Accurate Segmentation and Classification of Skin Lesions

Abstract

This repository contains the research code for skin lesion segmentation and classification using deep learning. The project provides an end-to-end workflow for preprocessing dermoscopic images, segmenting skin lesions, and performing automated skin lesion classification. The segmentation component uses a SwinUNet-based architecture with Neural Cellular Automata (NCA), while the classification component uses a Swin Transformer V2 Tiny model. The repository is intended to support research reproducibility and provide the source code, pretrained model weights, and example segmentation results associated with the study.