Automated choroidal layer segmentation from en face swept-source optical coherence tomography images of normal eyes using machine learning

Abstract

Abstract The study aims to use machine learning in healthy eyes to develop an automated method to segment the choroidal layers of en-face swept-source optical coherence tomography (SS-OCT) images. We included 117 eyes of 117 healthy subjects who underwent an SS-OCT volume scan with a 12 x 9 mm range. SS-OCT en face images of the choroid were obtained every 2.6 µm from Bruch’s membrane (BM) to the chorioscleral border. The images at the start of the choriocapillaris, the onset of Sattler’s layer, and the beginning of Haller’s layer were identified, and the image numbers from the BM line were taken as the teacher data. Through the Boundary-Enhancing undersampling and sub-class ensemble learning, we obtained a balanced accuracy of 85.54% with an error range of 0 and 92.82% with an error range of 2. Automated stratification of the choroid in en- face SS-OCT images, including choroidal vessels outside the macula, can be done accurately through machine learning.