Future direction for the deployment of deep learning artificial intelligence: Vision threatening disease detection in underserved communities during COVID-19

摘要

PURPOSE: Vision-threatening diseases (VTDs) are the leading causes of vision loss and blindness. Use of deep learning artificial intelligence (DLAI) in early detection and subspecialty referral is critical to saving vision years and maintaining quality of life. To address this, we propose a comprehensive community-based screening retinal approach that incorporates DLAI to mitigate disparities and address need in an underserved urban community. METHODS: We evaluated two DLAI software designed for 45° retinal image analysis. DLAI was deployed in clinical settings to triage cases for ophthalmic referrals. Functionality was evaluated to propose implementation in a community screening. RESULTS: Our community screenings have incorporated various imaging modalities to improve VTD pick up rate: nonmydriatic color retinal imaging (18%), fundus autofluorescence (AF) (23%), and ocular coherence tomography B and angiography scans (35%). Robotic teleconsultation increased follow-up reached 100%. In clinical settings, DLAI reduced image analysis time (EyeArt™ = under 38 s, SELENA+™ =10.6 s) and highlighted multiple VTDs. High concordance was observed between human graders and DLAI ( k = 0.68 in the department of endocrinology and k = 1 in the emergency department). CONCLUSION: Integration of DLAI in our ocular screening protocol can be used to reach underserved communities, especially when traditional health-care access is limited.