[1Yin-A] Poster 1

[1Yin-A-62] Investigating Three-Dimensional Optic Nerve Head Structure Estimation from Monocular Fundus Images

〇Zongxian Li¹, Goshiro Yamamoto¹, Sho Mitarai¹, Chang Liu¹, Kazumasa Kishimoto¹, Akihiro Tsutsumino¹, Kenji Suda¹, Hiroshi Tamura¹ (1. Kyoto University)

Keywords:

Machine Learning,3D Reconstruction,Medical Image Analysis

Early diagnosis of glaucoma requires three-dimensional structural assessment of the optic nerve head (ONH). While Optical Coherence Tomography (OCT) provides high-fidelity volumetric structural information, its high cost and operational constraints limit its applicability to large-scale screening. In contrast, fundus photography is widely available but lacks explicit depth information. The objective of this study is to investigate the feasibility of estimating ONH three-dimensional structures with clinical value from fundus images using paired fundus–OCT data. A three-dimensional ONH reconstruction pipeline based on OCT is first established as a structural reference, followed by depth estimation and three-dimensional surface reconstruction from fundus images using deep learning. At the current stage, analysis indicates that fundus-based depth estimation preserves relative structural trends observed in OCT-derived structures, particularly with respect to major ONH structural features such as cup depth and rim configuration. This work focuses on establishing the reconstruction pipeline and examining structural consistency, providing a foundation for subsequent quantitative evaluation and clinical application.