Portable magnetic resonance imaging (pMRI) is a promising tool for the detection of white matter hyperintensities (WMH) as conventional MRI (cMRI) often poses logistical and financial burdens. Automated segmentation of pMRI may further facilitate detection and characterization of WMH. Here we assess the accuracy of an automated WMH segmentation platform in a cohort of patients with vascular risk factors.

Patients with at least one vascular risk factor were prospectively recruited at an academic medical center chest pain center between December 2021 to July 2024. In a cross-sectional design, participants underwent fluid-attenuated inversion recovery (FLAIR) acquisition on a 0.064 T MRI device (Hyperfine Inc) and medical history questionnaire. Each pMRI scan was processed using WMH-SynthSeg, a machine learning algorithm designed to process sequences of any resolution. Patient cMRIs obtained within 1 year of pMRI were also processed with WMH-SynthSeg as a benchmark to assess volume agreement (n=36). Subsequently, we evaluated the relationship between WMH volume and known risk factors including age, hypertension, hyperlipidemia, atrial fibrillation, and congestive heart failure using Pearson’s correlations and unpaired t-tests.

We found that pMRI and cMRI WMH volumes were strongly correlated (n=36) (r=0.87, p<0.001). To account for the time difference between pMRI and cMRI scans, we adjusted WMH volumes for an established rate of WMH progression, which did not impact the correlations. Among 161 total participants (62.2±12.4 years, 54% male, 26% black), those with atrial fibrillation exhibited greater WMH volumes (mean 12.2±8.8 mL vs 8.8±5.6 mL, p=0.009). Additionally, WMH volume was significantly correlated with age (r=0.52, p<0.001).

Automatic segmentation of WMH from pMRI scans showed significant agreement with conventional imaging. Additionally, greater age and atrial fibrillation were associated with higher WMH volume, reproducing previous findings from cMRI.