CT perfusion analysis by nonlinear regression for predicting hemorrhagic transformation in ischemic stroke
Bennink, Edwin, Horsch, Alexander D., Dankbaar, Jan Willem, Velthuis, BK, Viergever, Max A., de Jong, Hugo W. A. M.
DOI: https://doi.org/10.1118/1.4923751
Medical Physics 42 (8), p. 4610-4618
Abstract
Purpose: Intravenous thrombolysis can improve clinical outcome in acute ischemic stroke patients but increases the risk of hemorrhagic transformation (HT). Blood-brain barrier damage, which can be quantified by the vascular permeability for contrast agents, is a potential predictor for HT. This study aimed to assess whether this prediction can be improved by measuring vascular permeability using a novel fast nonlinear regression (NLR) method instead of Patlak analysis.
Methods: From a prospective ischemic stroke multicenter cohort study, 20 patients with HT on follow-up imaging and 40 patients without HT were selected. The permeability transfer constant Ktrans was measured in three ways; using standard Patlak analysis, Patlak analysis with a fixed offset, and the NLR method. In addition, the permeability-surface (PS) area product and the conventional perfusion parameters (blood volume, flow, and mean transit time) were measured using the NLR method. Relative values were calculated in two ways, i.e., by dividing the average in the infarct core by the average in the contralateral hemisphere, and by dividing the average in the ipsilateral hemisphere by the average in the contralateral hemisphere. Mann-Whitney U tests and receiver operating characteristic (ROC) analyses were performed to assess the discriminative power of each of the relative parameters.
Results: Both the infarct-core and whole-hemisphere averaged relative Ktrans (rKtrans) values, measured with the NLR method, were significantly higher in the patients who developed HT as compared with those who did not. The rKtrans measured with standard Patlak analysis was not significantly different. The relative PS (rPS), measured with NLR, had the highest discriminative power (P = 0.002). ROC analysis of rPS showed an area under the curve (AUC) of 0.75 (95% confidence interval: 0.62-0.89) and a sensitivity of 0.75 at a specificity of 0.75. The AUCs of the Patlak rKtrans, the Patlak rKtrans with fixed offset, and the NLR rKtrans were 0.58, 0.66, and 0.67, respectively.
Conclusions: CT perfusion analysis may aid in predicting HT, but standard Patlak analysis did not provide estimates for rKtrans that were significantly higher in the HT group. The rPS, measured in the infarct core with NLR, had superior discriminative power compared with Ktrans measured with either Patlak analysis with a fixed offset or NLR, and conventional perfusion parameters. (C) 2015 American Association of Physicists in Medicine.