Fractal-guided self-supervised dual-stream attention network for interpretable nail-based liver disease detection

Liver disease is an illness that requires early diagnosis to ensure effective treatment, and traditional methods of diagnosis are very invasive, expensive and unavailable in resource-limited areas. In the current study, a new non-invasive model, which is named Fractal-Guided Self-Supervised Dual-Stream Attention Network (FG-SSDSAN), is proposed to detect liver diseases with nail texture input. The methodology proposed combines fractal geometry and deep learning to identify structural complexity of changes in pathology. To deal with limited labeled data, a self-supervised learning approach is used to allow the model to derive intrinsic texture representations. The architecture is a combination of convolutional neural networks and vision transformers that extract complementary local and global features, and a fractal-guided attention mechanism that refines attention in structurally irregular regions. There is also a multi-scale consistency constraint and entropy-regularized adaptive fusion, which enhances robustness and discrimination of features. The proposed model is experimentally evaluated using a curated dataset of 1000 nail images and it is observed that the model has a high level of accuracy of 97.4% and outperforms state-of-the-art methods. The results show that combining fractal analysis with deep learning is an effective approach for liver disease classification based on nail texture images, which is interpretable. The suggested framework is a convenient and cost-effective method for the early detection of liver disease and represents a promising base for future clinical validation studies.