Fig 1.
Non-pathological segments labeled as 1 and the plaque labeled as 2.
Fig 2.
Example of stretched reconstruction of coronary arteries.
2D view of 5 largest arteries of a patient (a), Annotation overlay on same arteries with light gray as segment with plaques and dark gray as non-pathological segments (b).
Table 1.
Hyper parameter tuning.
Table 2.
Cross-fold performance across validation splits.
Fig 3.
Examples of Predictions with dice scores more than 0.88.
The model showed a promising performance in finding medium to large plaque annotations.
Fig 4.
Example of small plaque annotations (< 99 mm3).
The model has failed to identify some smaller non-calcified plaques.
Table 3.
Detection performance stratified by calcification volume.
Table 4.
Detection performance stratified by plaque volume.
Fig 5.
Illustration of model’s performance.
Dice score distribution based on plaque annotation volumes (top). A frequency curve illustrated the distribution of plaque volumes within the dataset (bottom).
Fig 6.
A motion artifact which caused a false prediction as plaque.