Abstract
Purpose: :
Recent advances in optical coherence tomography (OCT) have elucidated the clinical relevance in several novel findings in diabetic macular edema (DME), although integrated evaluation of these parameters remains to be performed. Self-organizing map (SOM) is an algorithm in which similar nodes are arranged near each other, resulting in the clustering without supervisors. In this study, we applied SOM for the morphological patterns of the macula in diabetic retinopathy (DR).
Methods: :
Consecutive 178 eyes from 124 patients with DR on which the images of spectral domain (SD)-OCT (Spectralis OCT, Heidelberg Engineering) with enough quality were obtained were included. We evaluated several parameters; retinal thickness, status of external limiting membrane (ELM) at the fovea, height of foveal cystoid spaces, height of foveal serous retinal detachment (SRD), cystoid spaces in the parafovea, and logMAR. All cases with such multiple parameters were analyzed using batch-learning SOM algorithm, Viscovery SOMine trial version (Viscovery Software GmbH, Vienna, Austria), and arranged within 2-dimensional map.
Results: :
SOM demonstrated several patterns of macular morphologies, and the eyes with foveal SRD rarely had foveal cystoid spaces, and vice versa. Especially, eyes with foveal cystoid spaces might have several populations; cases with very severe cystoid spaces at the fovea suffering very poor logMAR, cases with moderately poor logMAR presenting cystoid spaces in the parafovea, and cases with better logMAR in the presence of smaller foveal cystoid spaces, but not parafoveal cystoid spaces. We did not find the association between the height of SRD and logMAR. Additionally, disrupted ELM was associated with poor logMAR in eyes without foveal thickening.
Conclusions: :
SOM automatically has provided the clustering of macular pathomorphologies in eyes with DR, suggesting the several patterns of pathogenesis in DME.
Keywords: diabetic retinopathy • retina