Abstract 319: Computational Analysis Of Brain Imaging To Predict Post-arrest Brain Death

Abstract

Background: Grey-white matter ratio ( GWR) can estimate severity of cerebral edema after cardiac arrest and predict progression to brain death. Current approaches to calculating GWR have variable interrater reliability. We tested if a computational technique to quantify the distribution of whole-brain x-ray attenuation at hospital admission could predict ultimate progression to brain death. Methods: We identified in- and out-of-hospital post-arrest patients at our single center treated between 2018-2019. We extracted demographic data from our registry and Digital Imaging and Communication in Medicine (DICOM) files for the first post-arrest CT. We analyzed the 17th slice of each DICOM, corresponding to the level of the basal ganglia. We extracted the 512x512 array of Hounsfield units (HU). We excluded HU >40 or <10 to focus on brain tissue densities. We summarized each patient’s distribution of HU and calculated the variance as a proxy for GWR. We randomly divided our data into 75% training set and 25% test set, used univariate logistic regression to predict brain death, identified a highly specific cutoff to rule out progression to brain death, and calculated test set sensitivity and specificity at this cutoff. We performed all analyses using Python. Results: Of 413 included patients, 163 (39%) were female, median [IQR] age was 63 [52 - 73] years, and 39 (9%) progressed to brain death. CTs were obtained a median [IQR] of 4.47 [3.03 to 6.35] hours post-arrest. HU variation predicted brain death with an area under the receiver operating characteristic curve (AUROC) of 0.81 in the training set and 0.88 in the test set. The odds ratio (OR) of progression to brain death with decreasing HU variation was 0.85 (95% confidence interval [CI], 0.80 - 0.91; P < 0.001). A cutoff of 53.9 had a sensitivity of 36% (95% CI, 30.10% to 41.33%) at 97% specificity (95% CI, 89.91% to 100%) in the training set and a sensitivity of 45% (95% CI, 34.63% to 54.73%) at 100% specificity in the test set. Conclusion: Preserved HU variation excludes progression to brain death in comatose post-arrest patients with high specificity.