(CCSP097) THE CLINICAL UTILITY OF A RAPID 2D BALANCED STEADY STATE FREE PRECESSION SEQUENCE WITH DEEP LEARNING RECONSTRUCTION

Background: Cardiac magnetic resonance (CMR) cine imaging is considered a gold-standard modality for measuring left- and right-ventricular (LV and RV) volumetry, but is limited by long acquisition times (1,2). This study evaluated the clinical utility of an accelerated 2D cine sequence with deep learning (DL) reconstruction (3,4) to decrease acquisition time without affecting quantitative volumetry nor image quality.

METHODS AND RESULTS: 98 patients and 15 healthy volunteers (72 [67%] men, mean age 53.3 ± 15.3 years) were prospectively recruited and scanned on a 3 Tesla system (GE Healthcare, Milwaukee, USA) using both a standard (ASSET) and accelerated (Sonic DL) cine method. Images were acquired in 2-chamber, 3-chamber, 4-chamber views, and as a short axis (SAX) stack through both ventricles (11-12 slices). Two experienced clinical readers rated images based on diagnostic confidence using a 4-point ordinal scale where: 1 represented no diagnostic confidence (non-interpretable), 2 represented low diagnostic confidence (poor image quality, many artifacts), 3 represented medium diagnostic confidence (good overall image quality with one or two views with poorer IQ), and 4 represented high diagnostic confidence (very high image quality). LV and RV volumetry and left-ventricular mass (LVM) were compared using Bland-Altman plots and paired t-tests. Comparisons in image quality were measured using a Wilcoxon-signed-rank test.

Sonic DL significantly reduced the acquisition time of cine images when compared to ASSET (p < 210-16). Diagnostic image quality ratings were similar between methods (Table 1). LV- end-systolic volume (ESV), ejection fraction (EF), and RV end-diastolic volume (EDV), RVESV and RVEF were statistically similar between cine methods (p > 0.05). However, Sonic DL under-estimated LVEDV (Sonic DL: 71.95  16 mL/m2, ASSET: 73.32  17 mL/m2, p = 0.0437) and over-estimated LVM (Sonic DL: 66.29  17 g/m2, ASSET: 62.77  17 g/m2, p = 5.6610-5) compared to ASSET (Table 1). The average difference between methods for each parameter were 1.37  5.2 mL/m2, -0.13  4.1 mL/m2, 1.06  4.4%, -3.36  5.9 g/m2, 1.24  6 mL/m2, -0.04  3.6 mL/m2, and 1.37  8.02% for LVEDV, LVESV, LVEF, LVM, RVEDV, RVESV, and RVEF, respectively (Table 1, Figure 1).

Conclusion: This study found that an accelerated cine method with DL reconstruction can reduce acquisition time without affecting volumetry nor image quality. This work contributes to the growing body of evidence showing utility for accelerated acquisition methods with DL reconstruction for clinical CMR, nudging CMR towards the direction of becoming a more accessible diagnostic tool.