Validation of a method to differentiate arterial and venous vessels in CT perfusion data using linear combinations of quantitative time-density curve characteristics
- 231 Downloads
We aimed to develop and evaluate a new method that reliably differentiates between cerebral arteries and veins using voxel-wise CT-perfusion-derived parameters.
Materials and Methods
Fourteen consecutive patients with suspected stroke but without pathological findings were examined on a multi-detector CT system: 32 dynamic phases (∆t = 1.5 s) during application of 35 mL iomeprol-350 were acquired at 80 kV/200mAs. Three hemodynamic parameters were calculated for 18 arterial and venous vessel segments: A (maximum of the time-density-curve), T (time-to-peak), and W (full-width-at-half-maximum). Using receiver operator characteristic (ROC) curve analysis and Fisher’s linear discriminant analysis (FLDA), the performance of every classifier (A, T, W) and of all linear combinations for the differentiation of arterial and venous vessels was determined.
A maximum area under the ROC-curve (AUC) of 0.945 (accuracy = 86.8 %) was obtained using the FLDA combination of A&T or the triplet FLDA of A&T&W for the classification of venous and arterial vessels. The best single parameter was T with an AUC of 0.871 (accuracy = 79.0 %), which performed significantly worse than the combination A&T (p < 0.001).
Arteries and veins can be accurately differentiated based on dynamic CT perfusion data using the maximum of the time-density curve, its time-to-peak, its width, and FLDA combinations of these parameters, which yield accuracies up to 87 %.
• For classification of cerebral vasculature, time-to-peak has the best single-parameter accuracy.
• Fisher’s linear discriminant analysis improves the performance of the individual classifiers.
• Combining signal maximum and time-to-peak parameters significantly increased the classifying potential.
• Pre-processing of time-density-curves by Gaussian filtering or fitting can improve diagnostic accuracy.
KeywordsX-ray computed tomography Diagnostic imaging Angiography Discriminant analysis Brain
The scientific guarantor of this publication is Olaf Dietrich. The authors of this manuscript declare no relationships with any companies, whose products or services may be related to the subject matter of the article. The authors state that this work has not received any funding. No complex statistical methods were necessary for this paper. Institutional Review Board approval was obtained. Written informed consent was waived by the Institutional Review Board. Methodology: retrospective, experimental, performed at one institution.
- 15.Sourbron S, Biffar AF, Ingrisch M, et al. (2009) PMI0.4: platform for research in medical imaging. Proc. ESMRMB, AntalyaGoogle Scholar
- 16.R Core Team (2014) R: A Language and Environment for Statistical Computing.Google Scholar
- 19.Bonferroni C (1936) Teoria statistica delle classi e calcolo delle probabilita. Pubbl del R Ist Super di Sci Econ e Commer di Firenze 8:3–62Google Scholar
- 24.Slawski M, Boulesteix A-L, Bernau C (2009) CMA: Synthesis of microarray-based classificationGoogle Scholar