Multi-parametric predictive model for end infarct volume in stroke patients

We investigated the intra- and inter-rater reliability of ischemic lesion volumes measurements assessed by different MRI sequences at various times from onset. The reliability of these measurements are essential for validating their usage in stroke outcome prediction models. Ischemic lesion volumes were measured for infra-rater reliability using diffusion weighted (DWI), mean transit time (MTT) perfusion and Fluid Attenuated Inversion Recovery (FLAIR) MRI at chronic time points. There was good concordance of the mean sample volumes of the two infra-rater reads (deviations were \u3c4% and 2 cc globally, \u3c2% and 2 cc for DWI, \u3c6% and 7 cc for MTT and \u3c2% and 1 cc for FLAIR). There was also good concordance of the inter-rater reads (\u3c5% and 2 cc globally). Repeat measurements of stroke lesion volumes show excellent intra- and inter-rater correlations and concordance for DWI, MTT and FLAIR at acute through chronic time points. ^ We also investigated two methods of measuring MRI perfusion-diffusion mismatch to determine whether reliability is improved by direct measurement on a single, blended map. Image software was used for measurement of lesion volumes from diffusion-weighted (DWI) and mean transit time (MTT) calculated from perfusion weighted (PWI) images on 64 acute stroke patients. For the first method the DWI and MTT lesions were measured separately. For the second method the mismatch volume was measured directly on the blended images created from the registered DWI and MTT images. Test-retest agreement was 100% and 97% for the separate and blended methods using mismatch cutoffs of \u3e20% versus \u3c20%. There were no significant differences in the mismatch statistics between the methods. Mismatch volumes by a single reader can provide highly reliable and consistent results even when separately measuring DWI and MTT lesions. Propagation of measurement error was not demonstrated and the methods were statistically comparable. ^ Combined clinical and imaging prediction models were created to best describe the outcome of a stroke patient in a meaningful way. For comparison to these combined prediction models, we investigated imaging based prediction models based on pixel or voxel classification techniques using multiple MRI sequences. These predictive maps can be interpreted as the potential for disease growth given the conditions immediately after the patient has suffered a stroke. We measured the amount of abnormal tissue in these predictive maps to compare to the volume measurement results from the two combined models.