UIM-Net: A New Frontier in Automated Skin Disease Diagnosis

Abstract

This paper introduces UIM-Net, a novel network model designed for the segmentation of skin lesions in medical images. Tested on the ISIC 2018 dataset, UIM-Net demonstrates superior performance compared to established models such as UNet, AttU-Net, UNet++, DeepLabV3, and UNeXt. Quantitative analysis reveals that UIM-Net achieves the highest IOU scores and surpasses other models in F1 scores, as illustrated in Figure 3.9. The model's parameter count and computational complexity are significantly lower, making it a lightweight yet efficient choice. Through TOPSIS assessment, UIM-Net earns the highest score, indicating its advanced segmentation capabilities on the ISIC 2018 skin disease dataset. This study underscores the potential of UIM-Net in enhancing dermatological image analysis.