Clinical Validation of Accelerated Breath-held and Free Breathing Cine Cardiac MR Imaging Using Model-Based Deep Learning Reconstruction (SonicDL) in Children and Young Adult

Background: Cine bSSFP imaging using consecutive breath-holds (BH) for ventricular volumetry and myocardial mass assessment remains challenging, particularly in pediatric populations where shorter BH durations or free-breathing (FB) acquisitions are needed to accommodate limited breath-hold capacity.

Methods: In this HIPAA-compliant, IRB-approved retrospective study, cardiac MRI exams from April to July 2025 included standard-of-care BH cine imaging (ASSET [Array Spatial Sensitivity Encoding Technique] R=2) and accelerated Sonic DL cine imaging (R=7 and R=12) in ventricular short-axis orientation (Table 1). All acquisitions used retrospective cardiac gating and prospective arrhythmia rejection. One radiologist and two cardiologists independently scored image quality (IQ) on a five-point scale. Automated DL-based biventricular segmentation was performed for volumetric analysis. Comparative statistics were applied to IQ scores, biventricular volumes, and LV mass.

Results: Twenty-five patients (mean age 16.2 ± 1.8 years; BSA 1.73 ± 0.18 m²; HR 66.5 ± 10.0 bpm) underwent three BH cine acquisitions (ASSET R=2 [9-RR], Sonic DL R=7 [4-RR], Sonic DL R=12 [1-RR]) and one FB Sonic DL R=12 [1-RR] acquisition. All pairwise comparisons showed significant scan time reductions (p < 0.001), with mean durations decreasing from 9-RR BH (10.2 ± 3 min) to 4-RR BH (4.3 ± 0.8 min), 1-RR BH (2.3 ± 0.9 min), and 1-RR FB (1.3 ± 0.5 min). 4-RR BH reduced scan time by ~54%, while 1-RR BH and 1-RR FB achieved reductions of ~79% and ~86%, respectively. IQ ratings showed fair inter-rater agreement (Fleiss’ kappa = 0.2653), with score 4 most frequently assigned. Only one subject received scores < 3 by two raters for 9-RR BH and 4-RR BH. Mean IQ scores were highest for 4-RR BH (4.43 ± 0.59), significantly better than 9-RR BH (3.64 ± 0.52), 1-RR BH (3.85 ± 0.33), and 1-RR FB (3.58 ± 0.37) (p < 0.0001). Volumetric measurements (LVEDV, LVESV, RVEDV, RVESV, LV mass) showed high consistency across protocols (ANOVA p > 0.94), with select paired comparisons revealing subtle but statistically significant differences as shown in Table 2. All protocols underestimated stroke volumes by volumetry compared to QFlow for both LV (bias < 5ml/m²) and RV (bias < 5ml/m²).

Conclusion: Retrospectively gated cine bSSFP imaging with SonicDL acceleration factors of 7 and 12 yields biventricular volumetric and functional indices comparable to standard ASSET R=2 acquisitions, while significantly reducing scan time and maintaining diagnostic image quality with nominally equivalent spatial and temporal resolution.

Acquisition Parameters

Parameter	9-RR BH	4-RR BH	1-RR BH	1-RR FB
Acq Voxel Size (mm3)	1.7×1.7×7-8	1.7×1.3×7-8	1.7×1.7×7-8	1.7×1.7×7-8
Slices	14-17	14-17	14-17	14-17
FOV (mm2)	340×340	340×340	340×340	340×340
Acquisition Matrix	200×200	200×256	200×160	200×160
Phase Undersampling	0.9	0.9	0.9	0.9
Acceleration (R)	ASSET 2	Sonic DL 7	Sonic DL 12	Sonic DL 12
TR (ms) / TE (ms)	3.65-3.8 / 1.6-1.7	3.3-3.5 / 1.2-1.3	3.2-3.8 / 1.1-1.5	3.2-3.9 / 1.1-1.5
FA(°)	55	55	55	55
Views per segment	10	10	12	12
Acq Temp Res (ms)	37-38	33-35	38-46	38-46
Rec Cardiac Phases	30	30	30	30
Breah-hold	9-12 s / 1 slice	11-16 s / 3 slices	8-11 s / 5 slices	N/A
Total Scan Duration (min)	5-13	3-6	1.5-3	1-2

Comparison of Cardiac Functional Parameters Across Protocols

Comparison	LVEDV ml/m2	LVESV ml/m2	RVEDV ml/m2	RVESV ml/m2	LV Mass mg/m2	LVEF %	RVEF %
9RR vs 4RR	1.07 ± 3.38 (p=0.1263)	0.84 ± 3.26 (p=0.2084)	2.51 ± 4.78 (p=0.0148*)	1.71 ± 3.12 (p=0.0112*)	-1.63 ± 3.47 (p=0.0276*)	-0.32 ± 2.29 (p=0.4921)	-0.47 ± 1.84 (p=0.2115)
9RR vs 1RRBH	1.08 ± 3.59 (p=0.1464)	0.10 ± 3.35 (p=0.8821)	2.66 ± 5.25 (p=0.0183*)	1.12 ± 3.20 (p=0.0924)	-0.93 ± 3.20 (p=0.1597)	0.31 ± 2.33 (p=0.5103)	0.05 ± 1.93 (p=0.8991)
9RR vs 1RRFB	1.95 ± 3.18 (p=0.0054*)	-0.43 ± 3.13 (p=0.4964)	1.22 ± 5.26 (p=0.2581)	-0.42 ± 5.09 (p=0.6814)	-0.99 ± 3.33 (p=0.1499)	1.11 ± 2.46 (p=0.0335*)	0.61 ± 2.91 (p=0.3081)