E-POSTER DETAIL | 11th Asia Pacific Tele Ophthalmology Society Symposium

Title
Artificial Intelligence-Enhanced Multi-modal Analysis of Choroidal Structural and Vascular Remodelling in Geographic Atrophy

Authors
Anais Monet, Kah Meng Ang, Sabrina Noritake, Denise Ng, Yann Malato, Weiwei Luo

Presenting
Anais Monet

PURPOSE:
To investigate the spatial relationship between retinal atrophy and underlying choroidal alterations using AI (Artificial Intelligence)-powered imaging analytics in a preclinical model of GA (Geographic Atrophy), with histological validation.

METHODS:
A laser‑induced GA mouse model (n=14) underwent longitudinal 3D‑OCT (Optical Coherence Tomography) imaging. Retinal and choroidal layers were segmented using a customized EfficientNet‑B7 pipeline. To enhance vascular visualisation without additional hardware, a pseudo‑OCTA (OCT Angiography) motion‑contrast method isolated decorrelation signals from retinal and choroidal networks. Structural en face maps quantified hyper‑transmission (“window defect”), while pseudo‑OCTA assessed vascular remodelling. OPL (Outer plexiform layer)‑derived lesion masks were projected onto choroidal maps to compare choroidal involvement with retinal collapse.

METHODS:

A laser‑induced GA mouse model (n=14) underwent longitudinal 3D‑OCT (Optical Coherence Tomography) imaging. Retinal and choroidal layers were segmented using a customized EfficientNet‑B7 pipeline. To enhance vascular visualisation without additional hardware, a pseudo‑OCTA (OCT Angiography) motion‑contrast method isolated decorrelation signals from retinal and choroidal networks. Structural en face maps quantified hyper‑transmission (“window defect”), while pseudo‑OCTA assessed vascular remodelling. OPL (Outer plexiform layer)‑derived lesion masks were projected onto choroidal maps to compare choroidal involvement with retinal collapse.

RESULTS:
Quantitative analysis showed that choroidal structural changes consistently extended beyond OPL-defined lesion boundaries, with choroidal signal spread exceeding focal retinal atrophy. By Day 14, a clear “window defect” appeared, marked by hyper-reflective visualisation of large choroidal vessels. Pseudo-OCTA demonstrated disrupted choriocapillaris signal within the lesion. Longitudinal imaging confirmed that retinal atrophy remained localised, whereas choroidal signal abnormalities persisted with a wider footprint through Day 28. Histology (Hematoxylin and Eosin) corroborated these findings, revealing RPE (Retinal Pigment Epithelium) fragmentation and outer retinal collapse.

RESULTS:

Quantitative analysis showed that choroidal structural changes consistently extended beyond OPL-defined lesion boundaries, with choroidal signal spread exceeding focal retinal atrophy. By Day 14, a clear “window defect” appeared, marked by hyper-reflective visualisation of large choroidal vessels. Pseudo-OCTA demonstrated disrupted choriocapillaris signal within the lesion. Longitudinal imaging confirmed that retinal atrophy remained localised, whereas choroidal signal abnormalities persisted with a wider footprint through Day 28. Histology (Hematoxylin and Eosin) corroborated these findings, revealing RPE (Retinal Pigment Epithelium) fragmentation and outer retinal collapse.

CONCLUSIONS:
OCT-derived choroidal metrics reveal spatially expanded alterations that exceed focal retinal collapse. This analytics framework demonstrates that pseudo-OCTA post-processing can enhance vascular signals when specialised motion-contrast OCTA is unavailable. By characterizing structural window defects and vascular remodelling, this pipeline provides a scalable solution for image analytics in high-throughput preclinical research.

CONCLUSIONS:

OCT-derived choroidal metrics reveal spatially expanded alterations that exceed focal retinal collapse. This analytics framework demonstrates that pseudo-OCTA post-processing can enhance vascular signals when specialised motion-contrast OCTA is unavailable. By characterizing structural window defects and vascular remodelling, this pipeline provides a scalable solution for image analytics in high-throughput preclinical research.