A Platform for Automatic Extraction of Imaging Biomarkers in CT Scans from Ischemic and Hemorrhagic Stroke Patients

Quantitative analysis of brain imaging after stroke provides a window into disease course and severity, critical factors for pathophysiology and outcome studies. While millions of patients experience a stroke each year, there are barriers to systematically analyzing the high volume of imaging studies across centers. This study demonstrates the capabilities of a cloud-based imaging repository and computational platform for extracting multi-dimensional imaging biomarkers from CT scans of ischemic and hemorrhagic stroke patients. The Stroke NeuroImaging Phenotype Repository (SNIPR) is built on the XNAT platform and has archived serial imaging data from several large stroke cohorts. The platform supports Docker-containerized analysis for the evaluation of lesion and edema volumes. This study outlines the deployment of these comprehensive image analysis pipelines, encompassing steps from brain scan classification to the segmentation of ischemic and hemorrhage regions and subsequent quantification of edema biomarkers, including net water uptake (NWU) and the ratio of hemispheric CSF volumes. Analysis pipelines were executed on 8029 CT sessions across ischemic stroke, intracerebral hemorrhage, and subarachnoid hemorrhage cohorts, yielding successful biomarker extraction in 91% of cases, demonstrating the platform's efficiency. Biomarker profiles extracted from these large cohorts illustrated the plausibility of big data analyses in stroke using SNIPR. The incorporation of containerized analysis pipelines into a stroke imaging platform significantly enhances the capacity for multi-centric collaborative stroke research, enabling the processing of thousands of brain CT images for biomarker extraction.