Deep Learning Predicts Heart Failure With Preserved, Mid-Range, and Reduced Left Ventricular Ejection Fraction From Patient Clinical Profiles

Circadian assessment of heart failure using explainable deep learning and novel multi-parameter polar images

BACKGROUND AND OBJECTIVE

Heart failure (HF) is a multi-faceted and life-threatening syndrome that affects more than 64.3 million people worldwide. Current gold-standard screening technique, echocardiography, neglects cardiovascular information regulated by the circadian rhythm and does not incorporate knowledge from patient profiles. In this study, we propose a novel multi-parameter approach to assess heart failure using heart rate variability (HRV) and patient clinical information.

METHODS

In this approach, features from 24-hour HRV and clinical information were combined as a single polar image and fed to a 2D deep learning model to infer the HF condition. The edges of the polar image correspond to the timely variation of different features, each of which carries information on the function of the heart, and internal illustrates color-coded patient clinical information.

RESULTS

Under a leave-one-subject-out cross-validation scheme and using 7,575 polar images from a multi-center cohort (American and Greek) of 303 coronary artery disease patients (median age: 58 years [50-65], median body mass index (BMI): 27.28 kg/m2 [24.91-29.41]), the model yielded mean values for the area under the receiver operating characteristics curve (AUC), sensitivity, specificity, normalized Matthews correlation coefficient (NMCC), and accuracy of 0.883, 90.68%, 95.19%, 0.93, and 92.62%, respectively. Moreover, interpretation of the model showed proper attention to key hourly intervals and clinical information for each HF stage.

CONCLUSIONS

The proposed approach could be a powerful early HF screening tool and a supplemental circadian enhancement to echocardiography which sets the basis for next-generation personalized healthcare.

Identification of Congenital Valvular Murmurs in Young Patients Using Deep Learning-Based Attention Transformers and Phonocardiograms

One in every four newborns suffers from congenital heart disease (CHD) that causes defects in the heart structure. The current gold-standard assessment technique, echocardiography, causes delays in the diagnosis owing to the need for experts who vary markedly in their ability to detect and interpret pathological patterns. Moreover, echo is still causing cost difficulties for low- and middle-income countries. Here, we developed a deep learning-based attention transformer model to automate the detection of heart murmurs caused by CHD at an early stage of life using cost-effective and widely available phonocardiography (PCG). PCG recordings were obtained from 942 young patients at four major auscultation locations, including the aortic valve (AV), mitral valve (MV), pulmonary valve (PV), and tricuspid valve (TV), and they were annotated by experts as absent, present, or unknown murmurs. A transformation to wavelet features was performed to reduce the dimensionality before the deep learning stage for inferring the medical condition. The performance was validated through 10-fold cross-validation and yielded an average accuracy and sensitivity of 90.23 % and 72.41 %, respectively. The accuracy of discriminating between murmurs' absence and presence reached 76.10 % when evaluated on unseen data. The model had accuracies of 70 %, 88 %, and 86 % in predicting murmur presence in infants, children, and adolescents, respectively. The interpretation of the model revealed proper discrimination between the learned attributes, and AV channel was found important (score 0.75) for the murmur absence predictions while MV and TV were more important for murmur presence predictions. The findings potentiate deep learning as a powerful front-line tool for inferring CHD status in PCG recordings leveraging early detection of heart anomalies in young people. It is suggested as a tool that can be used independently from high-cost machinery or expert assessment.

Investigating automated regression models for estimating left ventricular ejection fraction levels in heart failure patients using circadian ECG features

Heart Failure (HF) significantly impacts approximately 26 million people worldwide, causing disruptions in the normal functioning of their hearts. The estimation of left ventricular ejection fraction (LVEF) plays a crucial role in the diagnosis, risk stratification, treatment selection, and monitoring of heart failure. However, achieving a definitive assessment is challenging, necessitating the use of echocardiography. Electrocardiogram (ECG) is a relatively simple, quick to obtain, provides continuous monitoring of patient's cardiac rhythm, and cost-effective procedure compared to echocardiography. In this study, we compare several regression models (support vector machine (SVM), extreme gradient boosting (XGBOOST), gaussian process regression (GPR) and decision tree) for the estimation of LVEF for three groups of HF patients at hourly intervals using 24-hour ECG recordings. Data from 303 HF patients with preserved, mid-range, or reduced LVEF were obtained from a multicentre cohort (American and Greek). ECG extracted features were used to train the different regression models in one-hour intervals. To enhance the best possible LVEF level estimations, hyperparameters tuning in nested loop approach was implemented (the outer loop divides the data into training and testing sets, while the inner loop further divides the training set into smaller sets for cross-validation). LVEF levels were best estimated using rational quadratic GPR and fine decision tree regression models with an average root mean square error (RMSE) of 3.83% and 3.42%, and correlation coefficients of 0.92 (p<0.01) and 0.91 (p<0.01), respectively. Furthermore, according to the experimental findings, the time periods of midnight-1 am, 8-9 am, and 10-11 pm demonstrated to be the lowest RMSE values between the actual and predicted LVEF levels. The findings could potentially lead to the development of an automated screening system for patients with coronary artery disease (CAD) by using the best measurement timings during their circadian cycles.

Prediction of fetal RR intervals from maternal factors using machine learning models

Previous literature has highlighted the importance of maternal behavior during the prenatal period for the upbringing of healthy adults. During pregnancy, fetal health assessments are mainly carried out non-invasively by monitoring fetal growth and heart rate (HR) or RR interval (RRI). Despite this, research entailing prediction of fHRs from mHRs is scarce mainly due to the difficulty in non-invasive measurements of fetal electrocardiogram (fECG). Also, so far, it is unknown how mHRs are associated with fHR over the short term. In this study, we used two machine learning models, support vector regression (SVR) and random forest (RF), for predicting average fetal RRI (fRRI). The predicted fRRI values were compared with actual fRRI values calculated from non-invasive fECG. fRRI was predicted from 13 maternal features that consisted of age, weight, and non-invasive ECG-derived parameters that included HR variability (HRV) and R wave amplitude variability. 156 records were used for training the models and the results showed that the SVR model outperformed the RF model with a root mean square error (RMSE) of 29 ms and an average error percentage (< 5%). Correlation analysis between predicted and actual fRRI values showed that the Spearman coefficient for the SVR and RF models were 0.31 (P < 0.001) and 0.19 (P < 0.05), respectively. The SVR model was further used to predict fRRI of 14 subjects who were not included in the training. The latter prediction results showed that individual error percentages were (≤ 5%) except in 3 subjects. The results of this study show that maternal factors can be potentially used for the assessment of fetal well-being based on fetal HR or RRI.

Baroreflex activity through the analysis of the cardio-respiratory variability influence over blood pressure in cardiomyopathy patients

A large portion of the elderly population are affected by cardiovascular diseases. Early prognosis of cardiomyopathies remains a challenge. The aim of this study was to classify cardiomyopathy patients by their etiology based on significant indexes extracted from the characterization of the baroreflex mechanism in function of the influence of the cardio-respiratory activity over the blood pressure. Forty-one cardiomyopathy patients (CMP) classified as ischemic (ICM-24 patients) and dilated (DCM-17 patients) were considered. In addition, thirty-nine control (CON) subjects were used as reference. The beat-to-beat (BBI) time series, from the electrocardiographic (ECG) signal, the systolic (SBP), and diastolic (DBP) time series, from the blood pressure signal (BP), and the respiratory time (TT), from the respiratory flow (RF) signal, were extracted. The three-dimensional representation of the cardiorespiratory and vascular activities was characterized geometrically, by fitting a polygon that contains 95% of data, and by statistical descriptive indices. DCM patients presented specific patterns in the respiratory response to decreasing blood pressure activity. ICM patients presented more stable cardiorespiratory activity in comparison with DCM patients. In general, CMP shown limited ability to regulate changes in blood pressure. In addition, patients also shown a limited ability of their cardiac and respiratory systems response to regulate incremental changes of the vascular variability and a lower heart rate variability. The best classifiers were used to build support vector machine models. The optimal model to classify ICM versus DCM patients achieved 92.7% accuracy, 94.1% sensitivity, and 91.7% specificity. When comparing CMP patients and CON subjects, the best model achieved 86.2% accuracy, 82.9% sensitivity, and 89.7% specificity. When comparing ICM patients and CON subjects, the best model achieved 88.9% accuracy, 87.5% sensitivity, and 89.7% specificity. When comparing DCM patients and CON subjects, the best model achieved 87.5% accuracy, 76.5% sensitivity, and 92.3% specificity. In conclusion, this study introduced a new method for the classification of patients by their etiology based on new indices from the analysis of the baroreflex mechanism.

Heart Failure Assessment Using Multiparameter Polar Representations and Deep Learning. Annu Int Conf IEEE Eng Med Biol Soc 2023;2023:1-4. [PMID: 38083567 DOI: 10.1109/embc40787.2023.10341132] [Citation(s) in RCA: 0] [Impact Index Per Article: 0]

Heart failure refers to the inability of the heart to pump enough amount of blood to the body. Nearly 7 million people die every year because of its complications. Current gold-standard screening techniques through echocardiography do not incorporate information about the circadian rhythm of the heart and clinical information of patients. In this vein, we propose a novel approach to integrate 24-hour heart rate variability (HRV) features and patient profile information in a single multi-parameter and color-coded polar representation. The proposed approach was validated by training a deep learning model from 7,575 generated images to predict heart failure groups, i.e., preserved, mid-range, and reduced left ventricular ejection fraction. The developed model had overall accuracy, sensitivity, and specificity of 93%, 88%, and 95%, respectively. Moreover, it had a high area under the receiver operating characteristics curve (AUROC) of 0.88 and an area under the precision-recalled curve (AUPR) of 0.79. The novel approach proposed in this study suggests a new protocol for assessing cardiovascular diseases to act as a complementary tool to echocardiography as it provides insights on the circadian rhythm of the heart and can be potentially personalized according to patient clinical profile information.Clinical relevance- Implementing polar representations with deep learning in clinical settings to supplement echocardiography leverages continuous monitoring of the heart's circadian rhythm and personalized cardiovascular medicine while reducing the burden on medical practitioners.

Leveraging Machine Learning Techniques to Forecast Chronic Total Occlusion before Coronary Angiography

BACKGROUND

Chronic total occlusion (CTO) remains the most challenging procedure in coronary artery disease (CAD) for interventional cardiology. Although some clinical risk factors for CAD have been identified, there is no personalized prognosis test available to confidently identify patients at high or low risk for CTO CAD. This investigation aimed to use a machine learning algorithm for clinical features from clinical routine to develop a precision medicine tool to predict CTO before CAG.

METHODS

Data from 1473 CAD patients were obtained, including 1105 in the training cohort and 368 in the testing cohort. The baseline clinical characteristics were collected. Univariate and multivariate logistic regression analyses were conducted to identify independent risk factors that impact the diagnosis of CTO. A CTO predicting model was established and validated based on the independent predictors using a machine learning algorithm. The area under the curve (AUC) was used to evaluate the model.

RESULTS

The CTO prediction model was developed with the training cohort using the machine learning algorithm. Eight variables were confirmed as 'important': gender (male), neutrophil percentage (NE%), hematocrit (HCT), total cholesterol (TC), high-density lipoprotein cholesterol (HDL), ejection fraction (EF), troponin I (TnI), and N-terminal pro-B-type natriuretic peptide (NT-proBNP). The model achieved good concordance indices of 0.724 and 0.719 in the training and testing cohorts, respectively.

CONCLUSIONS

An easy-to-use tool to predict CTO in patients with CAD was developed and validated. More research with larger cohorts are warranted to improve the prediction model, which can support clinician decisions on the early discerning CTO in CAD patients.

A systemic review and meta-analysis comparing the ability of diagnostic of the third heart sound and left ventricular ejection fraction in heart failure

Objective

This study aimed to compare the sensitivity and specificity of diagnosis between the third heart sound (S3) and left ventricular ejection fraction (LVEF) in heart failure (HF).

Methods

Relevant studies were searched in PubMed, SinoMed, China National Knowledge Infrastructure, and the Cochrane Trial Register until February 20, 2022. The sensitivity, specificity, likelihood ratio (LR), and diagnostic odds ratio (DOR) were pooled. The symmetric receiver operator characteristic curve (SROC) and Fagan’s nomogram were drawn. The source of heterogeneity was explored by meta-regression and subgroup analysis.

Results

A total of 19 studies, involving 5,614 participants, were included. The combined sensitivity of S3 was 0.23 [95% confidence interval (CI) (0.15–0.33), specificity was 0.94 [95% CI (0.82–0.98)], area under the SROC curve was 0.49, and the DOR was 4.55; while the sensitivity of LVEF was 0.70 [95% CI (0.53–0.83)], specificity was 0.79 [95% CI (0.75–0.82)], area under the SROC curve was 0.79, and the DOR was 8.64. No publication bias was detected in Deeks’ funnel plot. The prospective design, partial verification bias, and blind contributed to the heterogeneity in specificity, while adequate description of study participants contributed to the heterogeneity in sensitivity. In Fagan’s nomogram, the post-test probability was 48% when the pre-test probability was set as 20%, while in LVEF, the post-test probability was 45% when the pre-test probability was set as 20%.

Conclusion

The use of S3 alone presented lower sensitivity in diagnosing HF compared with LVEF, whereas it was useful in early pathological assessment.

Deep learning identifies cardiac coupling between mother and fetus during gestation

In the last two decades, stillbirth has caused around 2 million fetal deaths worldwide. Although current ultrasound tools are reliably used for the assessment of fetal growth during pregnancy, it still raises safety issues on the fetus, requires skilled providers, and has economic concerns in less developed countries. Here, we propose deep coherence, a novel artificial intelligence (AI) approach that relies on 1 min non-invasive electrocardiography (ECG) to explain the association between maternal and fetal heartbeats during pregnancy. We validated the performance of this approach using a trained deep learning tool on a total of 941 one minute maternal-fetal R-peaks segments collected from 172 pregnant women (20–40 weeks). The high accuracy achieved by the tool (90%) in identifying coupling scenarios demonstrated the potential of using AI as a monitoring tool for frequent evaluation of fetal development. The interpretability of deep learning was significant in explaining synchronization mechanisms between the maternal and fetal heartbeats. This study could potentially pave the way toward the integration of automated deep learning tools in clinical practice to provide timely and continuous fetal monitoring while reducing triage, side-effects, and costs associated with current clinical devices.

Artificial intelligence-enabled electrocardiogram screens low left ventricular ejection fraction with a degree of confidence

Background

Artificial intelligence-enabled electrocardiogram has become a substitute tool for echocardiography in left ventricular ejection fraction estimation. However, the direct use of artificial intelligence-enabled electrocardiogram may be not trustable due to the uncertainty of the prediction.

Objective

The study aimed to establish an artificial intelligence-enabled electrocardiogram with a degree of confidence to identify left ventricular dysfunction.

Methods

The study collected 76,081 and 11,771 electrocardiograms from an academic medical center and a community hospital to establish and validate the deep learning model, respectively. The proposed deep learning model provided the point estimation of the actual ejection fraction and its standard deviation derived from the maximum probability density function of a normal distribution. The primary analysis focused on the accuracy of identifying patients with left ventricular dysfunction (ejection fraction ≤ 40%). Since the standard deviation was an uncertainty indicator in a normal distribution, we used it as a degree of confidence in the artificial intelligence-enabled electrocardiogram. We further explored the clinical application of estimated standard deviation and followed up on the new-onset left ventricular dysfunction in patients with initially normal ejection fraction.

Results

The area under receiver operating characteristic curves (AUC) of detecting left ventricular dysfunction were 0.9549 and 0.9365 in internal and external validation sets. After excluding the cases with a lower degree of confidence, the artificial intelligence-enabled electrocardiogram performed better in the remaining cases in internal (AUC = 0.9759) and external (AUC = 0.9653) validation sets. For the application of future left ventricular dysfunction risk stratification in patients with initially normal ejection fraction, a 4.57-fold risk of future left ventricular dysfunction when the artificial intelligence-enabled electrocardiogram is positive in the internal validation set. The hazard ratio was increased to 8.67 after excluding the cases with a lower degree of confidence. This trend was also validated in the external validation set.

Conclusion

The deep learning model with a degree of confidence can provide advanced improvements in identifying left ventricular dysfunction and serve as a decision support and management-guided screening tool for prognosis.