Improving Arterial Tree Classification in 3D MRA Using Feature Selection and Machine Learning

Abstract

A comprehensive cerebral vascular atlas is crucial for advancing the understanding of brain function and for studying highly complex neurological diseases. The Circle of Willis (CoW), central to cerebral blood circulation, requires precise modeling of its vascular bifurcations to capture inter-individual variability and provide reliable clinical insights. However, the construction of such an atlas is challenging because of the wide range of vascular geometries and the high dimensionality of their features. The classification of vascular bifurcations can be improved by identifying and prioritizing the most discriminating geometric features, which are essential for creating an accurate atlas. In this study, we used several feature selection methods to extract the most informative features from the entire feature set. The results demonstrate that our proposed method not only reduces the dimensionality of the data but also improves the accuracy and robustness of the classifiers.