Deep learning for interactive and automated inner retinal layer segmentation in OCT images of patients with retinitis pigmentosa using limited training data

The study presents two deep learning models for segmentation of inner retinal layers in patients with retinitis pigmentosa (RP) using OCT images. The first model is based on Segment Anything Model (SAM), the second on nnU-Net. The research included SD-OCT images from 37 RP patients and 1,700 segmented B-Scans from open databases for pretraining. Models were trained on 228 annotated RP scans. nnU-Net achieved precision of 0.96 and F-1 score of 0.96, while OCT-SAM showed lower values of 0.93, 0.8, and 0.85. OCT-SAM demonstrated good agreement with manual annotation and longitudinal reproducibility. Both models successfully detected characteristic pathologies and degenerative changes of RP.