Abstract 2697: Fully-automated Identification of Acute Stroke Lesion Volumes with CT Perfusion
Purpose: Mismatch between volumes of infarct core and critically hypoperfused tissue (CHT) may be used to identify acute stroke patients who could benefit from reperfusion therapies. We present a fully-automated, operator-free approach for identifying the core and CHT lesion volumes with CT perfusion (CTP).
Methods: 31 scans of 25 acute stroke patients who underwent CTP followed by MRI (range: 23-120 min) were analyzed. CTP was obtained as a one or two 2cm slabs. MRI included DWI and PWI. Reference stroke lesion metrics were MRI-based: core via DWI (ADC<615x10-6mm2/s), and CHT via PWI (Tmax>6s). CTP and PWI scans were processed with an automated image analysis program (RAPID) with delay-independent deconvolution. MRI maps were coregistered to CTP. Contralaterally-relative CBVCT and CBFCT (crCBFCT, crCBFCT) maps were computed by putting into ratio the original and corresponding laterally-mirrored and smoothed rCBVCT and rCBFCT maps (obtained by vertical flip and coregistration of the anatomic images, see Fig). Stroke core in CTP was delineated by thresholding crCBVCT and crCBFCT, and CHT by thresholding TmaxCT. Optimal thresholds were obtained by ROC analysis and minimization of lesion volume differences between CT and MRI.
Results: For identification of stroke core in CTP, crCBFCT performed better than crCBVCT. Optimal threshold was crCBFCT < 0.30 with sensitivity 60% (CI95% 57-63%) and specificity 88%, (CI95% 87-89%); median volume difference between CBFCT and DWI lesions was 0 ml (IQR: -6ml to 6 ml); correlation of volumes was r2=0.72 (p<0.0001). For identification of CHT, reference MRI lesions (TmaxMR >6s) were most accurately identified by TmaxCT >6s with sensitivity 72% (CI95% 70-74%), specificity 97% (CI95% 96-97%); median volume difference between TmaxCT and TmaxMR was -3ml (IQR: -10ml to 0ml); correlation of CHT volumes r2=0.89 (p<0.0001).
Conclusions: The processing methods and CTP thresholds presented in this study show a great promise for fully-automated outlining of stroke lesions using CTP. Such a technique could be of great value for CTP-based patient selection in clinical trials and clinical practice.
- © 2012 by American Heart Association, Inc.