Bridging the Gap Between Vitiligo Segmentation and Clinical Scores

Area of Science:

• Dermatology
• Medical Imaging
• Computer Vision

Background:

• Quantitative vitiligo evaluation is essential for treatment monitoring but faces challenges with objectivity and workload.
• Current automatic segmentation methods focus on patch-wise images, limiting clinical score calculation and full-body assessment.
• Full-body vitiligo segmentation is needed for comprehensive patient monitoring and objective clinical scoring.

Purpose of the Study:

• To develop and validate a novel algorithm for full-body vitiligo segmentation.
• To address challenges in segmenting vitiligo from full-body images, including varied lighting and lesion distribution.
• To enable objective, clinically translatable scoring of vitiligo extent from full-body images.

Main Methods:

• Establishment of the first full-body vitiligo dataset (1740 images) adhering to international standards.
• Development of a custom algorithm incorporating contrast enhancement and long-range comparison for segmentation.
• Introduction of a confidence score refinement module to handle challenging cases like fully affected or unaffected skin areas.

Main Results:

• The proposed algorithm significantly reduced average per-image vitiligo involvement percentage error from 3.69% to 1.81%.
• Top 10% per-image errors decreased substantially from 23.17% to 8.29%.
• Per-patient vitiligo involvement percentage errors (mean 1.17%, max 3.11%) surpassed experienced dermatologist evaluations.

Conclusions:

• The developed full-body vitiligo segmentation algorithm provides objective and clinically applicable quantitative evaluations.
• The method overcomes limitations of previous patch-wise approaches and handles complex imaging conditions.
• This advancement offers a more accurate and efficient tool for assessing vitiligo progression and treatment response.