JMIR Publications

ABSTRACT

Background:

Cardiovascular magnetic resonance (CMR) was widely utilized across various cardiac conditions, systematically assessed the cardiac anatomical structure and functional dynamics. Machine learning (ML) can accurately predict outcomes and understand inherent features of clinical data.

Objective:

To derive CMR phenotypes using an unsupervised ML model, investigate the relationship between the phenotypes and stroke risk, and relearn the phenotypes through supervised ML.

Methods:

We enrolled 36,467 stroke-free participants and extracted the CMR parameters from the UK Biobank, with follow-up data extending until September 30, 2023. Utilizing the Generative Topographic Mapping (GTM) technique, we identified micro-clusters of participants and then derived macro-clusters through agglomerative hierarchical clustering. We employed supervised ML models to predict cardiac function phenotypes and used Cox proportional hazards models to assess the association between these phenotypes and long-term stroke risk.

Results:

We enrolled 36,467 participants in the study. During a mean follow-up time of 14.7 years, 1.4% participants developed stroke. After GTM modeling, we identified 2 distinct phenotypes: phenotype 1, characterized by an increased risk of left ventricular and left atrial dysfunction, excessive volume or pressure overload, and impaired myocardial function; and phenotype 2, which was significantly associated with a reduced risk of stroke (hazard ratio, 0.691; 95% confidence interval [CI], 0.556-0.860; P = 0.001).We selected the random forest model as the optimal model for the phenotypes, demonstrating high accuracy (area under the curve [95% CI] training = 0.914 [0.911-0.918] and validation = 0.867 [0.858-0.876]) and calibration ability (Brier score [95% CI] training = 0.111 [0.109-0.113]), and validation = 0.132 [0.127-0.137]).

Conclusions:

By integrating unsupervised and supervised ML methods, we identified phenotypes of cardiac function based on CMR parameters, demonstrating robust predictive ability for incident stroke, which may have the potential to improve preventive and therapeutic strategies for high-risk populations.