Fig. 3: Illustration of prediction results in three patients with multiple occlusions.

Correctly predicted bounding boxes by the artificial neural network (ANN) are shown in red, whereas false positive bounding boxes are depicted in orange, whereas for false-negative predictions only the ground truth annotations are shown (green circle). A Patient with a correctly ANN- identified occlusion of the left carotid bifurcation (chronic; first slice, coronal), the right middle cerebral artery (MCA) M2-segment (second slice, coronal) and anterior cerebral artery (ACA) A3-segment (third slice, sagittal). In addition, a false positive finding of left MCA M2-segment occlusion was identified by the ANN (second slice, coronal) possibly due to decreased opacification of the vessel due to chronic occlusion of the left internal carotid artery. B Patient with occlusion of the right common carotid artery (CCA) not detected by the ANN (first slice, coronal) however false positive prediction of a right carotid bifurcation occlusion distal from the CCA occlusion (second slice, axial). True positive detection of an additional right proximal M1-segment occlusion and false positive of a more distally located M1 occlusion (third slice, coronal). False positive detection of a left MCA M2-segment occlusion (fourth slice, axial), and false negative detection of a right ACA A2-segment occlusion (fifth slice, sagittal). C Patient with tandem occlusion of the carotid bifurcation and the proximal MCA M1-segment correctly detected by the ANN (first slice, both occlusion visible, sagittal; second slice - ICA M1-segment occlusion visible, axial). Images are depicted using radiological orientation (left image side is the patient’s right side).