Abstract:Unsupervised skin lesion segmentation offers several benefits, including conserving expert human resources, reducing discrepancies due to subjective human labeling, and adapting to novel environments. However, segmenting dermoscopic images without manual labeling guidance presents significant challenges due to dermoscopic image artifacts such as hair noise, blister noise, and subtle edge differences. To address these challenges, we introduce an innovative Uncertainty Self-Learning Network (USL-Net) designed for skin lesion segmentation. The USL-Net can effectively segment a range of lesions, eliminating the need for manual labeling guidance. Initially, features are extracted using contrastive learning, followed by the generation of Class Activation Maps (CAMs) as saliency maps using these features. The different CAM locations correspond to the importance of the lesion region based on their saliency. High-saliency regions in the map serve as pseudo-labels for lesion regions while low-saliency regions represent the background. However, intermediate regions can be hard to classify, often due to their proximity to lesion edges or interference from hair or blisters. Rather than risk potential pseudo-labeling errors or learning confusion by forcefully classifying these regions, we consider them as uncertainty regions, exempting them from pseudo-labeling and allowing the network to self-learn. Further, we employ connectivity detection and centrality detection to refine foreground pseudo-labels and reduce noise-induced errors. The application of cycle refining enhances performance further. Our method underwent thorough experimental validation on the ISIC-2017, ISIC-2018, and PH2 datasets, demonstrating that its performance is on par with weakly supervised and supervised methods, and exceeds that of other existing unsupervised methods.

Skin lesion segmentation using U-Net and good training strategies

Attention-Guided Network with Densely Connected Convolution for Skin Lesion Segmentation

SkinNet: A Deep Learning Framework for Skin Lesion Segmentation

SEACU-Net: Attentive ConvLSTM U-Net with squeeze-and-excitation layer for skin lesion segmentation

Lesion Detection on Skin Images Using Improved U-Net

Semantic Segmentation of Skin Lesions using a Small Data Set

Enhancing skin lesion segmentation with U-Net++: Design, analysis, and performance evaluation

U-Net-based Models for Skin Lesion Segmentation: More Attention and Augmentation

Lesion Net -- Skin Lesion Segmentation Using Coordinate Convolution and Deep Residual Units

Skin lesion region segmentation model based on improved U2Net

LesionSeg: Semantic segmentation of skin lesions using Deep Convolutional Neural Network

Skin lesion segmentation with attention-based SC-Conv U-Net and feature map distortion

Towards improved U-Net for efficient skin lesion segmentation

Deep Hybrid Convolutional Neural Network for Segmentation of Melanoma Skin Lesion

Skin lesion segmentation based on preprocessing, thresholding and neural networks

EIU-Net: Enhanced feature extraction and improved skip connections in U-Net for skin lesion segmentation

The Effects of Image Pre- and Post-Processing, Wavelet Decomposition, and Local Binary Patterns on U-Nets for Skin Lesion Segmentation

DSNet: Automatic Dermoscopic Skin Lesion Segmentation

USL-Net: Uncertainty Self-Learning Network for Unsupervised Skin Lesion Segmentation

GSCEU-Net: an End-to-End Lightweight Skin Lesion Segmentation Model with Feature Fusion Based on U-Net Enhancements

TESL-Net: A Transformer-Enhanced CNN for Accurate Skin Lesion Segmentation