Artificial Intelligence Techniques for Diagnosing and Quantifying Vitiligo: A Systematic Review and Meta-Analysis

The utilization of machine learning (ML) and deep learning (DL) for vitiligo, particularly in image and object recognition, allows for the creation of innovative techniques that may match the precision and accuracy of current clinical methodologies. This systematic review seeks to explore how imaging-based artificial intelligence (AI) models may improve the accuracy of vitiligo diagnosis, severity assessment, and monitoring. A comprehensive literature search was conducted across PubMed, IEEE Xplore, EMBASE, Compendex, and Cochrane. Fifty-four studies from January 1, 2000, to October 10th, 2024, focusing on vitiligo diagnosis, classification, treatment, or disease severity were included in this review. Meta-analysis showed models achieved high diagnostic accuracy with an area under the curve of over 0.90 and lesion segmentation with a Dice score of 0.84. Often results matched human expert performance. Studies comparing AI models with experts reported AI could perform equally to that of human experts. These studies highlight the ability of ML models to accurately detect vitiligo and provide quantitative data on the extent of lesions. While AI has shown considerable potential for the improvement of vitiligo management, larger, more diverse publicly available vitiligo datasets and studies focusing on vitiligo segmentation with full-body images are needed for full clinical integration.