Deep learning predicts OCT measures of diabetic macular thickening from color fundus photographs

Abstract

Purpose : As part of Roche's comprehensive initiative in Ophthalmology Personalized Healthcare, we have developed deep learning (DL) models for the automatic detection of optical coherence tomography (OCT) measures of diabetic macular thickening (MT) from color fundus photographs (CFPs).

Methods : Retrospective analysis on 17,997 CFPs and their associated time domain OCT (TD-OCT) measurements from the phase 3 RIDE/RISE (NCT00473382/NCT00473330) diabetic macular edema (DME) studies. The CFP dataset was split as follows: 80% for training, 10% for testing, and 10% for validation. DL was applied on CFPs to predict TD-OCT equivalent measures of MT, including central subfield thickness (CST) and central foveal thickness (CFT). MT was defined using 2 OCT cutoff points: 250 µm and 400 µm. The cutoff point of 250 µm measured with TD-OCT is traditionally used to discriminate normal patients from those with abnormal MT. The cutoff point of 400 µm has traditionally been used to identify cases of severe DME in TD-OCT studies. In order to directly quantify the actual CFT and CST from CFPs, a DL regression model was developed. The Inception-V3 architecture, trained using a transfer learning cascade, was used to address both the binary classification and the regression tasks. We estimated the area under the curve (AUC) to assess the performance of the DL models for classifications, while we utilized the R² value to benchmark the DL regression model.

Results : The best DL model was able to predict CST ≥ 250 µm and CFT ≥ 250 µm, with an AUC of 0.97 (95% CI, 0.89–1.00) and 0.91 (95% CI, 0.76–0.99), respectively. To predict CST ≥ 400 µm and CFT ≥ 400 µm, the best DL model had an AUC of 0.94 (95% CI, 0.82–1.00) and 0.96 (95% CI, 0.88–1.00), respectively. The best deep convolutional neural network regression model to quantify CST and CFT had an R² of 0.74 (95% CI, 0.49–0.91) and 0.54 (95% CI, 0.20–0.87), respectively. The performance of the DL models increased when trained on high-quality CFPs, but decreased with the presence of laser scars.

Conclusions : DL is capable of predicting key quantitative TD-OCT measurements related to MT from CFPs. The DL models presented here could enhance the efficiency of DME diagnosis in tele-ophthalmology programs, promoting better visual outcomes. Future research is needed to validate DL algorithms for MT in the real world.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.