Robust AI to develop risk models in retinopathy of prematurity using deep learning

ROP是一种影响早产儿的视网膜新生血管疾病，是儿童失明的主要原因

ROP is a retinal neovascular disease affecting preterm infants, and is a leading cause of childhood blindness worldwide. Known clinical risk factors include preterm birth, low birthweight and use of supplemental oxygen but improved risk models are needed to identify infants that progress to treatment requiring disease and blindness. Deep learning techniques have been used to successfully identify “plus” disease in multi- institutional cohorts and to provide a continuous measure of disease severity. A major limitation of deep learning, however, is the need for large amounts of well curated datasets. Other limitations include overfitting and “brittleness” that can cause model performance to drop on external data. There are, however, numerous barriers to building and hosting these large central repositories with multi-institutional data required for robust deep learning including concerns about data sharing, regulations costs, patient privacy and intellectual property. In this project, we aim to demonstrate the utility of distributed/federated deep learning approaches where the data are located within institutions, but model parameters are shared with a central server. A major challenge thwarting this research, however, is the requirement for large quantities of labeled image data to train deep learning models. Efforts to create large public centralized collections of image data are hindered by barriers to data sharing, costs of image de-identification, patient privacy concerns, and control over how data are used. Current deep learning models that are being built using data from one or a few institutions are limited by potential overfitting and poor generalizability. Instead of centralizing or sharing patient images, we aim to distribute the training of deep learning models across institutions with computations performed on their local image data. Specifically, we seek to build robust risk models for predicting treatment requiring disease. Two large cohorts will be used to validate the hypothesis that the performance of the risk models using distributed learning approaches that of centrally hosted and is more robust than models built on single institutional datasets. Grants Admin Updated 04.01.2019 JBou