Machine learning of the spatio-temporal characteristics of echocardiographic deformation curves for infarct classification

A novel multi-task machine learning classifier for rare disease patterning using cardiac strain imaging data

To provide accurate predictions, current machine learning-based solutions require large, manually labeled training datasets. We implement persistent homology (PH), a topological tool for studying the pattern of data, to analyze echocardiography-based strain data and differentiate between rare diseases like constrictive pericarditis (CP) and restrictive cardiomyopathy (RCM). Patient population (retrospectively registered) included those presenting with heart failure due to CP (n = 51), RCM (n = 47), and patients without heart failure symptoms (n = 53). Longitudinal, radial, and circumferential strains/strain rates for left ventricular segments were processed into topological feature vectors using Machine learning PH workflow. In differentiating CP and RCM, the PH workflow model had a ROC AUC of 0.94 (Sensitivity = 92%, Specificity = 81%), compared with the GLS model AUC of 0.69 (Sensitivity = 65%, Specificity = 66%). In differentiating between all three conditions, the PH workflow model had an AUC of 0.83 (Sensitivity = 68%, Specificity = 84%), compared with the GLS model AUC of 0.68 (Sensitivity = 52% and Specificity = 76%). By employing persistent homology to differentiate the "pattern" of cardiac deformations, our machine-learning approach provides reasonable accuracy when evaluating small datasets and aids in understanding and visualizing patterns of cardiac imaging data in clinically challenging disease states.

Feasibility of artificial intelligence its current status, clinical applications, and future direction in cardiovascular disease

In routine clinical practice, the diagnosis and treatment of cardiovascular disease (CVD) rely on data in a variety of formats. These formats comprise invasive angiography, laboratory data, non-invasive imaging diagnostics, and patient history. Artificial intelligence (AI) is a field of computer science that aims to mimic human thought processes, learning capacity, and knowledge storage. In cardiovascular medicine, artificial intelligence (AI) algorithms have been used to discover novel genotypes and phenotypes in established diseases enhance patient care, enable cost effectiveness, and lower readmission and mortality rates. AI will lead to a paradigm change toward precision cardiovascular medicine in the near future. The promise application of AI in cardiovascular medicine is immense; however, failure to recognize and ignorance of the challenges may overshadow its potential clinical impact. AI can facilitate every stage in cardiology in the imaging process, from acquisition and reconstruction, to segmentation, measurement, interpretation, and subsequent clinical pathways. Along with new possibilities, new threats arise, acknowledging and understanding them is as important as understanding the machine learning (ML) methodology itself. Therefore, attention is also paid to the current opinions and guidelines regarding the validation and safety of AI. This paper provides a outline for clinicians on relevant aspects of AI and machine learning, selection of applications and methods in cardiology to date, and identifies how cardiovascular medicine could incorporate AI in the future. With progress continuing in this emerging technology, the impact for cardiovascular medicine is highlighted to provide insight for the practicing clinician and to identify potential patient benefits.

Feature-based clustering of the left ventricular strain curve for cardiovascular risk stratification in the general population

Objective

Identifying individuals with subclinical cardiovascular (CV) disease could improve monitoring and risk stratification. While peak left ventricular (LV) systolic strain has emerged as a strong prognostic factor, few studies have analyzed the whole temporal profiles of the deformation curves during the complete cardiac cycle. Therefore, in this longitudinal study, we applied an unsupervised machine learning approach based on time-series-derived features from the LV strain curve to identify distinct strain phenogroups that might be related to the risk of adverse cardiovascular events in the general population.

Method

We prospectively studied 1,185 community-dwelling individuals (mean age, 53.2 years; 51.3% women), in whom we acquired clinical and echocardiographic data including LV strain traces at baseline and collected adverse events on average 9.1 years later. A Gaussian Mixture Model (GMM) was applied to features derived from LV strain curves, including the slopes during systole, early and late diastole, peak strain, and the duration and height of diastasis. We evaluated the performance of the model using the clinical characteristics of the participants and the incidence of adverse events in the training dataset. To ascertain the validity of the trained model, we used an additional community-based cohort (n = 545) as external validation cohort.

Results

The most appropriate number of clusters to separate the LV strain curves was four. In clusters 1 and 2, we observed differences in age and heart rate distributions, but they had similarly low prevalence of CV risk factors. Cluster 4 had the worst combination of CV risk factors, and a higher prevalence of LV hypertrophy and diastolic dysfunction than in other clusters. In cluster 3, the reported values were in between those of strain clusters 2 and 4. Adjusting for traditional covariables, we observed that clusters 3 and 4 had a significantly higher risk for CV (28% and 20%, P ≤ 0.038) and cardiac (57% and 43%, P ≤ 0.024) adverse events. Using SHAP values we observed that the features that incorporate temporal information, such as the slope during systole and early diastole, had a higher impact on the model's decision than peak LV systolic strain.

Conclusion

Employing a GMM on features derived from the raw LV strain curves, we extracted clinically significant phenogroups which could provide additive prognostic information over the peak LV strain.

Deep neural network-based clustering of deformation curves reveals novel disease features in PLN pathogenic variant carriers

Echocardiographic deformation curves provide detailed information on myocardial function. Deep neural networks (DNNs) may enable automated detection of disease features in deformation curves, and improve the clinical assessment of these curves. We aimed to investigate whether an explainable DNN-based pipeline can be used to detect and visualize disease features in echocardiographic deformation curves of phospholamban (PLN) p.Arg14del variant carriers. A DNN was trained to discriminate PLN variant carriers (n = 278) from control subjects (n = 621) using raw deformation curves obtained by 2D-speckle tracking in the longitudinal axis. A visualization technique was used to identify the parts of these curves that were used by the DNN for classification. The PLN variant carriers were clustered according to the output of the visualization technique. The DNN showed excellent discriminatory performance (C-statistic 0.93 [95% CI 0.87-0.97]). We identified four clusters with PLN-associated disease features in the deformation curves. Two clusters showed previously described features: apical post-systolic shortening and reduced systolic strain. The two other clusters revealed novel features, both reflecting delayed relaxation. Additionally, a fifth cluster was identified containing variant carriers without disease features in the deformation curves, who were classified as controls by the DNN. This latter cluster had a very benign disease course regarding development of ventricular arrhythmias. Applying an explainable DNN-based pipeline to myocardial deformation curves enables automated detection and visualization of disease features. In PLN variant carriers, we discovered novel disease features which may improve individual risk stratification. Applying this approach to other diseases will further expand our knowledge on disease-specific deformation patterns. Overview of the deep neural network-based pipeline for feature detection in myocardial deformation curves. Firstly, phospholamban (PLN) p.Arg14del variant carriers and controls were selected and a deep neural network (DNN) was trained to detect the PLN variant carriers. Subsequently, a clustering-based approach was performed on the attention maps of the DNN, which revealed 4 distinct phenotypes of PLN variant carriers with different prognoses. Moreover, a cluster without features and a benign prognosis was detected.

Quality Assurance of Segmental Strain Values Provided by Commercial 2-D Speckle Tracking Echocardiography Using in Silico Models: A Report from the EACVI-ASE Strain Standardization Task Force

The aim of this study was to determine the accuracy and reproducibility of vendor-specific regional strain values by echocardiography using in silico data. Synthetic 2-D ultrasound gray-scale images of the left ventricle (LV) were generated with knowledge of the longitudinal segmental strain values from the underlying electromechanical LV model. Four of five models mimicked transmural infarctions with systolic segmental stretching in different vascular areas. Cine loops in the three apical views were synthetically generated at four noise levels. All in silico images were repeatedly analyzed by a single investigator and some by another investigator. The absolute errors varied significantly between vendors from 3.3 ± 3.1% to 11.2 ± 5.9%. The area under the curve for the identification of segmental stretching ranged from 0.80 (confidence interval: 0.77-0.83) to 0.96 (0.95-0.98). The levels of agreement for intra-investigator variability varied between -3.0% to 2.9% and -5.2% to 4.8%, and for inter-investigator variability, between -3.6% to 3.5% and -14.5% to 8.5%. Segmental strain analysis allows the identification of areas with segmental stretching with good accuracy. However, single segmental peak-strain values are not accurate and should be interpreted with caution. Nevertheless, our results indicate the usefulness of semiquantitative strain assessment for the detection of regional dysfunction.

Artificial intelligence applied to support medical decisions for the automatic analysis of echocardiogram images: A systematic review

The echocardiogram is a test that is widely used in Heart Disease Diagnoses. However, its analysis is largely dependent on the physician's experience. In this regard, artificial intelligence has become an essential technology to assist physicians. This study is a Systematic Literature Review (SLR) of primary state-of-the-art studies that used Artificial Intelligence (AI) techniques to automate echocardiogram analyses. Searches on the leading scientific article indexing platforms using a search string returned approximately 1400 articles. After applying the inclusion and exclusion criteria, 118 articles were selected to compose the detailed SLR. This SLR presents a thorough investigation of AI applied to support medical decisions for the main types of echocardiogram (Transthoracic, Transesophageal, Doppler, Stress, and Fetal). The article's data extraction indicated that the primary research interest of the studies comprised four groups: 1) Improvement of image quality; 2) identification of the cardiac window vision plane; 3) quantification and analysis of cardiac functions, and; 4) detection and classification of cardiac diseases. The articles were categorized and grouped to show the main contributions of the literature to each type of ECHO. The results indicate that the Deep Learning (DL) methods presented the best results for the detection and segmentation of the heart walls, right and left atrium and ventricles, and classification of heart diseases using images/videos obtained by echocardiography. The models that used Convolutional Neural Network (CNN) and its variations showed the best results for all groups. The evidence produced by the results presented in the tabulation of the studies indicates that the DL contributed significantly to advances in echocardiogram automated analysis processes. Although several solutions were presented regarding the automated analysis of ECHO, this area of research still has great potential for further studies to improve the accuracy of results already known in the literature.

Electromechanical Wave Imaging With Machine Learning for Automated Isochrone Generation

Standard Electromechanical Wave Imaging isochrone generation relies on manual selection of zero-crossing (ZC) locations on incremental strain curves for a number of pixels in the segmented myocardium for each echocardiographic view and patient. When considering large populations, this becomes a time-consuming process, that can be limited by inter-observer variability and operator bias. In this study, we developed and optimized an automated ZC selection algorithm, towards a faster more robust isochrone generation approach. The algorithm either relies on heuristic-based baselines or machine learning classifiers. Manually generated isochrones, previously validated against 3D intracardiac mapping, were considered as ground truth during training and performance evaluation steps. The machine learning models applied herein for the first time were: i) logistic regression; ii) support vector machine (SVM); and iii) Random Forest. The SVM and Random Forest classifiers successfully identified accessory pathways in Wolff-Parkinson-White patients, characterized sinus rhythm in humans, and localized the pacing electrode location in left ventricular paced canines on the resulting isochrones. Nevertheless, the best performing classifier was proven to be Random Forest with a precision rising from 89.5% to 97%, obtained with the voting approach that sets a probability threshold upon ZC candidate selection. Furthermore, the predictivity was not dependent on the type of testing dataset it was applied to, contrary to SVM that exhibited a 5% drop in precision on the canine testing dataset. Finally, these findings indicate that a machine learning approach can reduce user variability and considerably decrease the durations required for isochrone generation, while preserving accurate activation patterns.

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.

Artificial Intelligence, Machine Learning, and Cardiovascular Disease

Artificial intelligence (AI)-based applications have found widespread applications in many fields of science, technology, and medicine. The use of enhanced computing power of machines in clinical medicine and diagnostics has been under exploration since the 1960s. More recently, with the advent of advances in computing, algorithms enabling machine learning, especially deep learning networks that mimic the human brain in function, there has been renewed interest to use them in clinical medicine. In cardiovascular medicine, AI-based systems have found new applications in cardiovascular imaging, cardiovascular risk prediction, and newer drug targets. This article aims to describe different AI applications including machine learning and deep learning and their applications in cardiovascular medicine. AI-based applications have enhanced our understanding of different phenotypes of heart failure and congenital heart disease. These applications have led to newer treatment strategies for different types of cardiovascular diseases, newer approach to cardiovascular drug therapy and postmarketing survey of prescription drugs. However, there are several challenges in the clinical use of AI-based applications and interpretation of the results including data privacy, poorly selected/outdated data, selection bias, and unintentional continuance of historical biases/stereotypes in the data which can lead to erroneous conclusions. Still, AI is a transformative technology and has immense potential in health care.

Strain Curve Classification Using Supervised Machine Learning Algorithm with Physiologic Constraints

Speckle tracking echocardiography (STE) enables quantification of myocardial deformation by a generation of spatiotemporal strain curves or time-strain curves (TSCs). Currently, only assessment of peak global longitudinal strain is employed in clinical practice because of the uncertainty in the accuracy of STE. We describe a supervised machine learning, physiologically constrained, fully automatic algorithm, trained with labeled data, for classification of TSCs into physiologic or artifactual classes. The data set of 415 healthy patients, with three cine loops per patient, corresponding to the three standard 2-D longitudinal views, was processed using a previously published, in-house STE software termed K-SAD. We report an accuracy of 86.4% for classifying TSCs as physiologic, artifactual and undetermined curves. The positive predictive value for a physiologic strain curve is 89%. This is as a necessary step for a similar separation of pathologic conditions, to allow full utilization of the temporal information concealed in layer-specific segmental TSCs.

Classification model based on strain measurements to identify patients with arrhythmogenic cardiomyopathy with left ventricular involvement

BACKGROUND AND OBJECTIVE

A heterogenous expression characterizes arrhythmogenic cardiomyopathy (AC). The evaluation of regional wall movement included in the current Task Force Criteria is only qualitative and restricted to the right ventricle. However, a strain-based approach could precisely quantify myocardial deformation in both ventricles. We aim to define and modelize the strain behavior of the left ventricle in AC patients with left ventricular (LV) involvement by applying algorithms such as Principal Component Analysis (PCA), clustering and naïve Bayes (NB) classifiers.

METHODS

Thirty-six AC patients with LV involvement and twenty-three non-affected family members (controls) were enrolled. Feature-tracking analysis was applied to cine cardiac magnetic resonance imaging to assess strain time series from a 3D approach, to which PCA was applied. A Two-Step clustering algorithm separated the patients' group into clusters according to their level of LV strain impairment. A statistical characterization between controls and the new AC subgroups was done. Finally, a NB classifier was built and new data from a small evolutive dataset was predicted.

RESULTS

60% of AC-LV patients showed mildly affected strain and 40% severely affected strain. Both groups and controls exhibited statistically significant differences, especially when comparing controls and severely affected AC-LV patients. The classification accuracy of the strain NB classifier reached 82.76%. The model performance was as good as to classify the individuals with a 100% sensitivity and specificity for severely impaired strain patients, 85.7% and 81.1% for mildly impaired strain patients, and 69.9% and 91.4% for normal strain, respectively. Even when the severely affected LV-AC group was excluded, LV strain showed a good accuracy to differentiate patients and controls. The prediction of the evolutive dataset revealed a progressive alteration of strain in time.

CONCLUSIONS

Our LV strain classification model may help to identify AC patients with LV involvement, at least in a setting of a high pretest probability, such as family screening.

Software for Post-Processing Analysis of Strain Curves: The D-Station

Fundamento

O emprego de Speckle Tracking para estudo da função cardíaca tem grande aplicabilidade em diversos cenários. A expansão do uso deste método requer ferramentas que permitam a extração de dados relevantes das curvas de deformação cardíaca e que sejam adicionais aos parâmetros habitualmente utilizados.

Objetivos

O presente trabalho visa apresentar e validar um software de uso livre, denominado D-station, para análise das curvas de deformação cardíaca.

Métodos

A partir de arquivos raw data, o D-Station realiza a separação das fases do ciclo cardíaco, exibe simultaneamente curvas de Strain e Strain Rate de diferentes câmaras cardíacas. Para validação do software utilizamos o parâmetro Global Longitudinal Strain (GLS) avaliando-o: 1) Graficamente, a partir da comparação das Medidas emparelhadas de GLS no EchoPAC e D-Station frente à linha de igualdade; 2) pelo Coeficiente de Correlação dessas medidas; 3) pelo Teste de Hipóteses (p > 0,05); e 4) pelo Método Gráfico de Bland-Altman.

Resultados

O Coeficiente rho de Spearman apontou forte correlação entre as medidas, o Teste de Hipóteses retornou um p-value = 0.6798 >> 0,05, que também indicou a equivalência entre elas; o Método gráfico de Bland-Altman mostrou um viés ≤ 1% e dispersão ≤ 2% entre as medidas. Os testes mostraram que para valores de GLS inferiores à 10% há a tendência de aumento das diferenças percentuais, mas seus valores absolutos ainda são baixos.

Conclusão

O D-Station foi validado como uma aplicação suplementar ao EchoPAC que utiliza o raw data das curvas de Strain e Strain Rate obtidos por software proprietário. (Arq Bras Cardiol. 2020; 114(3):496-506)

Variability in the Assessment of Myocardial Strain Patterns: Implications for Adequate Interpretation

Variability in global and regional peak strain has been thoroughly studied, but variability in spatiotemporal myocardial strain patterns has not been studied as well. This study reports on such variability and its implications for adequate disease interpretation. Forty in-training operators, distributed on 20 workstations, analyzed six cases with representative deformation patterns with commercial speckle tracking. Inter-operator differences were quantified through the variability in myocardial delineations, spatiotemporal longitudinal strain patterns and peak longitudinal strain. Intra-operator differences were assessed similarly using 10 repeated measurements from a single clinician expert. Delineations varied mainly along the lateral wall and at the valve level. Peak longitudinal strain variability was low to moderate. The spatiotemporal strain patterns were consistent despite high variability at the apex and near the valve. The results indicate that relevant pattern assessment is possible despite heterogeneous experience with speckle tracking and that careful interpretation of pattern abnormalities should be recommended before a more systematic quantitative analysis.

Image-Based Cardiac Diagnosis With Machine Learning: A Review

Cardiac imaging plays an important role in the diagnosis of cardiovascular disease (CVD). Until now, its role has been limited to visual and quantitative assessment of cardiac structure and function. However, with the advent of big data and machine learning, new opportunities are emerging to build artificial intelligence tools that will directly assist the clinician in the diagnosis of CVDs. This paper presents a thorough review of recent works in this field and provide the reader with a detailed presentation of the machine learning methods that can be further exploited to enable more automated, precise and early diagnosis of most CVDs.

Machine Learning Approaches for Myocardial Motion and Deformation Analysis

Information about myocardial motion and deformation is key to differentiate normal and abnormal conditions. With the advent of approaches relying on data rather than pre-conceived models, machine learning could either improve the robustness of motion quantification or reveal patterns of motion and deformation (rather than single parameters) that differentiate pathologies. We review machine learning strategies for extracting motion-related descriptors and analyzing such features among populations, keeping in mind constraints specific to the cardiac application.

Artificial Intelligence in Cardiovascular Imaging

The number of cardiovascular imaging studies is growing exponentially, and so is the need to improve clinical workflow efficiency and avoid missed diagnoses. With the availability and use of large datasets, artificial intelligence (AI) has the potential to improve patient care at every stage of the imaging chain. Current literature indicates that in the short-term, AI has the capacity to reduce human error and save time in the clinical workflow through automated segmentation of cardiac structures. In the future, AI may expand the informational value of diagnostic images based on images alone or a combination of images and clinical variables, thus facilitating disease detection, prognosis, and decision making. This review describes the role of AI, specifically machine learning, in multimodality imaging, including echocardiography, nuclear imaging, computed tomography, and cardiac magnetic resonance, and highlights current uses of AI as well as potential challenges to its widespread implementation.

Artificial intelligence in cardiovascular imaging: state of the art and implications for the imaging cardiologist

Healthcare, conceivably more than any other area of human endeavour, has the greatest potential to be affected by artificial intelligence (AI). This potential has been shown by several reports that demonstrate equal or superhuman performance in medical tasks that aim to improve efficiency, diagnosis and prognosis. This review focuses on the state of the art of AI applications in cardiovascular imaging. It provides an overview of the current applications and studies performed, including the potential value, implications, limitations and future directions of AI in cardiovascular imaging.It is envisioned that AI will dramatically change the way doctors practise medicine. In the short term, it will assist physicians with easy tasks, such as automating measurements, making predictions based on big data, and putting clinical findings into an evidence-based context. In the long term, AI will not only assist doctors, it has the potential to significantly improve access to health and well-being data for patients and their caretakers. This empowers patients. From a physician's perspective, reliable AI assistance will be available to support clinical decision-making. Although cardiovascular studies implementing AI are increasing in number, the applications have only just started to penetrate contemporary clinical care.

Machine learning-based phenogrouping in heart failure to identify responders to cardiac resynchronization therapy

AIMS

We tested the hypothesis that a machine learning (ML) algorithm utilizing both complex echocardiographic data and clinical parameters could be used to phenogroup a heart failure (HF) cohort and identify patients with beneficial response to cardiac resynchronization therapy (CRT).

METHODS AND RESULTS

We studied 1106 HF patients from the Multicenter Automatic Defibrillator Implantation Trial with Cardiac Resynchronization Therapy (MADIT-CRT) (left ventricular ejection fraction ≤ 30%, QRS ≥ 130 ms, New York Heart Association class ≤ II) randomized to CRT with a defibrillator (CRT-D, n = 677) or an implantable cardioverter defibrillator (ICD, n = 429). An unsupervised ML algorithm (Multiple Kernel Learning and K-means clustering) was used to categorize subjects by similarities in clinical parameters, and left ventricular volume and deformation traces at baseline into mutually exclusive groups. The treatment effect of CRT-D on the primary outcome (all-cause death or HF event) and on volume response was compared among these groups. Our analysis identified four phenogroups, significantly different in the majority of baseline clinical characteristics, biomarker values, measures of left and right ventricular structure and function and the primary outcome occurrence. Two phenogroups included a higher proportion of known clinical characteristics predictive of CRT response, and were associated with a substantially better treatment effect of CRT-D on the primary outcome [hazard ratio (HR) 0.35; 95% confidence interval (CI) 0.19-0.64; P = 0.0005 and HR 0.36; 95% CI 0.19-0.68; P = 0.001] than observed in the other groups (interaction P = 0.02).

CONCLUSIONS

Our results serve as a proof-of-concept that, by integrating clinical parameters and full heart cycle imaging data, unsupervised ML can provide a clinically meaningful classification of a phenotypically heterogeneous HF cohort and might aid in optimizing the rate of responders to specific therapies.