Purpose or Case Report: To assess the image quality and clinical performance of a novel accelerated, free-breathing 2D cine cardiac MR sequence with DCNN reconstruction in comparison to conventional breath-held 2D cine balanced steady-state free precession (bSSFP).
Methods & Materials: 17 consecutive children (mean age: 14.2 yr, range: 9.7-22.0 yr, 10 male) were prospectively recruited for both short-axis cine acquisition strategies in one non-sedated 1.5T (Signa Artist, GE Healthcare) exam for cardiomyopathy or congenital heart disease. The respiratory-triggered 12-fold accelerated sequence was undersampled to acquire one slice over 1 RR interval. The reconstruction network inputs are zero-filled images generated using coil sensitivity maps that are extracted from a calibration region. Images are then reconstructed with a network that implements iterations of separable 3D convolutions and data consistency steps, completed within 1 minute. For each case, 2 blinded pediatric CV radiologists independently scored image quality of the DCNN and bSSFP recons on a 5-point scale (1-worst, 5-best) and manually segmented the LV/RV endocardial contours. Scan times (approximated as time between acquisition start and next sequence start) and image quality scores were compared using the Wilcoxon rank-sum test. Intra- and interobserver agreement in ventricular volumes/function for each recon type were assessed with the intraclass correlation coefficient (ICC).
Results: Accelerated scans were significantly faster than conventional bSSFP (mean time: 2.3 +/- 1.5 min vs. 11.3 +/- 2.9 min, p<0.001) with only minimally decreased mean image quality scores (3.6 +/- 0.6 vs. 4.1 +/- 0.5, p=0.001); scores for all recons were >=3, indicating at least diagnostic quality. For both raters, there was excellent agreement between volume/function measurements derived from conventional and DCNN images; ICCs for LVEDV, LVESV, LVSV, LVEF, RVEDV, RVESV, RVSV, and RVEF were 0.959-0.960, 0.953-0.967, 0.925-0.952, 0.839-0.977, 0.965-0.970, 0.945-0.958, 0.945-0.979, 0.894-0.958, respectively. There was also excellent interobserver measurement agreement in these parameters both for DCNN and conventional images, with ICCs ranging from 0.879-0.990 and 0.887-0.992, respectively.
Conclusions: The accelerated, free-breathing DCNN technique was much faster than bSSFP with comparable image quality and measurement agreement for key LV/RV parameters. This approach may thus allow substantially shorter cardiac MRI with a primary goal of ventricular volume and function analysis.