Characterization of Responder Profiles for Cardiac Resynchronization Therapy through Unsupervised Clustering of Clinical and Strain Data

Clinical Phenotypes in Relation to Outcomes in Heart Failure Patients With Cardiac Resynchronization Therapy and Defibrillators (CRT-D): An Unsupervised Cluster Analysis

BACKGROUND

Patients with heart failure undergoing cardiac resynchronization therapy (CRT) are a heterogenous and complex population.

OBJECTIVE

To identify different clusters of patients with CRT-D and to evaluate the associations with clinical outcomes, using cluster analysis (CAs).

METHODS

Three agglomerative hierarchical CAs were performed in CRT-D patients seen between 2010 and 2019 in French hospitals. Associations between clusters and death at 1 year and death during the whole follow-up (FU) were evaluated.

RESULTS

The study included 23 029 CRT-D patients, who were analyzed in three ways, as follows: the first group was a 50% random sample of all patients (n = 11 514), the second group included patients who were dead at 1 year (n = 1604) and the third group included those alive at 3 years FU (n = 14 228). A CA was performed on each group of patients. Four clusters were identified: Cluster 1 corresponded to the low-risk phenotype; Cluster 2 to patients with coronary artery disease (CAD) with few cardiovascular (CV) risk factors and comorbidities; Cluster 3 included patients with CV risk factors and comorbidities, but low CAD; Cluster 4 corresponded to clinically complex phenotype (CAD with CV risk factors and comorbidities). Compared with Cluster 1, Clusters 2, 3, and 4 were independently associated with an increased risk of all-cause death at 1-year FU and during the whole FU (Cluster 2: hazard ratio (HR) 1.21, 95% confidence interval (CI) 1.08-1.36; Cluster 3: HR 1.15, 95% CI 1.04-1.26; and Cluster 4: HR 1.79, 95% CI 1.65-1.96).

CONCLUSION

CAs identified four statistically driven groups of CRT recipients, with specific clinical phenotypes and associated with different risks for all-cause death.

Myocardial Strain Imaging: Theory, Current Practice, and the Future

Myocardial strain imaging by echocardiography or cardiac magnetic resonance (CMR) is a powerful method to diagnose cardiac disease. Strain imaging provides measures of myocardial shortening, thickening, and lengthening and can be applied to any cardiac chamber. Left ventricular (LV) global longitudinal strain by speckle-tracking echocardiography is the most widely used clinical strain parameter. Several CMR-based modalities are available and are ready to be implemented clinically. Clinical applications of strain include global longitudinal strain as a more sensitive method than ejection fraction for diagnosing mild systolic dysfunction. This applies to patients suspected of having heart failure with normal LV ejection fraction, to early systolic dysfunction in valvular disease, and when monitoring myocardial function during cancer chemotherapy. Segmental LV strain maps provide diagnostic clues in specific cardiomyopathies, when evaluating LV dyssynchrony and ischemic dysfunction. Strain imaging is a promising modality to quantify right ventricular function. Left atrial strain may be used to evaluate LV diastolic function and filling pressure.

Characterization of cardiac resynchronization therapy response through machine learning and personalized models

INTRODUCTION

The characterization and selection of heart failure (HF) patients for cardiac resynchronization therapy (CRT) remain challenging, with around 30% non-responder rate despite following current guidelines. This study aims to propose a novel hybrid approach, integrating machine-learning and personalized models, to identify explainable phenogroups of HF patients and predict their CRT response.

METHODS

The paper proposes the creation of a complete personalized model population based on preoperative CRT patient strain curves. Based on the parameters and features extracted from these personalized models, phenotypes of patients are identified thanks to a clustering algorithm and a random forest classification is provided.

RESULTS

A close match was observed between the 162 experimental and simulated myocardial strain curves, with a mean RMSE of 4.48% (±1.08) for the 162 patients. Five phenogroups of personalized models were identified from the clustering, with response rates ranging from 52% to 94%. The classification results show a mean area under the curves (AUC) of 0.86 ± 0.06 and provided a feature importance analysis with 22 features selected. Results show both regional myocardial contractility (from 22.5% to 33.0%), tissue viability and electrical activation delays importance on CRT response for each HF patient (from 55.8 ms to 88.4 ms).

DISCUSSION

The patient-specific model parameters' analysis provides an explainable interpretation of HF patient phenogroups in relation to physiological mechanisms that seem predictive of the CRT response. These novel combined approaches appear as promising tools to improve understanding of LV mechanical dyssynchrony for HF patient characterization and CRT selection.

Risk profiles for ventricular arrhythmias in hypertrophic cardiomyopathy through clustering analysis including left ventricular strain

AIMS

The prediction of ventricular arrhythmia (VA) in hypertrophic cardiomyopathy (HCM) remains challenging. We sought to characterize the VA risk profile in HCM patients through clustering analysis combining clinical and conventional imaging parameters with information derived from left ventricular longitudinal strain analysis (LV-LS).

METHODS

A total of 434 HCM patients (65% men, mean age 56 years) were included from two referral centers and followed longitudinally (mean duration 6 years). Mechanical and temporal parameters were automatically extracted from the LV-LS segmental curves of each patient in addition to conventional clinical and imaging data. A total of 287 features were analyzed using a clustering approach (k-means). The principal endpoint was VA.

RESULTS

4 clusters were identified with a higher rhythmic risk for clusters 1 and 4 (VA rates of 26%(28/108), 13%(13/97), 12%(14/120), and 31%(34/109) for cluster 1,2,3 and 4 respectively). These 4 clusters differed mainly by LV-mechanics with a severe and homogeneous decrease of myocardial deformation for cluster 4, a small decrease for clusters 2 and 3 and a marked deformation delay and temporal dispersion for cluster 1 associated with a moderate decrease of the GLS (p < 0.0001 for GLS comparison between clusters). Patients from cluster 4 had the most severe phenotype (mean LV mass index 123 vs. 112 g/m2; p = 0.0003) with LV and left atrium (LA) remodeling (LA-volume index (LAVI) 46.6 vs. 41.5 ml/m2, p = 0.04 and LVEF 59.7 vs. 66.3%, p < 0.001) and impaired exercise capacity (% predicted peak VO2 58.6 vs. 69.5%; p = 0.025).

CONCLUSION

Processing LV-LS parameters in HCM patients 4 clusters with specific LV-strain patterns and different rhythmic risk levels are identified. Automatic extraction and analysis of LV strain parameters improves the risk stratification for VA in HCM patients.

Artificial intelligence models in prediction of response to cardiac resynchronization therapy: a systematic review

The aim of the presented review is to summarize the literature data on the accuracy and clinical applicability of artificial intelligence (AI) models as a valuable alternative to the current guidelines in predicting cardiac resynchronization therapy (CRT) response and phenotyping of patients eligible for CRT implantation. This systematic review was performed according to the PRISMA guidelines. After a search of Scopus, PubMed, Cochrane Library, and Embase databases, 675 records were identified. Twenty supervised (prediction of CRT response) and 9 unsupervised (clustering and phenotyping) AI models were analyzed qualitatively (22 studies, 14,258 patients). Fifty-five percent of AI models were based on retrospective studies. Unsupervised AI models were able to identify clusters of patients with significantly different rates of primary outcome events (death, heart failure event). In comparison to the guideline-based CRT response prediction accuracy of 70%, supervised AI models trained on cohorts with > 100 patients achieved up to 85% accuracy and an AUC of 0.86 in their prediction of response to CRT for echocardiographic and clinical outcomes, respectively. AI models seem to be an accurate and clinically applicable tool in phenotyping of patients eligible for CRT implantation and predicting potential responders. In the future, AI may help to increase CRT response rates to over 80% and improve clinical decision-making and prognosis of the patients, including reduction of mortality rates. However, these findings must be validated in randomized controlled trials.

Unsupervised Time-Series Clustering of Left Atrial Strain for Cardiovascular Risk Assessment

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.

Current role and future perspectives of artificial intelligence in echocardiography

Echocardiography is an essential tool in diagnostic cardiology and is fundamental to clinical care. Artificial intelligence (AI) can help health care providers serving as a valuable diagnostic tool for physicians in the field of echocardiography specially on the automation of measurements and interpretation of results. In addition, it can help expand the capabilities of research and discover alternative pathways in medical management specially on prognostication. In this review article, we describe the current role and future perspectives of AI in echocardiography.

Clinical and Experimental Evidence for a Strain-Based Classification of Left Bundle Branch Block-Induced Cardiac Remodeling

BACKGROUND

Septal strain patterns measured by echocardiography reflect the severity of left bundle branch block (LBBB)-induced left ventricular (LV) dysfunction. We investigated whether these LBBB strain stages predicted the response to cardiac resynchronization therapy in an observational study and developed a sheep model of LBBB-induced cardiomyopathy.

METHODS

The clinical study enrolled cardiac resynchronization therapy patients who underwent echocardiographic examination with speckle-tracking strain analysis before cardiac resynchronization therapy implant. In an experimental sheep model with pacing-induced dyssynchrony, LV remodeling and strain were assessed at baseline, at 8 and 16 weeks. Septal strain curves were classified into 5 patterns (LBBB-0 to LBBB-4).

RESULTS

The clinical study involved 250 patients (age 65 [58; 72] years; 79% men; 89% LBBB) with a median LV ejection fraction of 25 [21; 30]%. Across the stages, cardiac resynchronization therapy resulted in a gradual volumetric response, ranging from no response in LBBB-0 patients (ΔLV end-systolic volume 0 [-12; 15]%) to super-response in LBBB-4 patients (ΔLV end-systolic volume -44 [-64; -18]%) (P<0.001). LBBB-0 patients had a less favorable long-term outcome compared with those in stage LBBB≥1 (log-rank P=0.003). In 13 sheep, acute right ventricular pacing resulted in LBBB-1 (23%) and LBBB-2 (77%) patterns. Over the course of 8-16 weeks, continued pacing resulted in progressive LBBB-induced dysfunction, coincident with a transition to advanced strain patterns (92% LBBB-2 and 8% LBBB-3 at week 8; 75% LBBB-3 and 25% LBBB-4 at week 16) (P=0.023).

CONCLUSIONS

The strain-based LBBB classification reflects a pathophysiological continuum of LBBB-induced remodeling over time and is associated with the extent of reverse remodeling in observational cardiac resynchronization therapy-eligible patients.

Artificial intelligence applied to cardiovascular imaging, a critical focus on echocardiography: The point-of-view from "the other side of the coin"

Cardiovascular imaging has achieved a crucial role in the management of cardiovascular diseases. In this field, echocardiography advantages include wide availability, portability, and affordability, at a relatively low cost. However, echocardiographic assessment requires highly trained operators, and implies high observer variability, as compared with the other cardiac imaging modalities. Hence, artificial intelligence might be extremely helpful. From the point-of-view of the peripheral "Spoke" Hospital potential user ("the other side of the coin"), artificial intelligence development appears very slow in the clinical arena. Many limitations are still present, and require full involvement, cooperation, and coordination of professional operators into Hub-and-Spoke network.

Echocardiographic Advances in Dilated Cardiomyopathy

Although the overall survival of patients with dilated cardiomyopathy (DCM) has improved significantly in the last decades, a non-negligible proportion of DCM patients still shows an unfavorable prognosis. DCM patients not only need imaging techniques that are effective in diagnosis, but also suitable for long-term follow-up with frequent re-evaluations. The exponential growth of echocardiography’s technology and performance in recent years has resulted in improved diagnostic accuracy, stratification, management and follow-up of patients with DCM. This review summarizes some new developments in echocardiography and their promising applications in DCM. Although nowadays cardiac magnetic resonance (CMR) remains the gold standard technique in DCM, the echocardiographic advances and novelties proposed in the manuscript, if properly integrated into clinical practice, could bring echocardiography closer to CMR in terms of accuracy and may certify ultrasound as the technique of choice in the follow-up of DCM patients. The application in DCM patients of novel echocardiographic techniques represents an interesting emergent research area for scholars in the near future.

Automated Pattern Recognition in Whole-Cardiac Cycle Echocardiographic Data: Capturing Functional Phenotypes with Machine Learning

BACKGROUND

Echocardiography provides complex data on cardiac function that can be integrated into patterns of dysfunction related to the severity of cardiac disease. The aim of this study was to demonstrate the feasibility of applying machine learning (ML) to automate the integration of echocardiographic data from the whole cardiac cycle and to automatically recognize patterns in velocity profiles and deformation curves, allowing the identification of functional phenotypes.

METHODS

Echocardiography was performed in 189 clinically managed patients with hypertension and 97 healthy individuals without hypertension. Speckle-tracking analysis of the left ventricle and atrium was performed, and deformation curves were extracted. Aortic and mitral blood pool pulsed-wave Doppler and mitral annular tissue pulsed-wave Doppler velocity profiles were obtained. These whole-cardiac cycle deformation and velocity curves were used as ML input. Unsupervised ML was used to create a representation of patients with hypertension in a virtual space in which patients are positioned on the basis of the similarity of their integrated whole-cardiac cycle echocardiography data. Regression methods were used to explore patterns of echocardiographic traces within this virtual ML-derived space, while clustering was used to define phenogroups.

RESULTS

The algorithm captured different patterns in tissue and blood-pool velocity and deformation profiles and integrated the findings, yielding phenotypes related to normal cardiac function and others to advanced remodeling associated with pressure overload in hypertension. The addition of individuals without hypertension into the ML-derived space confirmed the interpretation of normal and remodeled phenotypes.

CONCLUSIONS

ML-based pattern recognition is feasible from echocardiographic data obtained during the whole cardiac cycle. Automated algorithms can consistently capture patterns in velocity and deformation data and, on the basis of these patterns, group patients into interpretable, clinically comprehensive phenogroups that describe structural and functional remodeling. Automated pattern recognition may potentially aid interpretation of imaging data and diagnostic accuracy.