Running pattern of choroidal vessel in en face OCT images determined by machine learning-based quantitative method

Purpose To evaluate the new method to quantitate the running pattern of the vessels in Haller's layer in en face optical coherence tomographic (OCT) images using the new algorithm. Methods A retrospective and cross-sectional study. The en face image of top 25% slab of Haller's layer was analyzed. The vascular area in these images was calculated after binarization. Then, the vessels were thinned, and the total length of the vessels and the mean vessel diameter were calculated. Based on the angle of vessel running, "natural oblique vessel" was defined. The ratio of the natural oblique vessel to the whole vessels was defined as the "symmetry index". To examine the reproducibility of the software, the images obtained on two different examination dates of the same subject (25 eyes of 25 healthy subjects) were analyzed. Also, to compare the symmetry index and subjective evaluations, 180 eyes and 180 healthy subjects were analyzed. The subjective evaluations classified the images into 3 groups, the Symmetrical, Semi-symmetrical, and Asymmetrical types. Symmetry index was compared in each group. Results The inter-measurement correlation coefficient (ICC) of the vessel area, vessel length, and vessel diameter were 0.955, 0.934, and 0.954, respectively. The ICC of the symmetry index was 0.926. The symmetry index of the Symmetrical type was 60.4 +/- 7.2%, that of the Semi-symmetry type was 56.2 +/- 4.6%, and that of the Asymmetry type was 52.6 +/- 5.2%. Conclusions The present algorithm can analyze vessels in Haller's layer of the en face images of choroid in an objective manner with good repeatability.