Super-resolution deep learning reconstruction enhances visualization of cerebral aneurysms on magnetic resonance angiography

To evaluate the depiction of cerebral aneurysms on magnetic resonance angiography (MRA) reconstructed using super-resolution deep learning reconstruction (SR-DLR). We retrospectively reviewed the MRA images of 79 patients (49 with cerebral aneurysms and 30 without). The MRA images were subjected to SR-DLR and compared with the original images using both quantitative (signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) for the basilar artery (BA) and aneurysms; full width at half maximum (FWHM), edge rise distance (ERD), and edge rise slope (ERS) for the BA) and qualitative metrics (depiction of aneurysms and BA, image sharpness, noise, artifacts, and overall image quality performed by three radiologists). Statistical comparisons were performed using a paired t-test for quantitative metrics and the Wilcoxon signed-rank test for qualitative metrics. SR-DLR significantly improved SNR, CNR, and ERD compared with the original images. ERS values showed a trend toward improvement with SR-DLR, whereas FWHM showed no significant difference. Depiction of the vasculature, image sharpness, noise, and overall image quality were rated as significantly better for SR-DLR by all readers. Most readers rated that the depiction of aneurysms was improved with SR-DLR, particularly for aneurysms < 5 mm. The assessment of artifacts varied among readers. SR-DLR significantly improves the image quality of MRA and enhances the visualization of cerebral aneurysms, with greater improvements in the depiction of smaller aneurysms.