Detecting patterns of atrophy in cognitively impaired individuals using portable, low-field MRI

Low-field MRI (LF-MRI) is an emerging neuroimaging approach for evaluating patients with dementia, offering greater accessibility and lower cost, albeit with reduced image resolution. In this study, we deployed LF-MRI in an outpatient clinic and analyzed images with a multi-functional artificial intelligence (AI) algorithm (WMH-SynthSeg) to generate segmentation volumes of 16 brain regions. We validated the accuracy of the quantifications compared with conventional, high-field (HF) MRI in healthy volunteers and subsequently applied the algorithm to aged subjects with mild cognitive impairment (MCI) or dementia due to Alzheimer's disease (AD), and to similarly aged subjects with cognitive impairment and vascular comorbidities (VC). Agreement between HF- and LF-derived brain volumes was high across cohorts and brain regions, with the highest correlations in the cortex, white matter, lateral ventricles, third ventricle, caudate, and amygdala (all r > 0.80, p < 0.001). The MCI and AD cohorts showed regional atrophy relative to the VC cohort, including the cortex, hippocampus, amygdala, putamen, and nucleus accumbens (all p < 0.001) but not the caudate, ventral diencephalon, and fourth ventricle (p > 0.05). Taken together, LF-MRI paired with an AI segmentation algorithm can generate brain volumes comparable with those derived from conventional MRI, allowing for differentiation between VC and AD/MCI subgroups. Our findings demonstrate that LF-MRI could be used at the point-of-care for evaluation of patients with dementia of different etiologies.