Skip to main content
. 2020 Dec 8;14:592352. doi: 10.3389/fnins.2020.592352

Figure 9.

Figure 9

Centerline prediction on MRA TOF (A) and Synthetic (B) test datasets using DeepVesselNet-FCN (centerline in green). There are more detections in smaller vessels than in larger vessels which can be explained by the network seeing more smaller vessels than bigger vessels during training. Again, centerline detection in MRA TOF covers all the vessels with missing points and can be improved by finetuning on annotated MRA data or by a post-processing strategy to fill in the missing points.