Impact of accelerated free-breathing cardiac MRI with deep learning reconstruction on image quality and scan duration in a pediatric cohort (RF_FR_324)

Friday, January 31, 2025

10:00 AM – 10:10 AM East Coast USA Time

Location: Blue Room PreFunction

Presenting Author(s)

AC

Adam B. Christopher, MD

Assistant professor
UPMC Children's Hospital of Pittsburgh

Primary Author(s)

AG

ADITI GUPTA, MD

FELLOW
Children Hospital of Pittsburgh

Co-Author(s)

AC

Adam B. Christopher, MD

Assistant professor
UPMC Children's Hospital of Pittsburgh
AS

ALESSANDRO SCOTTI, PhD

MR Clinical Scientist
GE HealthCare
KF

Kyoko Fujimoto, PhD

Director, Research and Scientific Affairs
GE HealthCare
Tarek Alsaied, MD

Assistant Professor of Pediatrics
Children's Hospital of Pittsburgh
Laura Olivieri, MD

Associate Professor of Pediatrics
Children's Hospital of Pittsburgh of UPMC

Background: Pediatric cardiac MRI poses unique demands of capturing smaller structures with faster heart rates in patients with often limited compliance. Cine imaging is limited by the need for breath holding and long acquisition times. The availability of more highly accelerated sequences in CMR using deep learning reconstruction in combination with free-breathing cine imaging has potential to impact pediatric cardiac MRI workflow^1,2. The aim of the study was to compare study duration and signal to noise ratio (SNR) in children < 18 years scanned using standard breath-held (BH) cine SSFP to children scanned using highly accelerated free-breathing (FB) SSFP ¹

Methods: All subjects were referred for a clinically indicated cardiac MRI, and standard lab protocols were followed based on indication for the study^1,2.Study duration, sequence selection, and image SNR were measured from a standard BH SSFP cohort prior to implementation of FB DL (Jan 2022 to Jun 2024) and those scanned following clinical implementation of FB DL imaging acquisition which is undersampled and reconstructed (July-August 2024). Diagnoses were categorized as Simple Congenital Heart Disease (CHD) (septal defects, aortic anomalies, isolated valve defects), Complex CHD (tetralogy of Fallot, left heart obstruction, conotruncal defects, Ebstein anomaly, single ventricle) and Structurally Normal (myocarditis/cardiomyopathy, ectopy, coronary abnormalities). Image quality was assessed by aortic signal to noise ratio (SNR).

Results: FB DL imaging was used in 58 patients (12.5 + 3.9 years, BSA 1.5 + 0.5), compared to 80 patients imaged with standard breath-held SSFP imaging (14.6 + 4 years, BSA 1.4 + 0.4). The average age was significantly lower in the FB DL suggestive of scanning younger population in the current era. The average duration of scan in the FB DL cohort was 29 +10.1 mins, vs in the BH standard cohort 44 +13.9 mins (p < 0.01). For the patients who had LGE performed (31% vs 51% pts, p=0.02 in the FB DL vs BH standard cohort), the study duration was significantly lower (29 + 9.7 mins vs 49 + 13.9 mins, p< 0.01) in the FB DL cohort. Similarly, for patients who had 4D flow acquired as part of the study (62%, vs 56% respectively, p=0.9) the study duration was significantly less (31 + 10.0mins vs 44 + 13.9 mins, p< 0.01) in the FB DL cohort. There was no statistical difference in the SNR of the two cohorts (p = 0.07).

Conclusion: FB DL reconstruction significantly reduces the acquisition times for pediatric cardiac MRI as compared to standard breath held SSFP imaging acquisition without compromising the SNR.