A deep learning project that classifies seven types of skin lesions using the HAM10000 dataset. Among four tested models, Swin Transformer achieved the best accuracy of 88.9%, showing AI’s potential in early skin cancer detection.

Implementation of BSC-DenseNet-121 in Pytorch from research paper "Adding Binary Search Connections to Improve DenseNet Performance".

RAD-XAI is an explainable-AI framework for chest X‑ray pneumonia detection that benchmarks CNNs and ViTs using Grad‑CAM-based saliency maps and quantitative XAI metrics (concentration, faithfulness, agreement) to audit model reasoning and surface clinically unsafe failure modes beyond standard AUC/accuracy.