Clinical InvestigationEchocardiography, Machine Learning, and Cardiac Resynchronization TherapyCharacterization of Responder Profiles for Cardiac Resynchronization Therapy through Unsupervised Clustering of Clinical and Strain Data
Graphical abstract
Section snippets
Study Population
We prospectively included patients from different centers in Europe who were eligible on the basis of clinical grounds for CRT implantation according to the current European Society of Cardiology guidelines and who consented to the study. These patients agreed to undergo follow-up at the implantation center, and they were not involved in a trial related to the device implanted. More specifically, these were patients with chronic heart failure with persistent New York Heart Association
Results
Two hundred fifty-four patients were analyzed, and 250 patients were included in the final analysis. Four patients were not included in the analysis, because of missing data. Table 1 shows the patient characteristics. One hundred eighty-five responders were identified, representing 74% of the population.
According to the Silhouette score, the optimal number of clusters was identified as five clusters. Cluster 1 included 52 patients, 50% of those being responders. Cluster 2 included 65 patients,
Discussion
Unsupervised machine learning allows an integration of echocardiographic data (deformation parameters quantified automatically), as well as electrocardiographic and clinically meaningful data, into a comprehensive analysis of CRT response. We identified five phenogroups with different profiles of CRT response. Our work exploits morphologic features of strain curves that have shown to be of particular significance.5,10
Conclusion
Clustering applied to CRT recipients allows the identification of specific subgroups of CRT response and outcome. This provides information about indicators of low response, as well as very good response. Automatic quantitative longitudinal strain curve analysis offers good discriminative value among clusters, more than classical clinical and echocardiographic features, and appears to be a promising tool to improve the understanding of LV mechanics and to improve patient characterization and
References (27)
- et al.
2017 ACC/AHA/HFSA focused update of the 2013 ACCF/AHA guideline for the management of heart failure: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines and the Heart Failure Society of America
J Am Coll Cardiol
(2017) - et al.
Mechanical dyssynchrony in heart failure: still a valid concept for optimizing treatment?
Arch Cardiovasc Dis
(2017) - et al.
Impact of cardiac resynchronization therapy on left ventricular mechanics: understanding the response through a new quantitative approach based on longitudinal strain integrals
J Am Soc Echocardiogr
(2015) - et al.
Resolution of left bundle branch block-induced cardiomyopathy by cardiac resynchronization therapy
J Am Coll Cardiol
(2013) - et al.
Identification of typical left bundle branch block contraction by strain echocardiography is additive to electrocardiography in prediction of long-term outcome after cardiac resynchronization therapy
J Am Coll Cardiol
(2015) - et al.
Value of myocardial work estimation in the prediction of response to cardiac resynchronization therapy
J Am Soc Echocardiogr
(2018) - et al.
Spectral density ratio based clustering methods for the binary segmentation of protein sequences: a comparative study
Biosystems
(2010) - et al.
Spectral density ratio based clustering methods for the binary segmentation of protein sequences: a comparative study
Biosystems
(2010) - et al.
Precision phenotyping in heart failure and pattern clustering of ultrasound data for the assessment of diastolic dysfunction
JACC Cardiovasc Imaging
(2017) - et al.
Phenotypic clustering of left ventricular diastolic function parameters: patterns and prognostic relevance
JACC Cardiovasc Imaging
(2019)
Artificial intelligence in cardiovascular imaging: JACC state-of-the-art review
J Am Coll Cardiol
Predictors of super-response to cardiac resynchronization therapy and associated improvement in clinical outcome: the MADIT-CRT (Multicenter Automatic Defibrillator Implantation Trial With Cardiac Resynchronization Therapy) study
J Am Coll Cardiol
Long-term echocardiographic outcome in super-responders to cardiac resynchronization therapy and the association with mortality and defibrillator therapy
Am J Cardiol
Cited by (13)
Unsupervised Time-Series Clustering of Left Atrial Strain for Cardiovascular Risk Assessment
2023, Journal of the American Society of EchocardiographyDiastolic function and its non-invasive assessment. The quest for the holy grail continues
2023, International Journal of CardiologyFuture Guidelines for Artificial Intelligence in Echocardiography
2022, Journal of the American Society of EchocardiographyAutomated Pattern Recognition in Whole-Cardiac Cycle Echocardiographic Data: Capturing Functional Phenotypes with Machine Learning
2021, Journal of the American Society of EchocardiographyCitation Excerpt :When interpreting an echocardiographic study, a clinician considers multiple parameters at the same time, comparing findings with previous patients or with textbook examples with the goal of grouping patients into disease phenogroups. Whereas the physician's pattern recognition is based on peak values, ratios, and guideline algorithms, an ML algorithm can integrate full data on deformation and velocity profiles in every time point of the cardiac cycle, thus providing incremental information about underlying patient pathology18,19 and automating analysis of complex and information-rich data.39 If combined with the aforementioned automated data extraction, the proposed automated pattern recognition could potentially be fully integrated to the clinical work flow,40 directly on the echocardiographic machines.
Machine Learning in Cardiac Imaging: Exploring the Art of Cluster Analysis
2021, Journal of the American Society of Echocardiography
Conflicts of interest: None.
Drs. Gallard and Bidaut contributed equally to this work, and Drs. Hernandez and Donal contributed equally to this work.