Inter-Relationships Between the Deep Learning-Based Pachychoroid Index and Clinical Features Associated with Neovascular Age-Related Macular Degeneration.

Background/Objectives: To investigate the impact of pachychoroid on the clinical features of neovascular age-related macular degeneration (nAMD) in Japan using the deep learning-based Hokkaido University pachychoroid index (HUPI), which has a high discriminative ability for pachychoroid. Methods: This retrospective observational study examined 124 eyes of 111 treatment-naïve nAMD patients, including 44 eyes with type 1 macular neovascularization (MNV), 26 eyes with type 2 MNV, and 54 eyes with polypoidal choroidal vasculopathy (PCV). HUPI was calculated for each eye from EDI-OCT choroidal images using our modified LeNet that had learned the image patterns of pachychoroid. Differences in HUPI between nAMD types and inter-relationships between nAMD parameters, including HUPI, were evaluated. Results: The mean HUPI was 0.53 ± 0.30 for type 1 MNV, 0.33 ± 0.23 for type 2 MNV, and 0.61 ± 0.3 for PCV, with significant differences between any two of the three groups (p < 0.05, for each). Round-robin multiple regression analysis for nAMD parameters showed the close associations of the HUPI with choroidal vascular hyperpermeability (CVH) and subretinal fluid (SRF) (p = 0.017 and p < 0.001 for each) and the clear division of nAMD parameters into the following two groups: one including intraretinal fluid and type 1 and type 2 MNV and the other including SRF, CVH, polypoidal lesions, and HUPI. Conclusions: HUPI revealed that eyes with type 1 MNV and PCV had more pachychoroid-like features than eyes with type 2 MNV. HUPI was tightly associated with CVH and SRF but not MNV per se in nAMD parameters, reinforcing the pathoetiological concept of differentiating pachychoroid from typical nAMD.