Linking Function and Structure with ReSenseNet: Predicting Retinal Sensitivity from Optical Coherence Tomography using Deep Learning


Purpose: Currently used measures of retinal function are limited by being subjective, non-localized and/or taxing for patients. To address these limitations, we sought to develop and evaluate a deep learning (DL) method to automatically predict a functional endpoint (retinal sensitivity) from structural optical coherence tomography (OCT) images. Design: Retrospective cross-sectional study. Subjects: In total, 714 volumes of 289 patients were used in this study. Methods: A novel deep learning algorithm was developed to automatically predict a comprehensive retinal sensitivity map from OCTs. 463 SD-OCT volumes from 174 patients and their corresponding microperimetry examinations (Nidek MP-1) were used for development and internal validation, with a total of 15,563 retinal sensitivity measurements. Patients presented either a healthy macula, early or intermediate AMD, choroidal neovascularization (CNV) or geographic atrophy (GA). In addition, an external validation with 251 volumes of 115 patients was performed, comprising three different patient populations suffering from diabetic macular edema (DME), retinal vein occlusion (RVO) or epiretinal membrane (ERM). Main Outcome Measures: We evaluated the performance of the algorithm using Mean Absolute Error (MAE), limits of agreement (LoA), and correlation coefficients of point-wise sensitivity (PWS) and mean sensitivity (MS). Results: The algorithm achieved a MAE of 2.34dB/1.30dB for point-wise sensitivity (PWS)/mean sensitivity (MS), LoA of 5.70/3.07 for PWS/MS and Pearson and Spearman correlation coefficients of 0.66/0.84 and 0.68/0.83 for PWS/MS. On the external test set, the method achieved a MAE of 2.73dB/1.66dB for PWS/MS. Conclusions: The proposed approach allows predicting retinal function at each measured location directly from the OCT scan, demonstrating how structural imaging can serve as a surrogate of visual function. Prospectively, the approach may help to complement retinal function measures, explore the linkage between image-based information and retinal functionality, improve disease progression monitoring or provide objective surrogate measures for future clinical trials.

In Ophthalmology: Retina.
José Ignacio Orlando
José Ignacio Orlando
Assistant Researcher

My research interests include machine learning and computer vision techniques for medical imaging applications, mostly centered in ophthalmology.