Automatic Selection of CT Perfusion Datasets Unsuitable for CTP Analysis due to Head Movement

Abstract

CT Brain Perfusion imaging (CTP) is a diagnostic tool for initial evaluation of acute ischemic stroke patients. Head movement of the patients during acquisition limits its applicability. CTP data with excessive head movement must be excluded or corrected for accurate CTP analysis. Instead of manual selection by visual inspection, this study provides an automatic method to select unsuitable CTP data subject to excessive head movement. We propose a 3D image-registration based movement measurement that provides 6 rigid transformation parameters: 3 rotation angles and 3 translations. This method is based on the registration of CTP datasets with a non contrast CT image with a larger volume of interest as reference, which is always available as part of a standard protocol for stroke patients. All parameters from the 3D registration are compared to a set of threshold value to objectively decide whether the CTP dataset suitable for accurate CTP analysis or not. Thresholds for unacceptable head movement were derived using controlled movement experiments with CTP phantom data. Validation was done by comparing the automatic selection of unsuitable data with radiologists’ manual selection using binary classification analysis. The accuracy of the method was 77% with a high sensitivity (95%) and fair specificity (56%). Since all these processes are carried automatically, it assures that clinical decision are not based upon faulty CTP analysis of data with head movement and it saves time in acute time-critically situations for acute ischemic stroke patients.