Smartphone-Ai-Based Keratoconus Detection Using Placido Rings

To investigate the possibility of detecting cone-shaped and inferior steepening of the cornea in keratoconus (KCN) patients using smartphone images with a deep learning model.

A customized placido-like ring pattern consisting of 19 red circles on a black background, with a white dot in the center of circles for monocular gaze in all patients, has been developed by our team. This pattern is displayed on one smartphone screen, while another smartphone is used to capture the reflected image of the pattern in the eye, maintaining a distance of 2 cm between the lens of the smartphone and the targeted eye. All images were taken by a single expert person in a dark room.

For this study, 30 keratoconus (KCN) eyes of 30 patients with randomized grades of the disease, while considering no intervention in the cornea, and 30 normal eyes were included.  This work explores possibility of developing a smartphone-based system for keratoconus (KCN) detection. The captured image is then loaded and preprocessed using OpenCV (cv2) and fed into a deep learning model implemented with PyTorch. Different models are used with different backbones to find the best feature extraction with transfer learning method. Furthermore, a MLP classifier is used to distinguish KCN cases from healthy patients. 

This model is able to distinguish between KCN and normal eyes with an accuracy of 98.84%, Considering that all patients have been undergo of ophthalmic visits and KCN patients have pentacam imaging and 2 expert ophthalmologists proof for their disease.However, to further improve accuracy for detecting the grade of KCN at different periods of a patient's life, it will be necessary to continue this idea with a larger sample size and follow-up of patients for monitoring.

The feasibility of this idea has been evaluated and has shown promise for the future sustainability of ophthalmic visits and monitoring of keratoconus (KCN) patients across mild, moderate, severe, and advanced grades. Additionally, the system's portability makes it a potential telemedicine application, aiding healthcare professionals.