Apparent diffusion coefficient threshold for delineation of ischemic core
- Conflicts of interest: GW Albers and R Bammer are equity shareholders in iSchemaView.
- Funding: The funding for this study was provided by National Institutes of Health (NIH) grants R01 NS39325 (GWA), R01 EB002711 (RB), K24 NS044848 (GWA), and K23 NS051372 (MGL). tPA was supplied at no charge by Genentech (United States and Canada sites) and Boehringer Ingelheim (Belgium site). Neither Genentech, Boehringer Ingelheim nor the NIH played a role in the design or conduct of the study, data collection, management, analysis, or interpretation, or preparation, review or approval of the manuscript.
MRI-based selection of patients for acute stroke interventions requires rapid accurate estimation of the infarct core on diffusion-weighted MRI. Typically used manual methods to delineate restricted diffusion lesions are subjective and time consuming. These limitations would be overcome by a fully automated method that can rapidly and objectively delineate the ischemic core. An automated method would require predefined criteria to identify the ischemic core.
The aim of this study is to determine apparent diffusion coefficient-based criteria that can be implemented in a fully automated software solution for identification of the ischemic core.
Imaging data from patients enrolled in the Diffusion and Perfusion Imaging Evaluation for Understanding Stroke Evolution (DEFUSE) study who had early revascularization following intravenous thrombolysis were included. The patients’ baseline restricted diffusion and 30-day T2-weighted fluid-attenuated inversion recovery lesions were manually delineated after coregistration. Parts of the restricted diffusion lesion that corresponded with 30-day infarct were considered ischemic core, whereas parts that corresponded with normal brain parenchyma at 30 days were considered noncore. The optimal apparent diffusion coefficient threshold to discriminate core from noncore voxels was determined by voxel-based receiver operating characteristics analysis using the Youden index.
51 045 diffusion positive voxels from 14 patients who met eligibility criteria were analyzed. The mean DWI lesion volume was 24 (± 23) ml. Of this, 18 (± 22) ml was ischemic core and 3 (± 5) ml was noncore. The remainder corresponded to preexisting gliosis, cerebrospinal fluid, or was lost to postinfarct atrophy. The apparent diffusion coefficient of core was lower than that of noncore voxels (P < 0·0001). The optimal threshold for identification of ischemic core was an apparent diffusion coefficient ≤620 × 10−6 mm2/s (sensitivity 69% and specificity 78%).
Our data suggest that the ischemic core can be identified with an absolute apparent diffusion coefficient threshold. This threshold can be implemented in image analysis software for fully automated segmentation of the ischemic core.