Abstract WMP104: Fully Automated Segmentation Algorithm for Volumetric Analysis of Perihematomal Edema After Spontaneous Intracerebral Hemorrhage
Background: Perihematomal edema (PHE) is a promising marker of secondary injury in patients with spontaneous intracerebral hemorrhage (ICH). It can be challenging to accurately and rapidly quantify. The aims of this study are to derive and internally validate a fully automated segmentation algorithm for volumetric PHE analysis. Methods: Inpatient CT scans of 400 consecutive adults with spontaneous supratentorial ICH enrolled in the Intracerebral Hemorrhage Outcomes Project (2009-2018) were separated into training (n=360) and test (n=40) datasets. A fully automated algorithm was derived from manual segmentations in the training dataset using convolutional neural networks and its performance was compared to manual and semi-automated segmentation methods in the test dataset. Results: The mean volumetric Dice similarity coefficients for the fully automated algorithm were 0.838±0.294 and 0.843±0.293 with manual and semi-automated segmentations as reference standards, respectively. PHE volumes derived from fully automated vs. manual (R 2 =0.959;p<0.001), fully automated vs. semi-automated (R 2 =0.960;p<0.001) and semi-automated vs. manual (R 2 =0.961; p<0.001) methods had strong between group correlations. The fully automated algorithm (mean 18.0±1.8 seconds/scan) quantified PHE volumes at a significantly faster rate than manual (mean 316.4±168.8 seconds/scan; p<0.001) and semi-automated (mean 480.5±295.3 seconds/scan; p<0.001) methods. Conclusions: The fully automated algorithm accurately quantified PHE from CT scans of supratentorial ICH patients with high fidelity and greater efficiency compared with manual and semi-automated segmentation methods.