Abstract TP45: One Threshold Does not Fit All

    loading  Checking for direct PDF access through Ovid

Abstract

Introduction: The hyperdense sign is a marker of thrombus on non-contrast computed tomography (NCCT). Currently, pre-specified Hounsfield unit (HU) threshold are used arbitrarily for clot segmentation. We aimed to test if HU thresholds best discriminating clot on thin slice NCCT are patient specific and dependent on factors such as age, hematocrit, and NCCT slice thickness.

Methods: Data are from the ESCAPE randomized controlled trial. Only patients with thin slice baseline NCCT (<2.5mm) and M1-MCA occlusion on baseline CT-Angiography (CTA) were included. CTA was co-registered to NCCT using in-house software. Proximal and distal clot interface was identified on CTA and super-imposed onto the NCCT. Three Regions of Interest (ROIs) (36- 100 voxels) were placed on the NCCT in the 1) clot as defined on co-registered CTA, 2) contralateral brain tissue, and 3) contralateral patent M1 MCA artery. Optimal patient specific HU thresholds differentiating “clot” from “normal vessel” and “brain tissue” voxels were calculated using receiver operating characteristic (ROC) analysis.

Results: 70 patients were included. ROC analysis showed that the optimal HU threshold discriminating clot on NCCT varies considerably between patients (Figure 1A). Hematocrit was associated with contralateral artery HU (p<0.001); no significant correlation was found with age (p=0.156) and slice thickness (p=0.473). Predictive ability of contralateral artery and brain tissue HU in determining optimal HU threshold discriminating clot was similar (p<0.001, Figure 1B). Model comparison using Bayesian Information Criterion suggests that we can reasonably use contralateral brain tissue HU or contralateral artery HU to derive patient-specific optimal HU thresholds that best discriminates clot.

Conclusion: HU thresholds on NCCT that best discriminates clot are patient specific. Signal from brain tissue or patent artery silhouette can be used to derive these patient specific thresholds.

Related Topics

    loading  Loading Related Articles