Abstract W P44: Predicting Omni-directional Lesion Growth in Acute Stroke using Multimodal Intensity Profiles
Introduction: Mechanisms of lesion growth in acute stroke remain poorly characterized. Favorable imaging signatures related to regional tissue status may be disclosed with diffusion-weighted (DWI) and perfusion-weighted (PWI) imaging. Certain values of individual parameters, such as prolonged and decreased CBF, are associated with poor tissue recovery, yet scant data are available regarding the directionality of lesion growth and how different perfusion imaging parameters may be combined to best characterize lesion growth. We developed a probabilistic model that exploits DWI and multi-parametric PWI to predict likelihood of lesion growth in every 3D direction.
Hypothesis: We test the hypothesis that combined intensity profiles of PWI features predict the likelihood of lesion growth, in every direction.
Methods: Retrospective analysis of DWI and PWI acquired within 24 hours of symptom onset with FLAIR sequences acquired four days later. DWI and PWI were co-registered and the lesions were manually delineated on the baseline DWI and follow-up FLAIR. Intensity profiles of perfusion parameters (including CBV, CBF, MTT, TTP, Tmax) were extracted along discrete spherical coordinates (every 5 degrees). A nonlinear regression model was used to capture the relationship between the intensity profile along a direction and the amount of growth in that direction. A cross-validation was performed to evaluate the accuracy of the model in predicting the lesion growth in every direction at day 4.
Results: A total of 49 patients were included in the analysis. Mean age was 68.7 (35-91). Median baseline NIHSS was 16 (2-31) and median mRS at discharge was 5 (1-6). Lowest prediction error (62 cm2 IQR [26 86]) in terms of average lesion surface and final directional growth error 7.91 mm IQR [5.2 10.3] was obtained by combining cBV, cBF, TTP, TMAX intensity profiles into a single input vector.
Conclusions: For the first time, a direction-specific model of infarct growth has been developed. It provides quantitative insights about the likelihood of lesion growth surrounding a stroke. This prediction is not only based on closeness to the infarct core or the presence of penumbra but relies also on the complex dependencies between joint evidence found in multiple perfusion parameters.
Author Disclosures: F. Scalzo: None. W. Chowdhury: None. D.S. Liebeskind: Research Grant; Significant; NIH-NINDS. Consultant/Advisory Board; Modest; Stryker, Covidien.
- © 2015 by American Heart Association, Inc.