Multi-method fusion for image segmentation in skin disease analysis

Objective: The morphological complexity of dermatologic diseases poses considerable challenges to clinical diagnosis. Conventional manual interpretation of skin images is time-consuming and influenced by subjective variability, which limits diagnostic accuracy. Hence, developing advanced medical image segmentation techniques through multi-method fusion is of particular importance. Methods: A comprehensive dataset of dermatologic images was utilized and rigorously preprocessed to ensure reliability and consistency. Two representative deep learning models, SegNet and U-Net, were optimized to achieve precise delineation of lesion areas. Building upon their complementary strengths, a novel fusion-based image segmentation framework was proposed, integrating both models to enhance performance through synergistic learning. The effectiveness of the ensemble strategy was validated through extensive experiments using standard evaluation metrics, including the Dice coefficient and Intersection over Union. Results: Compared with each individual model, the Ensemble Model yielded consistent improvements across all evaluation indices, with a notable reduction in loss values. These enhancements indicate markedly better learning efficiency and generalization in dermatologic image segmentation tasks. Conclusion: By integrating multiple deep learning algorithms, this fusion technique solves the misclassification and omission issues observed in single-model segmentation. It significantly improves overall segmentation accuracy and demonstrates superior performance, particularly in edge detection of complex skin lesions.