Improving classification of microembolus and artifact of HITS event by feature selection

Abstract

In monitoring system using transcranial Doppler ultrasound for stroke detection, the occurrence of high intensity transient signal can happen at different branch of arteries, i.e. internal cerebral artery (ICA), middle cerebral artery (MCA) and posterior cerebral artery (PCA). The representations of features can sometimes be redundant and not useful, which can degrade the classification performance. Thus, feature selection is studied and presented in this paper. The applied selection criteria are based on the unbounded Mahalanobis distance (referred as A) and single-feature-accuracy measure (referred as B). The result indicates that kinematic descriptor (SMV) is the most significant feature to predict HITS with 85.8% correct. However, the classification accuracy further improved when SMV is combined with other features in different feature subsets.