Segmentation of Leukoaraiosis on Noncontrast Head CT Using CT-MRI Paired Data Without Human Annotation

Evaluating leukoaraiosis (LA) on CT is challenging due to its low contrast and similarity to parenchymal gliosis. We developed and validated a deep learning algorithm for LA segmentation using CT-MRIFLAIR paired data from a multicenter Korean registry and tested it in a US dataset.

We constructed a large multicenter dataset of CT-FLAIR MRI pairs. Using validated software to segment white matter hyperintensity (WMH) on FLAIR, we generated pseudo-ground-truth LA labels on CT through deformable image registration. A 2D nnU-Net architecture was trained solely on CT images and registered masks. Performance was evaluated using the Dice similarity coefficient (DSC), concordance correlation coefficient (CCC), and Pearson correlation across internal, external, and US validation cohorts. Clinical associations of predicted LA volume with age, risk factors, and poststroke outcomes were also analyzed.

The external test set yielded a DSC of 0.527, with high volume correlations against registered LA (r = 0.953) and WMH (r = 0.951). In the external testing and US datasets, predicted LA volumes correlated with Fazekas grade (r = 0.832-0.891) and the correlations were consistent across CT vendors and infarct volumes. In an independent clinical cohort (n = 867), LA volume was independently associated with age, vascular risk factors, and 3-month functional outcomes.

Our deep learning algorithm offers a reproducible method for LA segmentation on CT, bridging the gap between CT and MRI assessments in patients with ischemic stroke.