Bayesian analysis of perfusion-weighted imaging to predict infarct volume: comparison with singular value decomposition.

PURPOSE: We compared the performances of a Bayesian estimation method and oscillation index singular value decomposition (oSVD) deconvolution for predicting final infarction using data previously obtained from 10 cynomolgus monkeys with permanent unilateral middle cerebral artery (MCA) occlusion. METHODS: We conducted baseline perfusion-weighted imaging 3 hours after MCA occlusion and generated time to peak, first moment of transit, cerebral blood flow, cerebral blood volume, and mean transit time maps using Bayesian and oSVD methods. Final infarct volume was determined by follow-up diffusion-weighted imaging (DWI) scanned 47 hours after MCA occlusion and from histological specimens. We used a region growing technique with various thresholds to determine perfusion abnormality volume. The best threshold was defined when the mean perfusion volume matched the mean final infarct volume, and Pearson's correlation coefficients (r) and intraclass correlations (ICC) were calculated between perfusion abnormality and final infarct volume at that threshold. These coefficients were compared between Bayesian and oSVD using Wilcoxon's signed rank test. P-value < 0.05 was considered a statistically significant difference. RESULTS: The Pearson's correlation coefficients were larger but not significantly different for the Bayesian technique than oSVD in 4 of 5 perfusion maps when final infarct was determined by specimen volume (P = 0.104). When final infarct volume was defined by DWI volume, all perfusion maps had a significantly higher correlation coefficient by Bayesian technique than oSVD (P = 0.043). For ICC, all perfusion maps had higher value in Bayesian than oSVD calculation, and significant differences were observed both on specimen- and DWI-defined volumes (P = 0.043 for both). CONCLUSION: The Bayesian method is more reliable than oSVD deconvolution in estimating final infarct volume.

Assessment of the accuracy of a Bayesian estimation algorithm for perfusion CT by using a digital phantom.

INTRODUCTION: A new deconvolution algorithm, the Bayesian estimation algorithm, was reported to improve the precision of parametric maps created using perfusion computed tomography. However, it remains unclear whether quantitative values generated by this method are more accurate than those generated using optimized deconvolution algorithms of other software packages. Hence, we compared the accuracy of the Bayesian and deconvolution algorithms by using a digital phantom. METHODS: The digital phantom data, in which concentration-time curves reflecting various known values for cerebral blood flow (CBF), cerebral blood volume (CBV), mean transit time (MTT), and tracer delays were embedded, were analyzed using the Bayesian estimation algorithm as well as delay-insensitive singular value decomposition (SVD) algorithms of two software packages that were the best benchmarks in a previous cross-validation study. Correlation and agreement of quantitative values of these algorithms with true values were examined. RESULTS: CBF, CBV, and MTT values estimated by all the algorithms showed strong correlations with the true values (r = 0.91-0.92, 0.97-0.99, and 0.91-0.96, respectively). In addition, the values generated by the Bayesian estimation algorithm for all of these parameters showed good agreement with the true values [intraclass correlation coefficient (ICC) = 0.90, 0.99, and 0.96, respectively], while MTT values from the SVD algorithms were suboptimal (ICC = 0.81-0.82). CONCLUSIONS: Quantitative analysis using a digital phantom revealed that the Bayesian estimation algorithm yielded CBF, CBV, and MTT maps strongly correlated with the true values and MTT maps with better agreement than those produced by delay-insensitive SVD algorithms.

Bayesian hemodynamic parameter estimation by bolus tracking perfusion weighted imaging.

A delay-insensitive probabilistic method for estimating hemodynamic parameters, delays, theoretical residue functions, and concentration time curves by computed tomography (CT) and magnetic resonance (MR) perfusion weighted imaging is presented. Only a mild stationarity hypothesis is made beyond the standard perfusion model. New microvascular parameters with simple hemodynamic interpretation are naturally introduced. Simulations on standard digital phantoms show that the method outperforms the oscillating singular value decomposition (oSVD) method in terms of goodness-of-fit, linearity, statistical and systematic errors on all parameters, especially at low signal-to-noise ratios (SNRs). Delay is always estimated sharply with user-supplied resolution and is purely arterial, by contrast to oSVD time-to-maximum TMAX that is very noisy and biased by mean transit time (MTT), blood volume, and SNR. Residue functions and signals estimates do not suffer overfitting anymore. One CT acute stroke case confirms simulation results and highlights the ability of the method to reliably estimate MTT when SNR is low. Delays look promising for delineating the arterial occlusion territory and collateral circulation.