OpthoAI

A conversational multi-disease AI Ophthalmology smartphone screener with a novel approach to monitor disease progression

Abstract

People who cannot afford annual eye checkups or do not have access to ophthalmologists are at high risk of preventable blindness from various undetected eye diseases, which presents the need for an easily accessible eye disease detector. OpthoAI is a low-cost Artificial Intelligence Ophthalmology screener that detects and monitors the progression of multiple common eye diseases, such as Drusens, Diabetic Retinopathy, Age-related Macular Degeneration, Hemorrhage, Glaucoma, Vascular Occlusion, Macular Edema, and Nevus. It utilizes an inexpensive lens apparatus to capture eye fundus images which can be uploaded into the conversational smartphone app. OpthoAI simulates an actual doctor-patient interaction, provides the patient with educational information, and recommends a follow-up appointment if needed. Chronic patients can reduce the frequency of follow-up visits to the ophthalmologist by using OpthoAI for interim follow-ups that check for treatment efficacy and disease progression. The dataset used for training the model was from the Brazilian Multilabel Ophthalmological Dataset, consisting of 16,266 Fundus images. Multiple disease-specific AI Convolutional Neural Networks (CNNs) were used for the initial disease diagnosis and progression, and each one used MobileNet as the base transfer learning model. Monitoring disease progression of a single disease solely using Artificial Intelligence is challenging and not as accurate as detecting different diseases. Therefore, this paper presents a novel approach to eye disease progression by combining Artificial Intelligence predictions based on eye images with additional inputs: an encoded Snellen eye chart and a patient's self-assessment/history. OpthoAI could successfully detect eight common eye diseases and measure the progression of Diabetic Retinopathy with a high confidence rate.