Figure 3.

Performance metrics of AOSLO-net and nnUNet as well as other deep-learning models on segmenting the AOSLO images. Histogram and density distribution of Dice score (a–e) and IoU (f–j) obtained from AOSLO-net and nnUNet when they are trained using (a, f) perfusion map and normal mask set, (b, g) enhanced AOSLO images and normal mask set, (c,h) two-channel MA images and normal mask set, (d, i) two-channel MA images and short mask set, (e, j) two-channel MA images and thick mask set. The result of AOSLO-net is denoted by cyan color whereas the result of nnUNet is denoted by red color. The dash lines denote the mean values of the Dice or IoU distributions. (k) The overall performance (mean value), performance stability (standard deviation), the worst and best case performance (minimum and maximum value) of Dice score of different AI models in segmenting AOSLO images.