Assessing Retinal Layers' Association with Diabetes using a Deep Learning Framework

Abstract

Purpose : To quantify the association between retinal layer thickness and diabetes diagnosis with a deep learning approach.

Methods : From the population-based Leipzig Research Center for Civilization Diseases (LIFE) Adult Study, we included participants with macular Spectralis optical coherence tomography (OCT) scans and American Diabetes Association-stratified diabetes status. We exacted the retinal layer thickness maps (RLTMs) for the 10 retinal layers (Figure 1) segmented by the Heidelberg Engineering software including retinal nerve fiber layer (RNFL), ganglion cell layer (GCL), inner plexiform layer (IPL), inner nuclear layer (INL), outer plexiform layer (OPL), outer nuclear layer (ONL), myoid zone (MZ), ellipsoid zone (EZ) and outer-photoreceptor segment (OS) combined (EZ+OS), interdigitation zone (IZ) and retinal pigment epithelium (RPE). We used each of the 10 layers individually and 10 layers together to predict diabetes diagnosis. We used the area under the receiver operating characteristic curve (AUC) to measure the association strength between each retinal layer and diabetes diagnosis. Participant level separation was applied to divide our data into training and testing subsets with two-thirds and one-third of the total data, respectively.

Results : 17,715 OCT scans from 8,947 participants with information on diabetes status were included in this study. There were 4,901 participants with normal glucose tolerance (NGT), 2,786 with prediabetes, and 1,260 diabetes patients, respectively. Combining 10 layers together, the AUCs to distinguish diabetes from NGT, diabetes from prediabetes and prediabetes from NGT were 0.76, 0.64, and 0.65, respectively. The inner layers were generally more affected including RNFL, GCL, IPL, INL, OPL, and ONL with AUCs ranging from 0.68 to 0.72 consistent with existing clinical knowledge. The top four layers that were most associated with diabetes diagnosis measured by AUCs were ONL, RNFL, RPE, and IPL with AUCs of 0.72, 0.71, 0.69, and 0.69, respectively.

Conclusions : The deep learning approach was able to quantify the association between retinal layer thickness and diabetes status.

This abstract was presented at the 2023 ARVO Imaging in the Eye Conference, held in New Orleans, LA, April 21-22, 2023.

 

Figure 1

 

Figure 2