Abstract Body:
Introduction:
Thrombectomy is an effective treatment for acute ischemic stroke patients with a large vessel occlusion, but even with successful treatment a large portion of patients have poor outcomes. New imaging metrics that predict outcomes could help to identify patients who might benefit from neuroprotective therapies. We determined if deep learning based CT (DLCT) volume and density measures are associated with patient outcomes.
Methods:
Patients from the prospective “CTP to predict Response to Recanalization in Ischemic Stroke Project 2” cohort study with baseline CT and CTP were included. Volume, density, and relative density compared to the contralateral hemisphere were measured using lesion segmentations from a validated CT-based deep learning segmentation model (DLCT). Additional clinical and CTP features were considered as independent variables for association with the modified Rankin Scale (mRS) 90 days after endovascular treatment. Variable selection was performed with stepwise ordinal regression (shift analysis) with combined forward/backward selection optimized using the Akaike Information Criterion (AIC). Log-likelihood (LL) and AIC were reported to compare models considering DLCT and CTP with DLCT or CTP variables besides clinical variables. Adjusted common odds ratios for favorable (reversed) mRS with 95% confidence intervals (aOR[95%CI]) were reported per variable for the optimized DLCT and CTP model.
Results:
We included 156/113 men/women, the median age was 71 (IQR:60-80) years. The DLCT-CTP model (LL=441, AIC=927) and DLCT (LL=443, AIC=927) had a similar performance that was higher than the CTP model performance alone (LL=452, AIC=938). DLCT measures association with mRS were hypodense (<26HU) lesion volume (aOR/10mL:0.69 [0.49; 0.99]), net water uptake (aOR/1%:0.93 [0.89; 0.98]), and density standard deviation in the lesion (aOR/HU:0.69 [0.56; 0.87]). CTP ischemic core volume was not significantly associated with favorable mRS (aOR/10mL:0.90 [0.81;1.01], p=0.08). Other independent variables with significant associations were female sex (aOR:0.60 [0.38; 0.96]), NIH stroke scale (aOR/point:0.90 [0.87; 0.94]), preexisting mRS (aOR/point:0.58 [0.47;0.71]), current smoking (aOR:0.45 [0.25;0.84]), and thrombolytic treatment (aOR:2.36 [1.42;3.91]).
Conclusion:
DLCT volume and density measures before interhospital transfer were associated with 90-day mRS and could be considered in addition or as a substitute for CTP as a prognostic marker.