Introduction: Studies investigating the value of CT perfusion (CTP) imaging in treatment decision or outcome prediction of acute stroke patients have found conflicting results. This may be attributed to variable accuracy and reliability across the different deconvolution algorithms. The aim of this study was to introduce a new, standardized and model-free method, based on similarities in signal time-curves (Pearson’s correlation coefficient) and to increase robustness of CTP analysis and perfusion deficit detection.
Methods: Acute stroke patients from the Dutch Acute Stroke Study were included. CTP data acquired at admission was analyzed using a deconvolution method (Philips Brain Perfusion software) and with TSP. Acute CTP and follow-up non-contrast images after 3 days were interpreted by experienced and inexperienced raters for presence of perfusion deficits, intra-rater and inter-rater agreement.
Results: 65 patients (68±13 years) were included. Example images are shown in the Figure. A perfusion deficit was detected in 56 patients on the MTT deconvolution maps; TSP detected 54 of these perfusion deficits. The agreement of MTT, TTP and TSP with the presence of ischemia on follow-up was comparable, but noticeably lower for CBV. CBV had the best relationship with final infarct volume (R2=0.77, p<0.001), closely followed by TSP (R2=0.63, p<0.001). Inter-rater agreement of experienced readers was comparable across maps. Intra-rater agreement of an inexperienced reader was higher for TSP than for CBV/MTT (kappa’s of 0.79-0.84 versus 0.63-0.7).
Conclusion: TSP offers a very fast, automated, non-proprietary, analysis of CTP data. TSP maps are highly comparable to deconvolution maps, but easier to interpret for inexperienced readers. The relationship of TSP perfusion deficit volume with final infarct volume is comparable to CBV. This suggests that TSP might be a valuable alternative to deconvolution-derived maps, which are known to be prone to tracer delays.