377 Risk Stratification Using Cortical Thickness for Prediction of Functional Outcomes After Mechanical Thrombectomy

INTRODUCTION: Mechanical thrombectomy(MT) is standard of care for the 30% of acute ischemic strokes(AIS) caused by large vessel occlusion(LVO). Structural brain reserves may play an important role in recovery after stroke, which may explain the large variability in recovery amongst patients despite successful reperfusion. This has led to investigation for an effective marker for potential brain reserve in order to guide prognostication. We explore a large cohort of patients using commercially available software to transform standard clinically-acquired MRIs into quantitative data for patients with LVO who underwent successful revascularization. METHODS: Retrospective analysis was performed on a prospectively maintained single-center thrombectomy registry. Patients that underwent MT for LVO with TICI 2B or greater, indicating significant revascularization, were included. Clinically obtained T1-weighted images were transformed using the convolutional neural network SynthSR. FreeSurfer was used to quantify baseline cortical thickness. Ordinal logistic regression was used to adjust for baseline mRS and relevant covariates. RESULTS: FreeSurfer segmentation could be performed in 213/223(95.6%) patients. Prior to FDR correction with adjustment for sex, age, and NIHSS, the Desikan-Killiany model demonstrated correlation between the precuneus(p=.002), insula(p=.02), and superior temporal gyrus(p=.05) with mRs>2 at 90 days and the Destrieux model showed significance in the postcentral sulcus(p=.002), circular sulcus(p=.02), parieto-occipital sulcus(p=.02), pericallosal sulcus(p=.02), and superior temporal gyrus(p=.04). CONCLUSIONS: Increased mean cortical thickness in key areas, especially the superior temporal gyrus, on clinically-obtained MRIs correlated with functional outcomes at 90-days following MT with successful reperfusion. Cortical thickness may be a radiographic biomarker of recovery after stroke and highlights how neural networks might be leveraged to utilize routinely-acquired clinical MRIs to perform research-level quantitative investigation in clinical trials for AIS.