Prediction of Causative Genes in Inherited Retinal Disorders from Fundus photography and Fundus Autofluorescence Imaging Utilising Deep Learning Techniques

Abstract

Purpose : Inherited retinal disorders (IRDs) are rare intractable diseases, and the access to specialists is difficult world widely. The purpose of this study is to investigate the utility of a data-driven deep learning approach in patients with IRDs, to predict the causative genes based on fundus photography and fundus autofluorescence (FAF) imaging.

Methods : Clinical and genetic data from 156 Japanese subjects with IRDs or no ocular diseases registered to the database of the Japan Eye Genetics Consortium were reviewed. Three categories of genetic diagnosis were selected, based on the highest prevalence of their causative genes: Stargardt disease (ABCA4), retinitis pigmentosa (EYS), and occult macular dystrophy (RP1L1). Fundus photographs and FAF images were cropped in a standardised manner with a macro algorithm. Algorithms for pipeline analyses, based on TensorFlow Inception V-3 were determined with learning parameters (provided by Medic Mind). Images for learning/testing were selected with a randomised 4-fold cross-validation method. The application program interface was established to reach the learning accuracy of concordance (aimed >80%) between the genetic diagnosis and the machine diagnosis (ABCA4, EYS, RP1L1, and normal).

Results : A total of 417 images were examined. The mean overall test accuracy for fundus photographs and FAF images was 88.2% (range, 81.5%-94.4%) and 81.3% (range, 73.5%-87.8%), respectively. The mean overall sensitivity/specificity for fundus photographs and FAF images were 88.3%/97.4% and 81.8%/95.5%, respectively. The mean test sensitivity/specificity per gene category for fundus photographs and FAF images was 88,2%/100% and 97.5%/94.8% for ABCA4-retinopathy, 88.4%/98.1% and 70.7%/99.2% for EYS-retinopathy, 94.4%/92.9% and 64.9%/96.3% for RP1L1-retinopathy, and 82.9%/96.7% and 92.9%/96.3% for normal.

Conclusions : A novel application of deep neural networks in the prediction of the causative gene in IRD from fundus photographs and FAF was highlighted, with a high prediction accuracy of over 80%. These achievements will extensively promote the quality of medical care by facilitating early diagnosis, reducing the cost for referrals, and allowing unnecessary clinical and genetic testing to be avoided.

This is a 2021 ARVO Annual Meeting abstract.