Mobile Strabismus Automatic Detection Method Combining Alternating Cover Test and Deep Learning

Strabismus is a common eye condition in children and the leading cause of amblyopia. Early treatment is more effective in children, making early screening crucial. However, traditional screening relies on professional ophthalmologists, and there is a lack of medical resources in remote areas. Additionally, the combination of image processing and the Hirschberg test has a limited scope for screening. To address these issues, this study proposes a novel strabismus screening algorithm that integrates deep learning, mobile applications, and the alternate cover test, enabling non-experts to perform detection with simple training. The method consists of four stages: facial eye detection (Faster-RCNN), alternate cover test hand detection, corneal and orbital region segmentation (Efficient_UNet), and corneal displacement calculation for strabismus diagnosis. ROC curve analysis of the entire system shows an area under the curve (AUC) of 0.901, sensitivity of 96.91%, and specificity of 83.33%. These results demonstrate the reliability and effectiveness of the screening system for strabismus detection, providing a reliable tool for large-scale screening and clinical practice.