepted for presentation at the 1st International Workshop on Responsible AI for Healthcare and Net Zero 2024.