Distributed machine learning cloud teleophthalmology IoT for predicting AMD disease progression

Abstract

The ability to perform screening of potential vision-impairing diseases remotely with an Ophthalmologist-in-the-loop is crucial in serving the Medically Underserved Areas/Population (MUA/P) and in acute medical settings, such as emergency departments. With an estimated 217 million individuals affected by moderate to severe vision-impairing diseases worldwide and an increasing number of new patients with such diseases, the need for access to faster (or real-time) diagnosis on a large scale is imperative. It is evident that early diagnosis of chronic diseases such as diabetic retinopathy and age-related macular degeneration (AMD) could better prevent vision loss. In this paper, a scalable cloud based teleophthalmology architecture via the Internet of Medical Things (IoMT) for diagnosis of AMD is presented. In the proposed architecture, patients wear a head-mounted camera (OphthoAI IoMT headset) to send their retinal fundus images to their secure and private cloud drive storage for personalized disease severity detection and predictive progression analysis. A proposed AMD-ResNet convolution neural network with 152 layers will then analyze the images to identify and determine AMD disease severity. The algorithm is trained with AREDS (age related eye disease study) images from the National Institute of Health (NIH) with over 130,000 fundus images captured over 12 years, and for determining AMD severity, we achieve a sensitivity and specificity of 94.97 ± 0.5% and 98.32 ± 0.1% respectively. A temporal Long–Short Term Memory (LSTM) deep neural network for precision medicine and AMD predictive progression is also proposed. Patient personalization allows better targeted care, lesser side effects, and a greater likelihood of responding to treatments by tailoring healthcare on a per-patient basis.