FRAMU: A Lightweight Fractal Framework with Abductive Reasoning for Skin Lesion Segmentation

Medical image segmentation, especially skin lesion segmentation, is a key technology in computer-aided diagnosis, but existing methods face an imbalance between efficiency, accuracy, and generalization capabilities. Despite significant advances in CNNs, Transformers, and SSMs, these approaches still encounter fundamental challenges in model design, particularly in parameter efficiency. In this article, we introduce Fractal Abductive Multi-Scale UNet (FRAMU), an innovative architecture combining fractal recursive structure, adaptive multi-scale processing, and abductive reasoning. Specifically, we design a fractal structure by recursively reusing modules across different scales, dramatically reducing parameters compared to traditional U-Net. We propose a dual-branch block combining convolution and multi-scale fractal extraction to capture both local details and global context. Additionally, we incorporate anatomical constraints through an abductive reflection mechanism, which generates a reflection vector to identify potentially inconsistent regions. Then we apply knowledge-driven verification to enhance continuity, compactness, and smoothness. Experiments on ISIC2017/2018 datasets show that FRAMU achieves comparable performance to SOTA methods while reducing parameters by nearly 70%, making it particularly suitable for resource-constrained medical environments. The code is available at https://github.com/Travis-go/FRAMU.