High-Resolution Late Gadolinium Enhancement Cardiac MRI: Feasibility Study

This feasibility study evaluated whether deep learning-based image denoising could enable high-spatial-resolution late gadolinium enhancement (LGE) cardiac MRI without increasing scan time or radiation dose, testing the approach in 120 patients referred for clinical cardiac MRI.

Deep learning denoising produced LGE images with higher signal-to-noise ratio and improved scar delineation compared with conventional spatial resolution LGE at the same acquisition time. Subendocardial scar patterns — critical for distinguishing ischemic from non-ischemic cardiomyopathy — were more sharply resolved in the denoised high-resolution images.

High-resolution LGE is clinically valuable but technically challenging: higher spatial resolution requires longer acquisition times or results in noisier images. Deep learning denoising offers a route to higher resolution without penalty — potentially enabling detection of small subendocardial scars that currently require invasive evaluation or are simply missed. This has implications for diagnosis of silent MI, microvascular disease, and fibrosis assessment in cardiomyopathy.

The study is a technical feasibility report, not a clinical validation. Outcomes data demonstrating improved diagnostic accuracy or changed clinical management are absent.

We rate the evidence limited. A small feasibility study demonstrating the technical viability of deep learning-enhanced high-resolution LGE MRI — promising methodology requiring larger validation.