TeleAEye: Low-Cost Automated Eye Disease Diagnosis Using a Novel Smartphone Fundus Camera With AI
Tienlan Sun
Eric Hamber Secondary
Floor Location : S 017 N

A billion people worldwide live with vision impairment resulting from a lack of access to eye care. Though the World Health Organization has established early diagnosis as a key to solving this global health crisis, the widespread lack of eye doctors has made implementation impossible. To combat this global health crisis, especially during the COVID-19 pandemic, TeleAEye combines a deep learning eye disease diagnosis software with a smartphone-based fundus camera that functions independently of eye doctors and mydriatics. The models are capable of diagnosing six common opthalmological diseases that account for 60% of all blindness: age-related macular degeneration, cataracts, diabetic retinopathy, glaucoma, hypertensive retinopathy and pathological myopia. The exceptional performance of the diagnosis models was achieved through a novel, experimental computer vision technique, “multi-step transfer learning”. These diagnosis models are hosted on a web application that also provides adjustable saliency heatmaps and a personalized artificial intelligence chatbot. The saliency heatmaps improve ophthalmologist diagnosis accuracy by highlighting the areas of the retinal images that contributed the most to the final diagnosis. The AI chatbot provides reliable information to the patient depending on individual diagnosis conditions. The total production cost of under $10 will allow for a previously unimaginable level of access, especially with its portability allowing for transport to remote locations. Currently, partnerships with telemedicine companies, non-profit organizations, and eye doctors are in place to bring TeleAEye to the frontlines of eye care. In the near future, TeleAEye could also have immense implications in democratizing healthcare through the diagnosis of other diseases using fundus images (e.g., Alzheimer’s, autonomic dysreflexia, cancer), the extension of datasets to underrepresented minorities, and the expansion of eye care access through automated diagnoses and telemedicine. The incredible effectiveness of “multi-step transfer learning ” could also transform industries using computer vision including agriculture, retail, automotive, and surveillance.