Artificial Intelligence-Assist Cardiac Magnetic Resonance Versus Standard-Of-Care in Assessment of Cardiomyopathy (RF_FR_377)

Friday, January 31, 2025

2:10 PM – 2:20 PM East Coast USA Time

Location: Blue Room PreFunction

Presenting Author(s)

MD

Marianna Daibes, MD

Postdoctoral Research Fellow
Brigham and Women's Hospital

Primary Author(s)

MD

Marianna Daibes, MD

Postdoctoral Research Fellow
Brigham and Women's Hospital

Co-Author(s)

IR

Isabela Reis Marques, MD

Postdoctoral Research Fellow
Brigham and Women's Hospital
BB

Benedikt Bernhard

MD
Cardiovascular Division, Department of Medicine, Brigham and Women's Hospital, Boston, MA, USA
BH

Bobak Heydari, MD

Professor, Harvard Medical School
Brigham and Women's Hospital
MJ

Michael Jerosch-Herold, PhD

Associate Professor
Harvard Medical School
NA

Nii O. Addy, PhD

Director of Artificial Intelligence
Vista.ai
LP

Lori Powell

Sr Manager Customer Success
Vista.ai
Juan Santos, PhD

Chief Architect and Head of Research and Co-Founder
Vista.ai
Bob Hu, MD

Chief Medical Officer
Vista.ai
Raymond Y. Kwong, MD

Professor of Medicine
Cardiovascular Division, Department of Medicine, Brigham and Women's Hospital, Boston, MA, USA

Background: CMR assessment of cardiac structures and myocardial characterization provides detailed guidance on heart failure etiology and management. However, its use is hindered by complex scanning procedure, availability of experienced operators, and long scan time. We aim to assess the benefits of a routine AI-assisted scanning method in a high-volume clinical CMR program.

Methods: Consecutive patients with signs and symptoms of heart failure were referred to CMR scanning using a standardized imaging protocol with either a standard method (SOC) or an AI-assist (AIA) scanning protocol. AIA scanning involves machine learning to automate scan planning, plane prescription, sequence tuning, and image reconstruction. Scan duration, percentage of scanning preparation time (preparation time ratio), and scan quality using a 5-point scale were compared between AIA and SOC. Clinical outcome was assessed by reviewing medical records and the Social Security Death Index. Cox regression models analyzed the association of either approach with a composite MACE encompassing all-cause death, heart failure hospitalization, myocardial infarction (MI), and unstable angina hospitalization.

Results: Among 3039 patients (58 ± 17 years), 33% underwent CMR with AIA. Overall, LGE was present in 45% and 37% had LVEF < 50%. Compared to SOC, AIA CMR resulted in 24% shorter scan duration (mean 38 ± 6 min vs. 50 ± 17 min; p< 0.001), lower preparation time ratio (mean 0.13 ± 0.07 vs. 0.27 ± 0.12, p< 0.001), and a higher mean image quality score (mean 4.08 ± 0.63 vs. 3.95 ± 0.58; p< 0.001). The Kaplan-Meier curve shows no significant difference in MACE risk between SOC and AIA.

Conclusion: AIA CMR for nonischemic cardiomyopathy is feasible, substantially more time-efficient, and provides better image quality. AIA improves scan effectiveness and CMR clinical accessibility.