Improving the prediction of final infarct size in acute stroke with bolus delay–corrected perfusion MRI measures

Abstract

To investigate whether bolus delay-corrected dynamic susceptibility contrast (DSC) perfusion MRI measures allowed a more accurate estimation of eventual infarct volume in 14 acute stroke patients using a predictive tissue classifier algorithm. Tissue classification was performed using a expectation maximization and k-means clustering algorithm utilizing diffusion and T2 measures (diffusion-weighted imaging [DWI], apparent diffusion coefficient [ADC], and T2) combined with uncorrected perfusion measures cerebral blood flow ((CBF) and mean transit time [MTT]), bolus delay-corrected perfusion measures (cCBF and cMTT), and bolus delay-corrected perfusion indices (cCBF and cMTT with bolus delay). The mean similarity index (SI), a kappa-based correlation statistic reflecting the pixel-by-pixel classification agreement between predicted and 30-day T2 lesion volumes, were 0.55 +/- 0.19, 0.61 +/- 0.15 (P < 0.02) and 0.60 +/- 0.17 (P <0.03), respectively. Spearman's correlation coefficients, comparing predicted and final lesion volumes were 0.56 (P < 0.05), 0.70 (P < 0.01), and 0.84 (P < 0.001), respectively. We found a more significant correlation between predicted infarct volumes derived from bolus delay-corrected perfusion measures than from conventional perfusion measures when combined with diffusion measures and compared with final lesion volumes measured on 30-day T2 MRI scans. Bolus delay-corrected perfusion measures enable an improved prediction of infarct evolution and evaluation of the hemodynamic status of neuronal tissue in acute stroke.