Focus Topic: Artificial Intelligence and Echocardiography
Clinical Investigations
Unsupervised Time-Series Clustering of Left Atrial Strain for Cardiovascular Risk Assessment

https://doi.org/10.1016/j.echo.2023.03.007Get rights and content

Highlights

  • Unsupervised ML grouped LA strain curves into clinically meaningful clusters.

  • Cluster 5 was associated with an increased risk for adverse cardiac events.

  • External validation yielded the same results.

  • Time-series clustering of LA strain provided incremental prognostic information.

Background

Early identification of individuals at high risk for developing cardiovascular (CV) events is of paramount importance for efficient risk management. Here, the authors investigated whether using unsupervised machine learning methods on time-series data of left atrial (LA) strain could distinguish clinically meaningful phenogroups associated with the risk for developing adverse events.

Methods

In 929 community-dwelling individuals (mean age, 51.6 years; 52.9% women), clinical and echocardiographic data were acquired, including LA strain traces, at baseline, and cardiac events were collected on average 6.3 years later. Two unsupervised learning techniques were used: (1) an ensemble of a deep convolutional neural network autoencoder with k-medoids and (2) a self-organizing map to cluster spatiotemporal patterns within LA strain curves. Clinical characteristics and cardiac outcome were used to evaluate the validity of the k clusters using the original cohort, while an external population cohort (n = 378) was used to validate the trained models.

Results

In both approaches, the optimal number of clusters was five. The first three clusters had differences in sex distribution and heart rate but had a similar low CV risk profile. On the other hand, cluster 5 had the worst CV profile and a higher prevalence of left ventricular remodeling and diastolic dysfunction compared with the other clusters. The respective indexes of cluster 4 were between those of clusters 1 to 3 and 5. After adjustment for traditional risk factors, cluster 5 had the highest risk for cardiac events compared with clusters 1, 2, and 3 (hazard ratio, 1.36; 95% CI, 1.09-1.70; P = .0063). Similar LA strain patterns were obtained when the models were applied to the external validation cohort, and clinical characteristics revealed similar CV risk profiles across all clusters.

Conclusion

Unsupervised machine learning algorithms used in time-series LA strain curves identified clinically meaningful clusters of LA deformation and provide incremental prognostic information over traditional risk factors.

Section snippets

Study Participants

The Flemish Study on Environment, Genes and Health Outcomes (FLEMENGHO) received ethical approval from the ethics committee of the University of Leuven (S64406). This study is a family-based population resource on the genetic epidemiology of CV phenotypes for which we randomly recruited a population sample within northeastern Belgium, as described elsewhere (https://flemengho.eu/en/).10 From 2005 to 2015, we invited 1,851 former FLEMENGHO participants for examinations including

Cluster Analysis of LA Strain Time Series

The 929 FLEMENGHO participants (52.9% women) included 433 subjects with hypertension (46.6%), of whom 238 (55.0%) were receiving antihypertensive drug treatment. The mean age at baseline was 51.6 ± 16.0 years.

On the basis of the inertia index, the optimal number of AE/k-medoids-based clusters was five (Figure 2). Relying on the same inertia index, we optimized the hyperparameters as outlined in Supplemental Table 2. Figure 3 shows the individual LA strain curves within each cluster as well as

Discussion

In this large prospective general population study, we applied unsupervised ML methods to group LA strain curves into clinically meaningful clusters. We identified five distinct patterns of LA strain curves associated with different clinical and echocardiographic characteristics and CV risk profiles. Although the first three clusters did not show differences in the risk for developing cardiac events, clinical characteristics revealed important differences in age and sex distribution. For

Conclusion

The use of unsupervised learning approaches integrating LA strain curves in the temporal domain could provide clinically meaningful phenogroups associated with risk for developing adverse cardiac events. Such phenogrouping might facilitate the early detection of cardiac dysfunction and, therefore, further improve CV risk stratification. As such, the constructed LA clustering model may effectively stratify individuals into low-, intermediate-, and high-risk groups. This could pave the way for

References (32)

  • T. Kuznetsova et al.

    Left ventricular strain and strain rate in a general population

    Eur Heart J

    (2008)
  • L. Al Saikhan et al.

    Prognostic implications of left ventricular strain by speckle-tracking echocardiography in the general population: a meta-analysis

    Vasc Health Risk Manag

    (2019)
  • A.M. Flores et al.

    Unsupervised learning for automated detection of coronary artery disease subgroups

    J Am Heart Assoc

    (2021)
  • F. Sabovčik et al.

    Echocardiographic phenogrouping by machine learning for risk stratification in the general population

    Eur Heart J–Digital Health

    (2021)
  • S. Sanchez-Martinez et al.

    Machine learning analysis of left ventricular function to characterize heart failure with preserved ejection fraction

    Circ Cardiovasc Imaging

    (2018)
  • T. Kuznetsova et al.

    Additive prognostic value of left ventricular systolic dysfunction in a population-based cohort

    Circ Cardiovasc Imaging

    (2016)
  • Cited by (0)

    The Research Unit Hypertension and Cardiovascular Epidemiology received grants from Internal Funds KU Leuven (C24M/21/025) and the Research Foundation Flanders (1225021N, 1S07421N, and G0C5319N).

    Conflicts of Interest: None.

    View full text