Abstract W P42: Automated Perfusion Computer Axial Tomography Predicts Acute Stroke Deficits

Abstract

Background: Perfusion Computerized Axial Tomography (pCT) is a modality gaining popularity for acute stroke management decisions. We investigated the role of automated pCT in prediction of deficits measured by NIH Stroke Scale (NIHSS). Method: Acute stroke patients underwent pCT and NIHSS. Digital Imaging and Communications in Medicine ( DICOM ) images were assessed for maximum intensity (MI) and cerebral blood flow (CBF). MI scans (where there is little acute abnormality) were used to transform each scan into standard space. The CBF images were the continuous measure for abnormality. The regional CBF was estimated for 150 regions of interest. Statistical analyses used 4000 permutations to correct for multiple comparisons. Additionally support vector machines (SVM) were used to assess the ability of machine learning to classify the deficit severity. Results: PCT scans for 83 patients where analyzed (mean age ± SD=64.2 ±14.8, 63% males). Traditional analyses revealed (p < 0.01) that reduced blood flow in the large regions of the right hemisphere were associated with left motor impairment, while regional reductions in the left hemisphere were associated with motor impairment on the right side as well as language impairments. We were accurately able to predict diagnosis based purely on perfusion. Reliable detection (all p < 0.001) of aphasia (74.7% accuracy or ACC, 86.1% specificity or SP, 56.8% sensitivity or SN), left upper motor (69.3% ACC, 87.9% SP, 42% SN), right upper motor (74.6% ACC, 84.4% SP, 57.3% SN), left lower (68.9% ACC, 87.5% SP, 43.7% SN) and right lower (71.2% ACC, 87.5% SP, 51.1% SN) impairments were accomplished. Conclusion: We describe a reliable automated pCT method to predict neurological deficit in acute stroke patients. The methodology may be useful in patients in who neurological assessment is not possible (e.g. unconscious or intubated patients) or assessment may be at risk of observer bias (e.g. in clinical trials).