AUTOMATED ANATOMICAL LABELING OF THE CEREBRAL ARTERIES USING BELIEF PROPAGATION

Abstract

Labeling of cerebral vasculature is important for characterization of anatomical variation, quantification of brain morphology with respect to specific vessels, and inter-subject comparisons of vessel properties and abnormalities. We propose an automated method to label the anterior portion of cerebral arteries using a statistical inference method on the Bayesian network representation of the vessel tree. Our approach combines the likelihoods obtained from a random forest classifier trained using vessel centerline features with a belief propagation method integrating the connection probabilities of the cerebral artery network. We evaluate our method on 30 subjects using a leave-one-out validation, and show that it achieves an average correct vessel labeling rate of over 92%.

Keywords: Automated labeling of vessels; belief propagation; cerebral arteries; random forest; statistical inference on Bayesian networks.

Figure 1

Vessel segmentation and network representation for one subject. Each node in the graph represents a vessel centerline segment, and the edges reflect the connections between the vessel segments. Edges are directed away from the internal carotid arteries. Legend: ICA=internal carotid a., ophth=ophthalmic a., ACA=anterior cerebral a., MCA=middle cerebral a., Acomm=anterior communicating a. First letter (L or R) indicates the cerebral hemisphere.

Figure 2

Fraction of vessels correctly labeled, averaged across 10 experiments for each subject. Blue columns are the results of the random forest classifier (RF), and red columns show the results of RF followed by belief propagation (RF+BP). The right-most column pair is the average correct labeling rates across all 30 ± 10 = 300 experiments.

Figure 3

Confusion matrices for the RF only method (left) and RF+BP (right) for all vessel segments across 30 subjects. True labels are on the horizontal axis, and the predicted labels on the vertical axis. The values presented in the matrices are fraction of vessels with true label x that were assigned label y (i.e. values in each column add up to 1).