Table 1 The per-lesion segmentation results of RICAU-Nets for the validation and test sets across all four groups. All four groups utilized non-contrast cardiac CT images for the training, validation, and test sets. The absolute differences between the validation and test sets were calculated to evaluate the generalizability of the architecture. In the extension study involving chest CT, CAC in LM was combined with CAC in LAD due to the challenges in identifying them separately. The two extension groups utilized non-contrast cardiac CT images for the training and validation sets, while non-contrast chest CT images were employed for the test sets. The segmentation performance for the background and bone were not included in the table, as their per-lesion dice scores exceeded 99% in all four groups. For the extension studies, the per-lesion dice scores of the background and bone exceeded 99% and 90%, respectively.

From: Evaluating the generalizability of an automated coronary artery calcium segmentation and scoring algorithm using multi-vendor dataset