Cross-Device Transfer Learning for Kayser-Fleischer Ring Detection: From Slit-Lamp to Smartphone

Wilson’s disease (WD) is a genetic disorder caused by copper metabolism abnormalities, one of the characteristics of which is the presence of Kayser-Fleischer (K-F) ring in the eyes, typically diagnosed through slit-lamp examination. This study explores the application of transfer learning in the YOLOv8 model for detecting K-F ring in smartphone images, transitioning from slit-lamp images. By training a slit-lamp model on slit-lamp images, transfer learning enables the adaptation of the model for K-F ring detection in smartphone images. The results demonstrate that the transfer learning-based model outperforms the model trained from scratch, achieving higher precision and better precision-recall balance in the K-F ring detection task using smartphone images. The developed smartphone K-F ring detection model holds significant clinical value for remote healthcare and the auxiliary diagnosis of Wilson disease, particularly in resource-limited settings, where it offers a convenient diagnostic tool for clinicians, potentially improving early diagnosis rates and supporting healthcare services in remote areas.