Deep learning of left atrial structure and function provides link to atrial fibrillation risk

Abstract

Increased left atrial volume and decreased left atrial function have long been associated with atrial fibrillation. The availability of large-scale cardiac magnetic resonance imaging data paired with genetic data provides a unique opportunity to assess the genetic contributions to left atrial structure and function, and understand their relationship with risk for atrial fibrillation. Here, we use deep learning and surface reconstruction models to measure left atrial minimum volume, maximum volume, stroke volume, and emptying fraction in 40,558 UK Biobank participants. In a genome-wide association study of 35,049 participants without pre-existing cardiovascular disease, we identify 20 common genetic loci associated with left atrial structure and function. We find that polygenic contributions to increased left atrial volume are associated with atrial fibrillation and its downstream consequences, including stroke. Through Mendelian randomization, we find evidence supporting a causal role for left atrial enlargement and dysfunction on atrial fibrillation risk.

Fig. 1. Surface reconstruction for left atrial volume.

Study overview. Top left panel: orientation of the different planes in which images of the atrium were captured. The art in this panel is derived from Servier Medical Art (licensed under creativecommons by attribution, CC-BY-4.0 [https://creativecommons.org/licenses/by/4.0/]). Right panel: Example images from each of the four imaging planes; after interpretation with the deep learning model, the left atrium is colored blue. Reproduced by kind permission of UK Biobank ©. Bottom left panel: schematic overview representing reconstruction of the left atrium based on information obtained from the deep learning output from the four imaging planes.

Fig. 2. Left atrial volume variation based on AF history and cardiac filling patterns.

In the left panel, a flow diagram breaks down the imaged population into groups with and without AF, and then further into groups that do and do not appear to have normal cardiac filling patterns. In the right panel, the LAmin volume is depicted for these groups with violin plots; the median for each group is demarcated with a vertical line. Source data are provided as a Source Data file.

Fig. 3. Epidemiological relationships between left atrial volume and disease.

Left panel (“Prevalent disease”): the difference in LA volumes (Y axis) between UK Biobank participants with atrial fibrillation (“AF”), heart failure (“CHF”), hypertension (“HTN”), or stroke occurring prior to MRI compared to participants without disease (X axis). N = 39,545 participants; 813 with AF, 149 with stroke, 210 with CHF, and 11,852 with HTN. Right panel (“Incident disease”): hazard ratios for incidence of AF, CHF, HTN, and stroke (Y axis) occurring after MRI per 1 standard deviation increase in LA volumes (X axis). N = 36,900 (fewer due to prevalent disease for CHF and HTN; Supplementary Table 3); 293 with incident AF, 98 with stroke, 125 with CHF, 469 with HTN. Mean volume difference (left panel) or hazard ratio per standard deviation (right panel) estimates are represented by a circle; 95% confidence intervals for the estimate are represented by error bars. Source data are provided as a Source Data file.

Fig. 4. Genome-wide association study Manhattan plots.

Manhattan plots showing the chromosomal position (X axis) and the strength of association (−log10 of the P value, Y axis) for all LA measurements and the BSA-indexed counterparts (except for LAEF, which is dimensionless). Loci that contain SNPs with two-tailed BOLT-LMM P < 5E-08 are colored red and labeled with the name of the nearest gene to the most strongly associated variant.

Fig. 5. Variants associated with left atrial structure and function and AF.

The 8 loci associated with LA measurements and AF are displayed. All loci (except those near CASQ2 and PITX2) have multiple patterns of linkage disequilibrium and are therefore represented multiple times. Black boxes represent an association with two-tailed BOLT-LMM P < 5E-8; lighter gray boxes represent P < 5E-6. Effect sizes are oriented with respect to the minor allele. Effect size for AF loci represents the logarithm of the odds ratio. Source data are provided as a Source Data file.

Fig. 6. Incident, atrial fibrillation risk, stratified by left atrial polygenic score.

Disease incidence curves for the 417,881 participants who were unrelated to within three degrees of the participants who underwent MRI in the UK Biobank. Those in the top 5% for the BSA-indexed LAmin PRS are depicted in red; the remaining 95% are in gray. The lighter-shaded bands around each line represent the 95% confidence interval. X axis: years since enrollment in the UK Biobank. Y axis: cumulative incidence of AF (19,875 cases in the bottom 95% and 1272 cases in the top 5%). Those in the top 5% of genetically predicted LAmin indexed had an increased risk of AF (Cox HR 1.19, P = 7.9E-10) compared with those in the remaining 95% in up to 12 years of follow-up time after UK Biobank enrollment.