GloW-VSNet: A scribble-based weakly supervised framework for global-view vitiligo lesion segmentation

Area of Science:

• Dermatology
• Medical Image Analysis
• Computer Vision

Background:

• Accurate vitiligo lesion segmentation is crucial for disease management.
• Challenges include indistinct borders, complex backgrounds, and artifacts in clinical images.
• Fully supervised methods demand extensive, costly data annotation.

Purpose of the Study:

• To introduce GloW-VSNet, a novel scribble-guided weakly supervised segmentation method for global-view vitiligo detection.
• To overcome limitations of fully supervised approaches in vitiligo lesion segmentation.
• To enhance objective quantification of vitiligo for clinical applications.

Main Methods:

• Developed GloW-VSNet, a weakly supervised segmentation model using scribble annotations.
• Integrated differentiable feature clustering and spatial attention mechanisms.
• Implemented spatial continuity optimization for natural lesion distribution and computational efficiency.

Main Results:

• GloW-VSNet achieved state-of-the-art performance on multiple public and private vitiligo datasets.
• Demonstrated improved segmentation accuracy despite challenging image conditions.
• Showcased effectiveness in handling small and sparse vitiligo lesions.

Conclusions:

• GloW-VSNet represents a significant advancement in weakly supervised global-view vitiligo segmentation.
• The method addresses a critical research gap, enabling more objective disease assessment.
• Offers a practical solution for improved vitiligo severity quantification and treatment monitoring.