Automatic Detection of Ocular Surface Disease on Smartphone Images Using Improved YOLOv5

The significant increase in the number of patients with ocular diseases has made medical institutions face great challenges. At present, computer aided diagnosis technology is urgently needed to assist doctors in clinical diagnosis of ocular surface diseases. In view of the application scenario of eye surface image disease detection, we proposed to improve the automatic detection of ocular surface diseases of YOLOv5. The CBAM attention module was introduced in the feature extraction stage of YOLOv5, and C3 module was improved to CBAMC3 module, which enhanced the feature extraction capability and made the backbone network more focused on the lesion area of the eye surface. Subsequently, the BiFPN module was introduced into the neck network to further enhance the feature fusion capability and improve the detection accuracy of eye surface disease images. The experimental results show that the mAP of ocular surface diseases detection by our proposed method in the test set is 97.9%, which realizes the automatic detection and localization of ocular surface diseases, and has good auxiliary diagnostic significance.