Figure 6.

Quantification results for the largest caliber and narrowest caliber factors for the segmented MAs predicted by AOSLO-net and nnUNet trained with three sets of MA masks (normal, short and thick masks). (a) The largest caliber (LC) results for the segmented MAs. The black dashed lines represent the reference MA LC values for different MAs, whereas the red and blue curves are the MA LC quantification results based on the segmentation maps from AOSLO- net and nnUNet using three different masks (from top to the bottom row). (b) Examples for the original enhanced perfusion maps (first column), segmented MAs using AOSLO-net (second column), and nnUNet (third column) trained with different MA masks. The MA segmentation results in yellow and green bounding boxes correspond to the images in (c) marked with yellow and green arrows. (c) Narrowest caliber (NC) quantification results for segmented MAs using AOSLO- net and nnUNet trained with three different MA masks; the black dashed lines represent the reference MA NC values computed using the original three types of MA masks; the red and blue curves represent the NC quantification results using segmentation maps obtained by AOSLO- net and nnUNet using three different masks (from top to the bottom row). Quantification of MA shapes using MAs segmented from AOSLO-net results in smaller discrepancy between the model predictions and groundtruth (dotted lines) due to the better performance of AOSLO-net in segmenting the parenting vessels of MAs.