Title:Choroidal Vessel Segmentation on Indocyanine Green Angiography Images via Human-in-the-Loop Labeling

Abstract:Human-in-the-loop (HITL) strategy has been recently introduced into the field of medical image processing. Indocyanine green angiography (ICGA) stands as a well-established examination for visualizing choroidal vasculature and detecting chorioretinal diseases. However, the intricate nature of choroidal vascular networks makes large-scale manual segmentation of ICGA images challenging. Thus, the study aims to develop a high-precision choroidal vessel segmentation model with limited labor using HITL framework. We utilized a multi-source ICGA dataset, including 55 degree view and ultra-widefield ICGA (UWF-ICGA) images for model development. The choroidal vessel network was pre-segmented by a pre-trained vessel segmentation model, and then manually modified by two ophthalmologists. Choroidal vascular diameter, density, complexity, tortuosity, and branching angle were automatically quantified based on the segmentation. We finally conducted four cycles of HITL. One hundred and fifty 55 degree view ICGA images were used for the first three cycles (50 images per cycle), and twenty UWF-ICGA images for the last cycle. The average time needed to manually correct a pre-segmented ICGA image per cycle reduced from 20 minutes to 1 minute. High segmentation accuracy has been achieved on both 55 degree view ICGA and UWF-ICGA images. Additionally, the multi-dimensional choroidal vascular parameters were significantly associated with various chorioretinal diseases. Our study not only demonstrated the feasibility of the HITL strategy in improving segmentation performance with reduced manual labeling, but also innovatively introduced several risk predictors for choroidal abnormalities.