Skip to content

AI Finds Associations Between Genetically Predicted LV Mass and Future CV Events

Key findings

  • In this study, deep learning was used to estimate left ventricular mass, indexed by body surface area (LVMI), from cardiac magnetic resonance (CMR) images of 43,230 participants in the UK Biobank
  • A genome-wide association study discovered 12 loci that met genome-wide significance for LVMI, including 11 that are novel, and multiple candidate genes were identified, including some previously associated with cardiac contractility and cardiomyopathy
  • Both CMR-derived LVMI and a polygenic risk score for LVMI were associated with a greater risk of incident cardiovascular events, including dilated cardiomyopathy and placement of an implantable cardioverter–defibrillator

Cardiac magnetic resonance (CMR) is the gold standard for quantifying left ventricular mass (LVM) and diagnosing LV hypertrophy. However, imaging-based measurement typically relies on manual LV segmentation, which requires substantial time and expertise. As a result, only relatively small genome-wide association studies (GWAS) have been conducted to explore the genetic variation underlying LVM.

Researchers at Massachusetts General Hospital previously reported in Cardiovascular Digital Health Journal their development of a deep learning (artificial intelligence [AI]) approach to enable estimation of LVM from CMR images. In a new study, they used their open-source model to find associations between genetically determined LVM and future heart disease.

Steven A. Lubitz, MD, MPH, cardiac electrophysiologist in the Telemachus & Irene Demoulas Family Foundation Center for Cardiac Arrhythmias at Mass General, Patrick T. Ellinor, MD, PhD, acting chief of Cardiology and the co-director of the Corrigan Minehan Heart Center, Shaan Khurshid, MD, MPH, a fellow in the Cardiovascular Research Center, and colleagues report in Nature Communications.

GWAS of CMR-derived LVM

The researchers conducted a GWAS of 43,230 participants (91% European ancestry) in the UK Biobank with CMR data available as of April 1, 2020. Because body size is a major influence on LV size and mass, they analyzed LVM indexed by body surface area (LVMI).

By applying machine learning, the team identified 12 single-nucleotide polymorphisms (SNPs) associated with LVMI at genome-wide significance. The SNP most strongly associated (rs2255167) is located at the TTN locus on chromosome 2 and has been previously associated with LVM.

The other 11 loci were novel, many located at or near genes implicated in arrhythmias, cardiomyopathy, and cardiomyocyte function. A GWAS restricted to individuals of European ancestry revealed two additional novel loci for LVMI.

Based on analyses across the 12 loci, the researchers propose multiple candidate genes related to stress response and neurohormonal regulation, cardiac contractility, cardiomyopathy, and cell signaling/function.

LVMI and Cardiovascular Disease

At a median follow-up of 2.7 years, greater CMR-derived LVMI was associated with a higher risk of multiple incident cardiovascular events/conditions:

  • Atrial fibrillation—HR, 1.46 per 1 standard deviation
  • Myocardial infarction—HR, 1.32
  • Heart failure—HR, 1.89
  • Ventricular arrhythmias—HR, 1.53
  • Dilated cardiomyopathy—HR, 2.75
  • Hypertrophic cardiomyopathy—HR, 2.39
  • Implantable cardioverter–defibrillator placed—HR, 2.43

Polygenic Risk Score

The researchers also developed a 465-variant polygenic risk score (PRS) for LVMI. In a set of UK Biobank participants separate from the GWAS sample (n=443,326), increased LVMI PRS was associated with a higher risk of most of the incident conditions listed above. Results were consistent when the PRS was applied to an independent sample of 29,354 Mass General Brigham patients.

Beyond adding to the understanding of common genetic variations underlying LVM, this study demonstrates the potential for using deep learning to make clinically relevant biological discoveries.

Learn more about the Telemachus & Irene Demoulas Family Foundation Center for Cardiac Arrhythmias

Refer a patient to the Corrigan Minehan Heart Center


Steven A. Lubitz, MD, MPH, and Shaan Khurshid, MD, MPH, of the Cardiovascular Research Center, and colleagues developed a deep learning computer model that predicts five-year risk of atrial fibrillation using 12-lead ECG data, performing as well as the CHARGE-AF score and even better when combined with that score.


Patrick T. Ellinor, MD, PhD, James P. Pirruccello, MD, and colleagues have developed an online calculator based on common demographic and clinical characteristics that estimates thoracic aortic diameter in asymptomatic patients, and thus may predict aortic dissection and rupture.