Abstract
Purpose:
Age-related macular degeneration (AMD), one of complex diseases, has various findings on spectral-domain optical coherence tomography (SD-OCT) images. We herein investigated the macular morphological patterns using multiple OCT parameters in AMD based on the self-organizing map.
Methods:
We retrospectively reviewed consecutive 271 eyes of 271 patients with treatment-naïve AMD. Fourteen OCT parameters and one indocyanine green angiography (IA) parameter, i.e., heights from the choroid (length from inner limiting membrane to Bruch’s membrane) or choroidal thicknesses in five subfields of Early Treatment Diabetic Retinopathy Study grid, heights of foveal cystoid spaces, hyperreflective lesions, retinal detachment (with lower reflectivity), and pigment epithelial detachment (PED), and PED area, were quantified. Using the self-organizing map algorithm, individual cases with mathematical similarity were arranged near each other on the two-dimensional map, followed by the objective clustering.
Results:
Two-dimensional self-organizing map demonstrated that eyes with higher heights from the choroid were arranged to the lower left corner, whereas those with thicker choroid to the lower right corner. Following clustering divided all cases into three segments. Eyes in segment 1 were characterized by lower height from the choroid and thinner choroid; those in segment 2 by lower height from the choroid and thicker choroid; those in segment 3 by higher height from the choroid and thinner choroid. We performed multiple regression analyses and found that logMAR was associated with PED area (β=0.616, P<0.001) and the nasal height from the choroid (β=0.208, P=0.029) in eyes of segment 2, and with heights of foveal cystoid spaces and hyperreflective lesions (β=0.554, P<0.001 and β=0.361, P=0.003) in eyes of segment 3.
Conclusions:
The self-organizing map enabled us to propose an objective clustering using multiple OCT findings and to facilitate integrative understanding of the morphological patterns and the structural/functional relationship in AMD.