Biomarker-Aware Super-Resolution for 4D Flow MRI in a Synthetic Pulmonary-Artery Benchmark

A synthetic pulmonary-artery benchmark was used to determine if biomarker-aware super-resolution enhances hemodynamic endpoint accuracy in 4D flow MRI. A 3D residual channel attention network (RCAN) was trained on 22 flow simulations with spatial degradation. The benchmark compared full biomarker-aware training versus a control that incorporated divergence and temporal regularization.

Biomarker-aware training led to a reduction in point-intensity (PI) error compared to the control, with a difference of 4.35 percentage points (95% confidence interval -5.58, -3.09; p = 0.0015). When biomarker supervision was combined with regularization, PI error was further reduced, with a difference of 2.235 percentage points (95% confidence interval -3.514, -0.849).

However, biomarker supervision did not result in improved reconstruction metrics such as peak signal-to-noise ratio (PSNR) or structural similarity (SSIM). Divergence regularization alone slightly improved PI without reducing divergence root mean square (DivRMS). Temporal-difference regularization, however, increased PI error by 1.562 percentage points (95% confidence interval 0.594, 2.439).

The results suggest that within this specific synthetic and fixed-case RCAN benchmark, biomarker supervision improves PI under both regularization conditions, but it does not necessarily lead to uniformly better reconstruction. The findings highlight the importance of in vivo validation to confirm these configuration-specific results.