Accelerated Cardiac MRI Using Deep Learning Cine Imaging: Validation in Pediatric Patients with and without Congenital Heart Disease

Thursday, February 5, 2026

10:15 AM - 10:22 AM

Location: Galapagos / Aruba / Ilhabela (2nd Lower Level)

Presenting Author(s)

Jonathan Kochav, MD, MSc

Assistant professor of Medicine
Columbia University
Ridgewwod, New Jersey, United States

Primary Author(s)

Jonathan Kochav, MD, MSc

Assistant professor of Medicine
Columbia University
Ridgewwod, New Jersey, United States

Co-Author(s)

JM

Junjie Ma, PhD

Cardiac MRI Scientist
GE HealthCare
Jersey City, New Jersey, United States
SJ

Sachin Jambawalikar, PhD

Chief Medical Physicist
Columbia University, United States
MF

Maggie Fung, MSc

GE ASL Team Leader & Senior Scientist, MR Application & Workflow Team
GE Healthcare, United States
PQ

Patrick Quarterman

Manager, MR Clinical Development
GE Healthcare, United States
AS

Amee Shah, MD

Associate Professor
Columbia University, United States
KF

Kanwal Farooqi, MD

Associate Professor of Pediatrics
Columbia University Irving Medical Center
New York, New York, United States
AM

Amara Majeed, MD

Assistant Professor
Columbia University, United States
SJ

Simone Jhaveri, MD

Pediatric cardiologist
Columbia University, United States
AC

Alexandra Channing, MD

Assistant Professor of Pediatrics
Columbia University
NEW YORK, New York, United States
MD

Michael P. DiLorenzo, MD, MSc

Associate Professor of Pediatrics
Columbia University/Morgan Stanley Children's Hospital
New York, New York, United States

Background: Balanced steady-state free precession (bSSFP) cine is the gold standard imaging sequence for assessment of ventricular volumes, but requires lengthy breath holds. This may be challenging in pediatric patients unable to comply or with underlying arrhythmias, and contributes to long scan time. A novel deep learning algorithm that accelerates image acquisition using a variable density undersampled k-t random sampling scheme (Sonic DL Cine) has previously been validated in adults¹, but data in pediatric patients with and without congenital heart disease (CHD) is limited.

Methods: Pediatric patients (age < 18 years) with biventricular circulation were prospectively recruited. All patients underwent a standard breath-held bSSFP cine short axis stack (Traditional SSFP), as well as breath-held short axis Sonic DL Cine using sampling from 3 heart beats (3RR), 1 heart beat (1RR), and a free-breathing acquisition from 1 heart beat (1RR-FB), acquired on 1.5T SIGNA Explorer (GE HealthCare). Representative images are shown in Figure 1. Ventricular measurements were performed using Circle cvi42, including left and right ventricular end-diastolic (LVEDV, RVEDV) and end-systolic (LVESV, RVESV) volumes, LV and RV ejection fraction (LVEF, RVEF), as well as LV mass. Major trabeculations and papillary muscles were considered part of ventricular mass, and as part of ventricular volume in separate analyses.

Results: 50 patients were recruited (16.1±1.3 years, 69% male). Twenty-eight (56%) had known CHD, 10 (20%) had a known cardiomyopathy, and 12 (24%) were performed to rule out CHD or cardiomyopathy. Short axis cine stack imaging time was substantially reduced using Sonic DL Cine sequences (Traditional SSFP 8.6±1.4 min; 3RR 3.0±3.0 min; 1RR 0.6±1.6 min; 1RR-FB 1.0±0.37 min; p< 0.001 for all Sonic DL Cine sequences versus traditional SSFP, Figure 2). When papillary muscles / trabeculations were considered ventricular mass (Table 1), LV and RV volumes measured lower, and LV mass measured higher on the Sonic DL Cine sequences compared to traditional SSFP, although mean differences were small and not clinically significant (< 5 ml mean differences for LV diastolic volumes, < 10 ml mean differences for RV diastolic volumes < 5 g mean differences for LV mass). Mean differences in LVEF and RVEF were also small (< 2% and < 3%, respectively). When papillary muscles / trabeculations were considered ventricular volume, RV volumetrics were not significantly different between traditional SSFP and Sonic DL Cine sequences.

Conclusion: This prospective analysis found that an accelerated 2D cine sequence with DL reconstruction can reduce CMR cine acquisition time by >50%, without significantly affecting volumetric analysis in a pediatric population with high prevalence of CHD. Differences between Sonic DL Cine and traditional SSFP were greater when papillary muscles were included in ventricular mass. These findings support use of Sonic DL Cine in younger patients with and without CHD.

Ventricular Volumetrics

Papillary Muscles / Trabeculations Included in Ventricular Mass
	Traditional SSFP	Sonic DL 3RR	p-value	Sonic DL 1RR	p-value	Sonic DL 1RR FB	p-value
LVEDV (ml)	147.2 ± 35.2	142.3 ± 33.9	< 0.01	141.5 ± 33.4	< 0.001	145.3 ± 34.8	0.18
LVEDV index (ml/m2)	86.0 ± 19.5	83.3 ± 19.8	< 0.01	83.1 ± 19.8	< 0.001	85.1 ± 19.6	0.19
LVESV (ml)	64.0 ± 19.9	60.9 ± 20.2	< 0.01	64.7 ± 20.4	0.64	63.8 ± 20.4	0.72
LVESV index (ml/m2)	37.8 ± 12.0	35.8 ± 12.7	< 0.01	38.1 ± 12.6	0.56	37.7 ± 12.4	0.85
LVEF (%)	56.9 ± 6.8	57.8 ± 7.1	0.02	54.9 ± 6.4	< 0.001	56.2 ± 8.2	0.62
LV Mass (g)	106.5 ± 35.4	111.5 ± 38.7	< 0.001	108.4 ± 37.6	0.02	109.6 ± 38.2	0.01
LV Mass index (g/m2)	62.7 ± 26.5	65.8 ± 29.1	< 0.001	64.7 ± 27.4	< 0.01	64.7 ± 27.6	0.01
RVEDV (ml)	185.1 ± 49.2	178.1 ± 50.1	< 0.01	171.4 ± 49.1	< 0.001	179.0 ± 50.2	< 0.01
RVEDV index	109.4 ± 32.6	105.2 ± 32.6	< 0.01	101.2 ± 31.7	< 0.001	105.6 ± 32.2	< 0.001
RVESV (ml)	93.9 ± 30.9	88.9 ± 29.2	< 0.001	87.2 ± 29.6	< 0.001	88.6 ± 30.2	< 0.001
RVESV index (ml/m2)	55.4 ± 19.6	52.3 ± 17.9	< 0.001	51.4 ± 17.8	< 0.001	52.1 ± 17.6	< 0.001
RVEF (%)	50.0 ± 8.0	47.7 ± 9.9	< 0.01	49.6 ± 8.8	0.65	51.1 ± 8.5	0.14
Papillary Muscles / Trabeculations Included in Ventricular Volume
	Traditional SSFP	Sonic DL 3RR	p-value	Sonic DL 1RR	p-value	Sonic DL 1RR FB	p-value
LVEDV (ml)	152.4 ± 36.1	147.9 ± 34.9	0.001	146.0 ± 34.6	< 0.001	149.3 ± 36.0	0.01
LVEDV index (ml/m2)	89.0 ± 19.8	86.6 ± 20.1	< 0.01	85.7 ± 20.2	< 0.001	87.4 ± 20.2	0.01
LVESV (ml)	68.3 ± 20.4	65.6 ± 20.7	< 0.01	69.2 ± 21.6	0.41	70.2 ± 21.0	0.04
LVESV index (ml/m2)	40.1 ± 12.3	38.6 ± 12.9	< 0.01	40.6 ± 13.0	0.40	41.2 ± 12.0	0.05
LVEF (%)	55.5 ± 6.6	56.1 ± 6.7	0.10	53.3 ± 6.3	< 0.001	53.5 ± 6.1	< 0.001
LV Mass (g)	100.7 ± 33.6	105.0 ± 36.7	< 0.01	103.7 ± 36.1	< 0.01	104.7 ± 36.7	< 0.01
LV Mass index (g/m2)	59.1 ± 25.7	62.1 ± 28.3	0.001	61.4 ± 27.6	< 0.001	61.8 ± 26.7	< 0.01
RVEDV (ml)	189.3 ± 50.4	185.6 ± 51.3	0.06	178.4 ± 51.6	< 0.001	186.0 ± 52.2	< 0.01
RVEDV index (ml/m2)	111.8 ± 33.5	109.5 ± 33.6	0.05	105.3 ± 33.1	< 0.001	109.9 ± 33.5	< 0.01
RVESV (ml)	95.0 ± 34.4	93.3 ± 31.6	0.52	92.3 ± 31.6	0.21	94.2 ± 31.4	0.47
RVESV index (ml/m2)	55.9 ± 21.4	55.0 ± 19.5	0.61	54.4 ± 18.9	0.27	55.5 ± 18.5	0.53
RVEF (%)	50.5 ± 10.1	49.3 ± 10.6	0.02	48.8 ± 8.8	0.33	49.9 ± 8.2	0.82

P-values relate to each Sonic DL Cine sequence sub-type to traditional SSFP via paired T-testVolumetrics with Papillary Muscles / Trabeculations Included in Ventricular Volume

	Traditional SSFP	Sonic DL 3RR	p-value	Sonic DL 1RR	p-value	Sonic DL 1RR FB	p-value
LVEDV (ml)	152.4 ± 36.1	147.9 ± 34.9	0.001	146.0 ± 34.6	< 0.001	149.3 ± 36.0	0.01
LVEDV index (ml/m2)	89.0 ± 19.8	86.6 ± 20.1	< 0.01	85.7 ± 20.2	< 0.001	87.4 ± 20.2	0.01
LVESV (ml)	68.3 ± 20.4	65.6 ± 20.7	< 0.01	69.2 ± 21.6	0.41	70.2 ± 21.0	0.04
LVESV index (ml/m2)	40.1 ± 12.3	38.6 ± 12.9	< 0.01	40.6 ± 13.0	0.40	41.2 ± 12.0	0.05
LVEF (%)	55.5 ± 6.6	56.1 ± 6.7	0.10	53.3 ± 6.3	< 0.001	53.5 ± 6.1	< 0.001
LV Mass (g)	100.7 ± 33.6	105.0 ± 36.7	< 0.01	103.7 ± 36.1	< 0.01	104.7 ± 36.7	< 0.01
LV Mass index (g/m2)	59.1 ± 25.7	62.1 ± 28.3	0.001	61.4 ± 27.6	< 0.001	61.8 ± 26.7	< 0.01
RVEDV (ml)	189.3 ± 50.4	185.6 ± 51.3	0.06	178.4 ± 51.6	< 0.001	186.0 ± 52.2	< 0.01
RVEDV index (ml/m2)	111.8 ± 33.5	109.5 ± 33.6	0.05	105.3 ± 33.1	< 0.001	109.9 ± 33.5	< 0.01
RVESV (ml)	95.0 ± 34.4	93.3 ± 31.6	0.52	92.3 ± 31.6	0.21	94.2 ± 31.4	0.47
RVESV index (ml/m2)	55.9 ± 21.4	55.0 ± 19.5	0.61	54.4 ± 18.9	0.27	55.5 ± 18.5	0.53
RVEF (%)	50.5 ± 10.1	49.3 ± 10.6	0.02	48.8 ± 8.8	0.33	49.9 ± 8.2	0.82

P-values relate each Sonic DL Cine sequence sub-type to traditional SSFP via paired T-testRepresentative Images
Time Required for Short Axis Cine Stack