Deep learning reconstruction enhances 1.5T MR angiography beyond 3T in vascular visualization for Moyamoya disease.

purposeDeep learning reconstruction (DLR) is increasingly being applied to clinical magnetic resonance angiography (MRA); however, its evaluation across different magnetic field strengths in moyamoya disease is limited. This study assessed whether DLR affects visualization of stenotic arteries and collateral moyamoya vessels (MMVs). MATERIALS AND

methodsThirty-two patients (mean age, 40 years; male/female, 9/23) with suspected or confirmed moyamoya disease between 2015 and 2024 were retrospectively selected. All patients underwent time-of-flight MRA using 1.5T and 3T scanners within a 400-day interval. Differences in image quality and vascular visualization were assessed across four types of maximum intensity projection images from 1.5T and 3T MRA with and without DLR. The rankings of imaging quality and vascular depiction, Houkin classification scores, and visualization scores of the MMVs were compared using the Wilcoxon signed-rank test.

resultsWhen all four groups (1.5T and 3T MRA with/without DLR of 32 patients) were simultaneously compared using ranking scores, both overall image quality and visualization of MMVs were consistently rated higher for the DLR-enhanced images; notably, the 1.5T DLR-enhanced MRA achieved higher quality rankings than the 3T original MRA (p < 0.048). Houkin’s scores significantly decreased in DLR-enhanced MRA (p < 0.016) when compared at the same field strength, indicating less severe stenosis in DLR-enhanced images. MMVs visualization scores tended to shift toward higher grades after DLR, although this difference was not significant.

conclusionDLR significantly improved vascular visualization in moyamoya disease, with 1.5T DLR-enhanced MRA outperforming the original 3T MRA in terms of image quality and MMVs depiction.