MSCAM-Net: A Multi-Scale and Context-Aware Model with Adaptive Fusion Module for Robust Skin Lesion Segmentation

Malignant melanoma remains one of the deadliest forms of skin cancer, emphasizing the need for precise and efficient segmentation of lesions in dermoscopic images. Traditional segmentation techniques often struggle with the diverse appearance of melanoma lesions, while U-Net encounters limitations in capturing intricate lesion features due to their fixed receptive fields. We propose a Multi-Scale Context-Aware Model (MSCAM-Net), a novel model integrating an Adaptive Fusion Module (AFM), Enhanced Feature Extraction Module (EFEM), and Context-Aware Feature Aggregation (CAFA) to improve segmentation performance on complex skin lesions in ISIC datasets. MSCAM-Net effectively captures multi-scale and context-aware features, dynamically adjusting feature fusion to maintain sensitivity to both fine-grained details and global context. Experimental results demonstrate that MSCAM-Net achieves superior segmentation accuracy over recent methods, code is available at https://github.com/Zumnan/MSCAM-Net.