Abstract 3319: Prediction Of Lesion Expansion In Stroke Patients Using Acute MRI

Abstract

Background: We investigated whether MRI-based algorithms combining acute DWI & PWI can be used to identify patients likely to experience lesion expansion. Methods: We analyzed MRI from 181 unilateral stroke patients who underwent DWI & PWI ≤ 12h from last seen well and had follow-up imaging (F/u) ≥4 days with lesions ≥ 1 cm 3 . Apparent diffusion coefficient, T2, DWI, CBF, CBV, MTT and Tmax (time of peak of deconvolved residue function) maps were co-registered, normalized and used as covariates in a model that produced infarction risk maps. Coefficients were calculated from 111 non-lysed patients and applied to 70 patients who received mechanical or drug lysis. Area under generated receiver operating characteristic curves (AUC) were calculated. Regional analyses were performed comparing mean risk values in Core (acute DWI), Growth (F/u - Core), Reverse (Core - F/u) and Normal (ipsilesional hemisphere - F/u) regions. Predicted lesion volumes (PLV) using a 50% risk threshold and post-hoc artifact removal for classifying infarcted tissue were correlated with the measured lesion volumes (MLV) on F/u. Values are median [IQR] or mean±SD. Results: The thrombolysis-treated group differed significantly from the non-lysis group in admission NIH Stroke Scale scores (12 [9-16] vs 6 [3-13]; P<0.001), time-to-MRI (4.8±1.9 h vs 5.9±2.8 h; P=0.005), acute DWI lesion volumes (21 [7-56] vs 11 [3-36] cm 3 ; P=0.02) and F/u lesion volumes (39 [12-72] vs 17 [6-56] cm 3 ; P=0.02). No significant difference was found between age (67±15 vs 65±17 years old), male sex (59% vs 67%) and time to F/u (15 [6-47] vs 10 [6-46] days). Significant differences in predicted risk between the 4 regions were found for both lysed (Core: 0.72±0.15; Growth: 0.41±0.13; Reverse: 0.64±0.16; Normal: 0.25±0.06; P<0.0001) and non-lysed groups (Core: 0.72±0. 15; Growth:0.42±0.12; Reverse: 0.64±0.15; Normal: 0.22±0.05; P<0.0001). AUCs of the prediction for the non-lysed group were higher than for the lysed group (0.86±0.09 vs 0.81±0.11; P<0.001). Correlations between PLV and MLV were significant for both lysed (R=0.57, P<0.001) and non-lysed groups (R=0.85, P<0.001). Mismatch between PLV and Core volumes (i.e. predicted lesion growth) was significantly correlated with actual lesion growth for the non-lysed group (R=0.49, P<0.001) but not for the lysed group (R=0.04, P=0.72). Conclusion: Mismatch in PLV and core can be used to identify patients most likely to experience lesion expansion if not given reperfusion therapy. The performance of the models was reduced in thrombolysed patients, which we propose is due to salvage of tissue that would have otherwise infarcted. MRI-based algorithms may identify patients who are not candidates for thrombolysis, yet have tissue to salvage. This may be useful as imaging selection criteria for future investigational trials of reperfusion therapy in extended time windows or for patients with unclear onsets.