Deep Learning Model of Diastolic Dysfunction Risk Stratifies the Progression of Early-Stage Aortic Stenosis.

BACKGROUND: The development and progression of aortic stenosis (AS) from aortic valve (AV) sclerosis is highly variable and difficult to predict. OBJECTIVES: The authors investigated whether a previously validated echocardiography-based deep learning (DL) model assessing diastolic dysfunction (DD) could identify the latent risk associated with the development and progression of AS. METHODS: The authors evaluated 898 participants with AV sclerosis from the ARIC (Atherosclerosis Risk In Communities) cohort study and associated the DL-predicted probability of DD with 2 endpoints: 1) the new diagnosis of AS; and 2) the composite of subsequent mortality or AV interventions. Validation was performed in 2 additional cohorts: 1) in 50 patients with mild-to-moderate AS undergoing cardiac magnetic resonance (CMR) imaging and serial echocardiographic assessments; and 2) in 18 patients with AV sclerosis undergoing 18F-sodium fluoride (NaF) and 18F-fluorodeoxyglucose positron emission tomography (PET) combined with computed tomography (CT) to assess valvular inflammation and calcification. RESULTS: In the ARIC cohort, a higher DL-predicted probability of DD was associated with the development of AS (adjusted HR: 3.482 [95% CI: 2.061-5.884]; P < 0.001) and subsequent mortality or AV interventions (adjusted HR: 7.033 [95% CI: 3.036-16.290]; P < 0.001). The multivariable Cox model (incorporating the DL-predicted probability of DD) derived from the ARIC cohort efficiently predicted the progression of AS (C-index: 0.798 [95% CI: 0.648-0.948]) in the CMR cohort. Moreover, the predictions of this multivariable Cox model correlated positively with valvular 18F-NaF mean standardized uptake values in the PET/CT cohort (r = 0.62; P = 0.008). CONCLUSIONS: Assessment of DD using DL can stratify the latent risk associated with the progression of early-stage AS.

A deep patient-similarity learning framework for the assessment of diastolic dysfunction in elderly patients.

AIMS: Age-related changes in cardiac structure and function are well recognized and make the clinical determination of abnormal left ventricular (LV) diastolic dysfunction (LVDD) particularly challenging in the elderly. We investigated whether a deep neural network (DeepNN) model of LVDD, previously validated in a younger cohort, can be implemented in an older population to predict incident heart failure (HF). METHODS AND RESULTS: A previously developed DeepNN was tested on 5596 older participants (66-90 years; 57% female; 20% Black) from the Atherosclerosis Risk in Communities Study. The association of DeepNN predictions with HF or all-cause death for the American College of Cardiology Foundation/American Heart Association Stage A/B (n = 4054) and Stage C/D (n = 1542) subgroups was assessed. The DeepNN-predicted high-risk compared with the low-risk phenogroup demonstrated an increased incidence of HF and death for both Stage A/B and Stage C/D (log-rank P < 0.0001 for all). In multi-variable analyses, the high-risk phenogroup remained an independent predictor of HF and death in both Stages A/B {adjusted hazard ratio [95% confidence interval (CI)] 6.52 [4.20-10.13] and 2.21 [1.68-2.91], both P < 0.0001} and Stage C/D [6.51 (4.06-10.44) and 1.03 (1.00-1.06), both P < 0.0001], respectively. In addition, DeepNN showed incremental value over the 2016 American Society of Echocardiography/European Association of Cardiovascular Imaging (ASE/EACVI) guidelines [net re-classification index, 0.5 (CI 0.4-0.6), P < 0.001; C-statistic improvement, DeepNN (0.76) vs. ASE/EACVI (0.70), P < 0.001] overall and maintained across stage groups. CONCLUSION: Despite training with a younger cohort, a deep patient-similarity-based learning framework for assessing LVDD provides a robust prediction of all-cause death and incident HF for older patients.

Electrocardiogram-Based Machine Learning Emulator Model for Predicting Novel Echocardiography-Derived Phenogroups for Cardiac Risk-Stratification: A Prospective Multicenter Cohort Study.

Purpose: Electrocardiography (ECG)-derived machine learning models can predict echocardiography (echo)-derived indices of systolic or diastolic function. However, systolic and diastolic dysfunction frequently coexists, which necessitates an integrated assessment for optimal risk-stratification. We explored an ECG-derived model that emulates an echo-derived model that combines multiple parameters for identifying patient phenogroups at risk for major adverse cardiac events (MACE). Methods: In this substudy of a prospective, multicenter study, patients from 3 institutions (n=727) formed an internal cohort, and the fourth institution was reserved as an external test set (n=518). A previously validated patient similarity analysis model was used for labeling the patients as low-/high-risk phenogroups. These labels were utilized for training an ECG-derived deep neural network model to predict MACE risk per phenogroup. After 5-fold cross-validation training, the model was tested on the reserved external dataset. Results: Our ECG-derived model showed robust classification of patients, with area under the receiver operating characteristic curve of 0.86 (95% CI: 0.79-0.91) and 0.84 (95% CI: 0.80-0.87), sensitivity of 80% and 76%, and specificity of 88% and 75% for the internal and external test sets, respectively. The ECG-derived model demonstrated an increased probability for MACE in high-risk vs low-risk patients (21% vs 3%; P<0.001), which was similar to the echo-trained model (21% vs 5%; P<0.001), suggesting comparable utility. Conclusions: This novel ECG-derived machine learning model provides a cost-effective strategy for predicting patient subgroups in whom an integrated milieu of systolic and diastolic dysfunction is associated with a high risk of MACE.

High Prevalence of Pericardial Involvement in College Student Athletes Recovering From COVID-19.

OBJECTIVES: This study sought to explore the spectrum of cardiac abnormalities in student athletes who returned to university campus in July 2020 with uncomplicated coronavirus disease 2019 (COVID-19). BACKGROUND: There is limited information on cardiovascular involvement in young individuals with mild or asymptomatic COVID-19. METHODS: Screening echocardiograms were performed in 54 consecutive student athletes (mean age 19 years; 85% male) who had positive results of reverse transcription polymerase chain reaction nasal swab testing of the upper respiratory tract or immunoglobulin G antibodies against severe acute respiratory syndrome coronavirus type 2. Sequential cardiac magnetic resonance imaging was performed in 48 (89%) subjects. RESULTS: A total of 16 (30%) athletes were asymptomatic, whereas 36 (66%) and 2 (4%) athletes reported mild and moderate COVID-19 related symptoms, respectively. For the 48 athletes completing both imaging studies, abnormal findings were identified in 27 (56.3%) individuals. This included 19 (39.5%) athletes with pericardial late enhancements with associated pericardial effusion. Of the individuals with pericardial enhancements, 6 (12.5%) had reduced global longitudinal strain and/or an increased native T1. One patient showed myocardial enhancement, and reduced left ventricular ejection fraction or reduced global longitudinal strain with or without increased native T1 values was also identified in an additional 7 (14.6%) individuals. Native T2 findings were normal in all subjects, and no specific imaging features of myocardial inflammation were identified. Hierarchical clustering of left ventricular regional strain identified 3 unique myopericardial phenotypes that showed significant association with the cardiac magnetic resonance findings (p = 0.03). CONCLUSIONS: More than 1 in 3 previously healthy college athletes recovering from COVID-19 infection showed imaging features of a resolving pericardial inflammation. Although subtle changes in myocardial structure and function were identified, no athlete showed specific imaging features to suggest an ongoing myocarditis. Further studies are needed to understand the clinical implications and long-term evolution of these abnormalities in uncomplicated COVID-19.

Machine Learning Assessment of Left Ventricular Diastolic Function Based on Electrocardiographic Features.

BACKGROUND: Left ventricular (LV) diastolic dysfunction is recognized as playing a major role in the pathophysiology of heart failure; however, clinical tools for identifying diastolic dysfunction before echocardiography remain imprecise. OBJECTIVES: This study sought to develop machine-learning models that quantitatively estimate myocardial relaxation using clinical and electrocardiography (ECG) variables as a first step in the detection of LV diastolic dysfunction. METHODS: A multicenter prospective study was conducted at 4 institutions in North America enrolling a total of 1,202 subjects. Patients from 3 institutions (n = 814) formed an internal cohort and were randomly divided into training and internal test sets (80:20). Machine-learning models were developed using signal-processed ECG, traditional ECG, and clinical features and were tested using the test set. Data from the fourth institution was reserved as an external test set (n = 388) to evaluate the model generalizability. RESULTS: Despite diversity in subjects, the machine-learning model predicted the quantitative values of the LV relaxation velocities (e') measured by echocardiography in both internal and external test sets (mean absolute error: 1.46 and 1.93 cm/s; adjusted R2 = 0.57 and 0.46, respectively). Analysis of the area under the receiver operating characteristic curve (AUC) revealed that the estimated e' discriminated the guideline-recommended thresholds for abnormal myocardial relaxation and diastolic and systolic dysfunction (LV ejection fraction) the internal (area under the curve [AUC]: 0.83, 0.76, and 0.75) and external test sets (0.84, 0.80, and 0.81), respectively. Moreover, the estimated e' allowed prediction of LV diastolic dysfunction based on multiple age- and sex-adjusted reference limits (AUC: 0.88 and 0.94 in the internal and external sets, respectively). CONCLUSIONS: A quantitative prediction of myocardial relaxation can be performed using easily obtained clinical and ECG features. This cost-effective strategy may be a valuable first clinical step for assessing the presence of LV dysfunction and may potentially aid in the early diagnosis and management of heart failure patients.

A Network-Based "Phenomics" Approach for Discovering Patient Subtypes From High-Throughput Cardiac Imaging Data.

OBJECTIVES: The authors present a method that focuses on cohort matching algorithms for performing patient-to-patient comparisons along multiple echocardiographic parameters for predicting meaningful patient subgroups. BACKGROUND: Recent efforts in collecting multiomics data open numerous opportunities for comprehensive integration of highly heterogenous data to classify a patient's cardiovascular state, eventually leading to tailored therapies. METHODS: A total of 42 echocardiography features, including 2-dimensional and Doppler measurements, left ventricular (LV) and atrial speckle-tracking, and vector flow mapping data, were obtained in 297 patients. A similarity network was developed to delineate distinct patient phenotypes, and then neural network models were trained for discriminating the phenotypic presentations. RESULTS: The patient similarity model identified 4 clusters (I to IV), with patients in each cluster showed distinctive clinical presentations based on American College of Cardiology/American Heart Association heart failure stage and the occurrence of short-term major adverse cardiac and cerebrovascular events. Compared with other clusters, cluster IV had a higher prevalence of stage C or D heart failure (78%; p < 0.001), New York Heart Association functional classes III or IV (61%; p < 0.001), and a higher incidence of major adverse cardiac and cerebrovascular events (p < 0.001). The neural network model showed robust prediction of patient clusters, with area under the receiver-operating characteristic curve ranging from 0.82 to 0.99 for the independent hold-out validation set. CONCLUSIONS: Automated computational methods for phenotyping can be an effective strategy to fuse multidimensional parameters of LV structure and function. It can identify distinct cardiac phenogroups in terms of clinical characteristics, cardiac structure and function, hemodynamics, and outcomes.

A low-cost texture-based pipeline for predicting myocardial tissue remodeling and fibrosis using cardiac ultrasound.

BACKGROUND: Maturation of ultrasound myocardial tissue characterization may have far-reaching implications as a widely available alternative to cardiac magnetic resonance (CMR) for risk stratification in left ventricular (LV) remodeling. METHODS: We extracted 328 texture-based features of myocardium from still ultrasound images. After we explored the phenotypes of myocardial textures using unsupervised similarity networks, global LV remodeling parameters were predicted using supervised machine learning models. Separately, we also developed supervised models for predicting the presence of myocardial fibrosis using another cohort who underwent cardiac magnetic resonance (CMR). For the prediction, patients were divided into a training and test set (80:20). FINDINGS: Texture-based tissue feature extraction was feasible in 97% of total 534 patients. Interpatient similarity analysis delineated two patient groups based on the texture features: one group had more advanced LV remodeling parameters compared to the other group. Furthermore, this group was associated with a higher incidence of cardiac deaths (p = 0.001) and major adverse cardiac events (p < 0.001). The supervised models predicted reduced LV ejection fraction (<50%) and global longitudinal strain (<16%) with area under the receiver-operator-characteristics curves (ROC AUC) of 0.83 and 0.87 in the hold-out test set, respectively. Furthermore, the presence of myocardial fibrosis was predicted from only ultrasound myocardial texture with an ROC AUC of 0.84 (sensitivity 86.4% and specificity 83.3%) in the test set. INTERPRETATION: Ultrasound texture-based myocardial tissue characterization identified phenotypic features of LV remodeling from still ultrasound images. Further clinical validation may address critical barriers in the adoption of ultrasound techniques for myocardial tissue characterization. FUNDING: None.

Interpatient Similarities in Cardiac Function: A Platform for Personalized Cardiovascular Medicine.

OBJECTIVES: The authors applied unsupervised machine-learning techniques for integrating echocardiographic features of left ventricular (LV) structure and function into a patient similarity network that predicted major adverse cardiac event(s) (MACE) in an individual patient. BACKGROUND: Patient similarity analysis is an evolving paradigm for precision medicine in which patients are clustered or classified based on their similarities in several clinical features. METHODS: A retrospective cohort of 866 patients was used to develop a network architecture using 9 echocardiographic features of LV structure and function. The data for 468 patients from 2 prospective cohort registries were then added to test the model's generalizability. RESULTS: The map of cross-sectional data in the retrospective cohort resulted in a looped patient network that persisted even after the addition of data from the prospective cohort registries. After subdividing the loop into 4 regions, patients in each region showed unique differences in LV function, with Kaplan-Meier curves demonstrating significant differences in MACE-related rehospitalization and death (both p < 0.001). Addition of network information to clinical risk predictors resulted in significant improvements in net reclassification, integrated discrimination, and median risk scores for predicting MACE (p < 0.05 for all). Furthermore, the network predicted the cardiac disease cycle in each of the 96 patients who had second echocardiographic evaluations. An improvement or remaining in low-risk regions was associated with lower MACE-related rehospitalization rates than worsening or remaining in high-risk regions (3% vs. 37%; p < 0.001). CONCLUSIONS: Patient similarity analysis integrates multiple features of cardiac function to develop a phenotypic network in which patients can be mapped to specific locations associated with specific disease stage and clinical outcomes. The use of patient similarity analysis may have relevance for automated staging of cardiac disease severity, personalized prediction of prognosis, and monitoring progression or response to therapies.

Prediction of coronary artery calcium scoring from surface electrocardiogram in atherosclerotic cardiovascular disease: a pilot study.

Aims: Coronary artery calcium (CAC) scoring is an established tool for cardiovascular risk stratification. However, the lack of widespread availability and concerns about radiation exposure have limited the universal clinical utilization of CAC. In this study, we sought to explore whether machine learning (ML) approaches can aid cardiovascular risk stratification by predicting guideline recommended CAC score categories from clinical features and surface electrocardiograms. Methods and results: In this substudy of a prospective, multicentre trial, a total of 534 subjects referred for CAC scores and electrocardiographic data were split into 80% training and 20% testing sets. Two binary outcome ML logistic regression models were developed for prediction of CAC scores equal to 0 and ≥400. Both CAC = 0 and CAC ≥400 models yielded values for the area under the curve, sensitivity, specificity, and accuracy of 84%, 92%, 70%, and 75%, and 87%, 91%, 75%, and 81%, respectively. We further tested the CAC ≥400 model to risk stratify a cohort of 87 subjects referred for invasive coronary angiography. Using an intermediate or higher pretest probability (≥15%) to predict CAC ≥400, the model predicted the presence of significant coronary artery stenosis (P = 0.025), the need for revascularization (P < 0.001), notably bypass surgery (P = 0.021), and major adverse cardiovascular events (P = 0.023) during a median follow-up period of 2 years. Conclusion: ML techniques can extract information from electrocardiographic data and clinical variables to predict CAC score categories and similarly risk-stratify patients with suspected coronary artery disease.