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

Pitfalls in Developing Machine Learning Models for Predicting Cardiovascular Diseases: Challenge and Solutions

UNSTRUCTURED

In recent years, there has been an explosive development of artificial intelligence (AI), which has been widely applied in the healthcare field. As a typical AI technology, machine learning (ML) models have emerged as great potential in predicting cardiovascular diseases (CVDs) by leveraging large amounts of medical data for training and optimization, which are expected to play a crucial role in reducing the incidence and mortality rates of CVDs. Although the field has become a research hotspot, there are still many pitfalls that researchers need to pay close attention to. These pitfalls may affect the predictive performance, credibility, reliability, reproducibility of the studied models, ultimately reducing the value of the research and affecting the prospects for clinical application. Therefore, identifying and avoiding these pitfalls is a crucial task before implementing the research. However, there is currently a lack of comprehensive summary on this topic. This viewpoint aims to analyze the existing problems in terms of data quality, dataset characteristics, model design and statistical methods as well as clinic implication, and provide possible solutions to these problems, like gathering objective data, improving training, repeating measurements, increasing sample size, preventing overfitting using statistical methods, utilizing specific AI algorithms to address targeted issues, standardizing outcomes and evaluation criteria, as well as enhancing fairness and replicability, with the goal of offering reference and assistance to researchers, algorithm developers, policy makers, and clinical practitioners.

Energy wavelet signal processed ECG and standard 12 lead ECG: Diagnosis of early diastolic dysfunction

BACKGROUND

Left ventricular (LV) diastolic dysfunction (LVDD) is the result of impaired LV relaxation and identifies those at risk of developing heart failure. Echocardiography has been used as the gold standard to identify early LVDD. The signal processed electrocardiogram (hsECG) has demonstrated effectiveness to detect early LVDD. Whether or not the standard 12‑lead electrocardiogram (ECG) can accurately predict early LVDD is not known.

METHODS

A standard 12‑lead ECG including signal processing (hsECG) was performed in 569 patients. Patients with atrial fibrillation, bundle branch block, pre-excitation, left ventricular hypertrophy or known cardiovascular disease were excluded, leaving 464 examinations for analysis. Early LVDD was diagnosed by established methods using echocardiography. Repolarization abnormalities (T wave discordance) in V1, V6, I and aVL and the hsECG were compared to the echocardiographic findings to establish diagnostic accuracy.

RESULTS

A total of 84 (18.1%) patients were diagnosed with early LVDD. A combination of a borderline or abnormal finding on the hsECG produced the best diagnostic model (sensitivity 84.5%, specificity 47.9%). The best performing ECG lead was V1 with a sensitivity of 38.1% and specificity of 92.1%. Regression analysis demonstrated increasing age and V1 to be predictive of LVDD.

CONCLUSIONS

The hsECG displayed reasonable ability to detect early LVDD. Other than V1, repolarization abnormalities on the standard 12‑lead ECG did not. While lead V1 showed promise in detecting LVDD, whether this or any other simple ECG variable can predict future LVDD would be of further interest.

Prognostic implications of heart failure stages among Chinese community populations: insight from a nationwide population-based study

Background

In light of high burden of heart failure (HF) in China, studies of prognostic implication of HF stages are important. We aimed to evaluate the relationship between HF stages and mortality risk in Chinese community populations.

Methods

Nationwide representative populations aged ≥35 years (n = 23,284, mean age 56.9 years, women 53.2%) were enrolled from 2012 to 2016. According to the international HF guidelines, participants were divided into stage A, B and C, and those who did not qualify these stages were categorized as apparently-healthy group. Association between HF stages and all-cause, cardiovascular [CV] and non-CV death was evaluated using multivariable-adjusted Cox proportional regression analysis.

Findings

During a median follow-up of 4.7 years (109,902.8 person-years), 1314 deaths occurred. Age-adjusted incidence rate of all-cause death was 5.3 in apparently-healthy, 7.8 in stage A, 8.6 in stage B and 24.6 in stage C groups per 1000 person-years. In reference to apparently-healthy group, adjusted hazard ratio for all-cause death was 1.90 (95% CI: 1.47-2.45), 2.43 (95% CI: 1.89-3.13) and 6.40 (95% CI: 4.56-8.99) for stage A, B and C. Advancing HF stages were associated with increasing risks for all-cause, CV and non-CV death (P-trend <0.05). For all-cause death, population attributable fraction due to stage A, B and C were 21.2%, 33.4% and 4.9%, accounting for 1,933,385, 3,045,993 and 446,867 deaths in China in 2018.

Interpretation

Advancing HF stages were associated with increasing risk mortality. Development and implementation of early screening and targeted interventions are urgently needed to reduce HF burdens in China.

Funding

This work was supported by the Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences (grant 2017-I2M-1-004), the Projects in the Chinese National Science & Technology Pillar Program during the Twelfth Five-year Plan Period (No.: 2011BAI11B01), and the Project Entrusted by the National Health Commission of the People's Republic of China (NHC2020-609).

Electrocardiogram-based prediction of conduction disturbances after transcatheter aortic valve replacement with convolutional neural network

Aims

Permanent pacemaker implantation and left bundle branch block are common complications after transcatheter aortic valve replacement (TAVR) and are associated with impaired prognosis. This study aimed to develop an artificial intelligence (AI) model for predicting conduction disturbances after TAVR using pre-procedural 12-lead electrocardiogram (ECG) images.

Methods and results

We collected pre-procedural 12-lead ECGs of patients who underwent TAVR at West China Hospital between March 2016 and March 2022. A hold-out testing set comprising 20% of the sample was randomly selected. We developed an AI model using a convolutional neural network, trained it using five-fold cross-validation and tested it on the hold-out testing cohort. We also developed and validated an enhanced model that included additional clinical features. After applying exclusion criteria, we included 1354 ECGs of 718 patients in the study. The AI model predicted conduction disturbances in the hold-out testing cohort with an area under the curve (AUC) of 0.764, accuracy of 0.743, F1 score of 0.752, sensitivity of 0.876, and specificity of 0.624, based solely on pre-procedural ECG images. The performance was better than the Emory score (AUC = 0.704), as well as the logistic (AUC = 0.574) and XGBoost (AUC = 0.520) models built with previously identified high-risk ECG patterns. After adding clinical features, there was an increase in the overall performance with an AUC of 0.779, accuracy of 0.774, F1 score of 0.776, sensitivity of 0.794, and specificity of 0.752.

Conclusion

Artificial intelligence-enhanced ECGs may offer better predictive value than traditionally defined high-risk ECG patterns.

Use of the energy waveform electrocardiogram to detect subclinical left ventricular dysfunction in patients with type 2 diabetes mellitus

BACKGROUND

Recent guidelines propose N-terminal pro-B-type natriuretic peptide (NT-proBNP) for recognition of asymptomatic left ventricular (LV) dysfunction (Stage B Heart Failure, SBHF) in type 2 diabetes mellitus (T2DM). Wavelet Transform based signal-processing transforms electrocardiogram (ECG) waveforms into an energy distribution waveform (ew)ECG, providing frequency and energy features that machine learning can use as additional inputs to improve the identification of SBHF. Accordingly, we sought whether machine learning model based on ewECG features was superior to NT-proBNP, as well as a conventional screening tool-the Atherosclerosis Risk in Communities (ARIC) HF risk score, in SBHF screening among patients with T2DM.

METHODS

Participants in two clinical trials of SBHF (defined as diastolic dysfunction [DD], reduced global longitudinal strain [GLS ≤ 18%] or LV hypertrophy [LVH]) in T2DM underwent 12-lead ECG with additional ewECG feature and echocardiography. Supervised machine learning was adopted to identify the optimal combination of ewECG extracted features for SBHF screening in 178 participants in one trial and tested in 97 participants in the other trial. The accuracy of the ewECG model in SBHF screening was compared with NT-proBNP and ARIC HF.

RESULTS

SBHF was identified in 128 (72%) participants in the training dataset (median 72 years, 41% female) and 64 (66%) in the validation dataset (median 70 years, 43% female). Fifteen ewECG features showed an area under the curve (AUC) of 0.81 (95% CI 0.787-0.794) in identifying SBHF, significantly better than both NT-proBNP (AUC 0.56, 95% CI 0.44-0.68, p < 0.001) and ARIC HF (AUC 0.67, 95%CI 0.56-0.79, p = 0.002). ewECG features were also led to robust models screening for DD (AUC 0.74, 95% CI 0.73-0.74), reduced GLS (AUC 0.76, 95% CI 0.73-0.74) and LVH (AUC 0.90, 95% CI 0.88-0.89).

CONCLUSIONS

Machine learning based modelling using additional ewECG extracted features are superior to NT-proBNP and ARIC HF in SBHF screening among patients with T2DM, providing an alternative HF screening strategy for asymptomatic patients and potentially act as a guidance tool to determine those who required echocardiogram to confirm diagnosis. Trial registration LEAVE-DM, ACTRN 12619001393145 and Vic-ELF, ACTRN 12617000116325.

Deep learning from latent spatiotemporal information of the heart: Identifying advanced bioimaging markers from echocardiograms

A key strength of echocardiography lies in its integration of comprehensive spatiotemporal cardiac imaging data in real-time, to aid frontline or bedside patient risk stratification and management. Nonetheless, its acquisition, processing, and interpretation are known to all be subject to heterogeneity from its reliance on manual and subjective human tracings, which challenges workflow and protocol standardization and final interpretation accuracy. In the era of advanced computational power, utilization of machine learning algorithms for big data analytics in echocardiography promises reduction in cost, cognitive errors, and intra- and inter-observer variability. Novel spatiotemporal deep learning (DL) models allow the integration of temporal arm information based on unlabeled pixel echocardiographic data for convolution of an adaptive semantic spatiotemporal calibration to construct personalized 4D heart meshes, assess global and regional cardiac function, detect early valve pathology, and differentiate uncommon cardiovascular disorders. Meanwhile, data visualization on spatiotemporal DL prediction models helps extract latent temporal imaging features to develop advanced imaging biomarkers in early disease stages and advance our understanding of pathophysiology to support the development of personalized prevention or treatment strategies. Since portable echocardiograms have been increasingly used as point-of-care imaging tools to aid rural care delivery, the application of these new spatiotemporal DL techniques show the potentials in streamlining echocardiographic acquisition, processing, and data analysis to improve workflow standardization and efficiencies, and provide risk stratification and decision supporting tools in real-time, to prompt the building of new imaging diagnostic networks to enhance rural healthcare engagement.

Adopting artificial intelligence in cardiovascular medicine: a scoping review

Recent years have witnessed significant transformations in cardiovascular medicine, driven by the rapid evolution of artificial intelligence (AI). This scoping review was conducted to capture the breadth of AI applications within cardiovascular science. Employing a structured approach, we sourced relevant articles from PubMed, with an emphasis on journals encompassing general cardiology and digital medicine. We applied filters to highlight cardiovascular articles published in journals focusing on general internal medicine, cardiology and digital medicine, thereby identifying the prevailing trends in the field. Following a comprehensive full-text screening, a total of 140 studies were identified. Over the preceding 5 years, cardiovascular medicine's interplay with AI has seen an over tenfold augmentation. This expansive growth encompasses multiple cardiovascular subspecialties, including but not limited to, general cardiology, ischemic heart disease, heart failure, and arrhythmia. Deep learning emerged as the predominant methodology. The majority of AI endeavors in this domain have been channeled toward enhancing diagnostic and prognostic capabilities, utilizing resources such as hospital datasets, electrocardiograms, and echocardiography. A significant uptrend was observed in AI's application for omics data analysis. However, a clear gap persists in AI's full-scale integration into the clinical decision-making framework. AI, particularly deep learning, has demonstrated robust applications across cardiovascular subspecialties, indicating its transformative potential in this field. As we continue on this trajectory, ensuring the alignment of technological progress with medical ethics becomes crucial. The abundant digital health data today further accentuates the need for meticulous systematic reviews, tailoring them to each cardiovascular subspecialty.

Value Creation Through Artificial Intelligence and Cardiovascular Imaging: A Scientific Statement From the American Heart Association

Multiple applications for machine learning and artificial intelligence (AI) in cardiovascular imaging are being proposed and developed. However, the processes involved in implementing AI in cardiovascular imaging are highly diverse, varying by imaging modality, patient subtype, features to be extracted and analyzed, and clinical application. This article establishes a framework that defines value from an organizational perspective, followed by value chain analysis to identify the activities in which AI might produce the greatest incremental value creation. The various perspectives that should be considered are highlighted, including clinicians, imagers, hospitals, patients, and payers. Integrating the perspectives of all health care stakeholders is critical for creating value and ensuring the successful deployment of AI tools in a real-world setting. Different AI tools are summarized, along with the unique aspects of AI applications to various cardiac imaging modalities, including cardiac computed tomography, magnetic resonance imaging, and positron emission tomography. AI is applicable and has the potential to add value to cardiovascular imaging at every step along the patient journey, from selecting the more appropriate test to optimizing image acquisition and analysis, interpreting the results for classification and diagnosis, and predicting the risk for major adverse cardiac events.

Artificial intelligence-enabled ECG for left ventricular diastolic function and filling pressure

Assessment of left ventricular diastolic function plays a major role in the diagnosis and prognosis of cardiac diseases, including heart failure with preserved ejection fraction. We aimed to develop an artificial intelligence (AI)-enabled electrocardiogram (ECG) model to identify echocardiographically determined diastolic dysfunction and increased filling pressure. We trained, validated, and tested an AI-enabled ECG in 98,736, 21,963, and 98,763 patients, respectively, who had an ECG and echocardiographic diastolic function assessment within 14 days with no exclusion criteria. It was also tested in 55,248 patients with indeterminate diastolic function by echocardiography. The model was evaluated using the area under the curve (AUC) of the receiver operating characteristic curve, and its prognostic performance was compared to echocardiography. The AUC for detecting increased filling pressure was 0.911. The AUCs to identify diastolic dysfunction grades ≥1, ≥2, and 3 were 0.847, 0.911, and 0.943, respectively. During a median follow-up of 5.9 years, 20,223 (20.5%) died. Patients with increased filling pressure predicted by AI-ECG had higher mortality than those with normal filling pressure, after adjusting for age, sex, and comorbidities in the test group (hazard ratio (HR) 1.7, 95% CI 1.645-1.757) similar to echocardiography and in the indeterminate group (HR 1.34, 95% CI 1.298-1.383). An AI-enabled ECG identifies increased filling pressure and diastolic function grades with a good prognostic value similar to echocardiography. AI-ECG is a simple and promising tool to enhance the detection of diseases associated with diastolic dysfunction and increased diastolic filling pressure.

A Novel Quantitative Electrocardiography Strategy Reveals the Electroinhibitory Effect of Tamoxifen on the Mouse Heart

Tamoxifen, a selective estrogen receptor modulator, was initially used to treat cancer in women and more recently to induce conditional gene editing in rodent hearts. However, little is known about the baseline biological effects of tamoxifen on the myocardium. In order to clarify the short-term effects of tamoxifen on cardiac electrophysiology of myocardium, we applied a single-chest-lead quantitative method and analyzed the short-term electrocardiographic phenotypes induced by tamoxifen in the heart of adult female mice. We found that tamoxifen prolonged the PP interval and caused a decreased heartbeat, and further induced atrioventricular block by gradually prolonging the PR interval. Further correlation analysis suggested that tamoxifen had a synergistic and dose-independent inhibition on the time course of the PP interval and PR interval. This prolongation of the critical time course may represent a tamoxifen-specific ECG excitatory-inhibitory mechanism, leading to a reduction in the number of supraventricular action potentials and thus bradycardia. Segmental reconstructions showed that tamoxifen induced a decrease in the conduction velocity of action potentials throughout the atria and parts of the ventricles, resulting in a flattening of the P wave and R wave. In addition, we detected the previously reported prolongation of the QT interval, which may be due to a prolonged duration of the ventricular repolarizing T wave rather than the depolarizing QRS complex. Our study highlights that tamoxifen can produce patterning alternations in the cardiac conduction system, including the formation of inhibitory electrical signals with reduced conduction velocity, implying its involvement in the regulation of myocardial ion transport and the mediation of arrhythmias. A Novel Quantitative Electrocardiography Strategy Reveals the Electroinhibitory Effect of Tamoxifen on the Mouse Heart（Figure 9）. A working model of tamoxifen producing acute electrical disturbances in the myocardium. SN, sinus node; AVN, atrioventricular node; RA, right atrium; LA, left atrium; RV, right ventricle; LV, left ventricle.

Artificial intelligence and heart failure: A state-of-the-art review

Heart failure (HF) is a heterogeneous syndrome affecting more than 60 million individuals globally. Despite recent advancements in understanding of the pathophysiology of HF, many issues remain including residual risk despite therapy, understanding the pathophysiology and phenotypes of patients with HF and preserved ejection fraction, and the challenges related to integrating a large amount of disparate information available for risk stratification and management of these patients. Risk prediction algorithms based on artificial intelligence (AI) may have superior predictive ability compared to traditional methods in certain instances. AI algorithms can play a pivotal role in the evolution of HF care by facilitating clinical decision making to overcome various challenges such as allocation of treatment to patients who are at highest risk or are more likely to benefit from therapies, prediction of adverse outcomes, and early identification of patients with subclinical disease or worsening HF. With the ability to integrate and synthesize large amounts of data with multidimensional interactions, AI algorithms can supply information with which physicians can improve their ability to make timely and better decisions. In this review, we provide an overview of the AI algorithms that have been developed for establishing early diagnosis of HF, phenotyping HF with preserved ejection fraction, and stratifying HF disease severity. This review also discusses the challenges in clinical deployment of AI algorithms in HF, and the potential path forward for developing future novel learning-based algorithms to improve HF care.

Improved delineation model of a standard 12-lead electrocardiogram based on a deep learning algorithm

BACKGROUND

Signal delineation of a standard 12-lead electrocardiogram (ECG) is a decisive step for retrieving complete information and extracting signal characteristics for each lead in cardiology clinical practice. However, it is arduous to manually assess the leads, as a variety of signal morphological variations in each lead have potential defects in recording, noise, or irregular heart rhythm/beat.

METHOD

A computer-aided deep-learning algorithm is considered a state-of-the-art delineation model to classify ECG waveform and boundary in terms of the P-wave, QRS-complex, and T-wave and indicated the satisfactory result. This study implemented convolution layers as a part of convolutional neural networks for automated feature extraction and bidirectional long short-term memory as a classifier. For beat segmentation, we have experimented beat-based and patient-based approach.

RESULTS

The empirical results using both beat segmentation approaches, with a total of 14,588 beats were showed that our proposed model performed excellently well. All performance metrics above 95% and 93%, for beat-based and patient-based segmentation, respectively.

CONCLUSIONS

This is a significant step towards the clinical pertinency of automated 12-lead ECG delineation using deep learning.

Recent Advances in Translational Research for Heart Failure in Japan

Despite decades of intensive research and therapeutic development, heart failure remains a leading cause of death worldwide. However, recent advances in several basic and translational research fields, such as genomic analysis and single-cell analysis, have increased the possibility of developing novel diagnostic approaches to heart failure. Most cardiovascular diseases that predispose individuals to heart failure are caused by genetic and environmental factors. It follows that genomic analysis can contribute to the diagnosis and prognostic stratification of patients with heart failure. In addition, single-cell analysis has shown great potential for unveiling the pathogenesis and/or pathophysiology and for discovering novel therapeutic targets for heart failure. Here, we summarize the recent advances in translational research on heart failure in Japan, based mainly on our studies.

Phonocardiogram transfer learning-based CatBoost model for diastolic dysfunction identification using multiple domain-specific deep feature fusion

Left ventricular diastolic dyfunction detection is particularly important in cardiac function screening. This paper proposed a phonocardiogram (PCG) transfer learning-based CatBoost model to detect diastolic dysfunction noninvasively. The Short-Time Fourier Transform (STFT), Mel Frequency Cepstral Coefficients (MFCCs), S-transform and gammatonegram were utilized to perform four different representations of spectrograms for learning the representative patterns of PCG signals in two-dimensional image modality. Then, four pre-trained convolutional neural networks (CNNs) such as VGG16, Xception, ResNet50 and InceptionResNetv2 were employed to extract multiple domain-specific deep features from PCG spectrograms using transfer learning, respectively. Further, principal component analysis and linear discriminant analysis (LDA) were applied to different feature subsets, respectively, and then these different selected features are fused and fed into CatBoost for classification and performance comparison. Finally, three typical machine learning classifiers such as multilayer perceptron, support vector machine and random forest were employed to compared with CatBoost. The hyperparameter optimization of the investigated models was determined through grid search. The visualized result of the global feature importance showed that deep features extracted from gammatonegram by ResNet50 contributed most to classification. Overall, the proposed multiple domain-specific feature fusion based CatBoost model with LDA achieved the best performance with an area under the curve of 0.911, accuracy of 0.882, sensitivity of 0.821, specificity of 0.927, F1-score of 0.892 on the testing set. The PCG transfer learning-based model developed in this study could aid in diastolic dysfunction detection and could contribute to non-invasive evaluation of diastolic function.

Multi-center retrospective cohort study applying deep learning to electrocardiograms to identify left heart valvular dysfunction

BACKGROUND

Aortic Stenosis and Mitral Regurgitation are common valvular conditions representing a hidden burden of disease within the population. The aim of this study was to develop and validate deep learning-based screening and diagnostic tools that can help guide clinical decision making.

METHODS

In this multi-center retrospective cohort study, we acquired Transthoracic Echocardiogram reports from five Mount Sinai hospitals within New York City representing a demographically diverse cohort of patients. We developed a Natural Language Processing pipeline to extract ground-truth labels about valvular status and paired these to Electrocardiograms (ECGs). We developed and externally validated deep learning models capable of detecting valvular disease, in addition to considering scenarios of clinical deployment.

RESULTS

We use 617,338 ECGs paired to transthoracic echocardiograms from 123,096 patients to develop a deep learning model for detection of Mitral Regurgitation. Area Under Receiver Operating Characteristic curve (AUROC) is 0.88 (95% CI:0.88-0.89) in internal testing, and 0.81 (95% CI:0.80-0.82) in external validation. To develop a model for detection of Aortic Stenosis, we use 617,338 Echo-ECG pairs for 128,628 patients. AUROC is 0.89 (95% CI: 0.88-0.89) in internal testing, going to 0.86 (95% CI: 0.85-0.87) in external validation. The model's performance increases leading up to the time of the diagnostic echo, and it performs well in validation against requirement of Transcatheter Aortic Valve Replacement procedures.

CONCLUSIONS

Deep learning based tools can increase the amount of information extracted from ubiquitous investigations such as the ECG. Such tools are inexpensive, can help in earlier disease detection, and potentially improve prognosis.

Prognostic factors and survival prediction for patients with metastatic lung adenocarcinoma: A population-based study

The prognosis of metastatic lung adenocarcinoma (MLUAD) varies greatly. At present, no studies have constructed a satisfactory prognostic model for MLUAD. We identified 44,878 patients with MLUAD. The patients were randomized into the training and validation cohorts. Cox regression models were performed to identify independent prognostic factors. Then, R software was employed to construct a new nomogram for predicting overall survival (OS) of patients with MLUAD. Accuracy was assessed by the concordance index (C-index), receiver operating characteristic curves and calibration plots. Finally, clinical practicability was examined via decision curve analysis. The OS time range for the included populations was 0 to 107 months, and the median OS was 7.00 months. Nineteen variables were significantly associated with the prognosis, and the top 5 prognostic factors were chemotherapy, grade, age, race and surgery. The nomogram has excellent predictive accuracy and clinical applicability compared to the TNM system (C-index: 0.723 vs 0.534). The C-index values were 0.723 (95% confidence interval: 0.719-0.726) and 0.723 (95% confidence interval: 0.718-0.729) in the training and validation cohorts, respectively. The area under the curve for 6-, 12-, and 18-month OS was 0.799, 0.764, and 0.750, respectively, in the training cohort and 0.799, 0.762, and 0.746, respectively, in the validation cohort. The calibration plots show good accuracy, and the decision curve analysis values indicate good clinical applicability and effectiveness. The nomogram model constructed with the above 19 prognostic factors is suitable for predicting the OS of MLUAD and has good predictive accuracy and clinical applicability.

Artificial intelligence in cardiology: Hope for the future and power for the present

Cardiovascular disease (CVD) is the principal cause of mortality and morbidity globally. With the pressures for improved care and translation of the latest medical advances and knowledge to an actionable plan, clinical decision-making for cardiologists is challenging. Artificial Intelligence (AI) is a field in computer science that studies the design of intelligent agents which take the best feasible action in a situation. It incorporates the use of computational algorithms which simulate and perform tasks that traditionally require human intelligence such as problem solving and learning. Whilst medicine is arguably the last to apply AI in its everyday routine, cardiology is at the forefront of AI revolution in the medical field. The development of AI methods for accurate prediction of CVD outcomes, non-invasive diagnosis of coronary artery disease (CAD), detection of malignant arrythmias through wearables, and diagnosis, treatment strategies and prediction of outcomes for heart failure (HF) patients, demonstrates the potential of AI in future cardiology. With the advancements of AI, Internet of Things (IoT) and the promotion of precision medicine, the future of cardiology will be heavily based on these innovative digital technologies. Despite this, ethical dilemmas regarding the implementation of AI technologies in real-world are still unaddressed.

Machine learning derived ECG risk score improves cardiovascular risk assessment in conjunction with coronary artery calcium scoring

Background

Precision estimation of cardiovascular risk remains the cornerstone of atherosclerotic cardiovascular disease (ASCVD) prevention. While coronary artery calcium (CAC) scoring is the best available non-invasive quantitative modality to evaluate risk of ASCVD, it excludes risk related to prior myocardial infarction, cardiomyopathy, and arrhythmia which are implicated in ASCVD. The high-dimensional and inter-correlated nature of ECG data makes it a good candidate for analysis using machine learning techniques and may provide additional prognostic information not captured by CAC. In this study, we aimed to develop a quantitative ECG risk score (eRiS) to predict major adverse cardiovascular events (MACE) alone, or when added to CAC. Further, we aimed to construct and validate a novel nomogram incorporating ECG, CAC and clinical factors for ASCVD.

Methods

We analyzed 5,864 patients with at least 1 cardiovascular risk factor who underwent CAC scoring and a standard ECG as part of the CLARIFY study (ClinicalTrials.gov Identifier: NCT04075162). Events were defined as myocardial infarction, coronary revascularization, stroke or death. A total of 649 ECG features, consisting of measurements such as amplitude and interval measurements from all deflections in the ECG waveform (53 per lead and 13 overall) were automatically extracted using a clinical software (GE Muse™ Cardiology Information System, GE Healthcare). The data was split into 4 training (S_tr) and internal validation (S_v) sets [S_tr (1): S_v (1): 50:50; S_tr (2): S_v (2): 60:40; S_tr (3): S_v (3): 70:30; S_tr (4): S_v (4): 80:20], and the results were compared across all the subsets. We used the ECG features derived from S_tr to develop eRiS. A least absolute shrinkage and selection operator-Cox (LASSO-Cox) regularization model was used for data dimension reduction, feature selection, and eRiS construction. A Cox-proportional hazards model was used to assess the benefit of using an eRiS alone (M_ecg), CAC alone (M_cac) and a combination of eRiS and CAC (M_ecg+cac) for MACE prediction. A nomogram (M_nom) was further constructed by integrating eRiS with CAC and demographics (age and sex). The primary endpoint of the study was the assessment of the performance of M_ecg, M_cac, M_ecg+cac and M_nom in predicting CV disease-free survival in ASCVD.

Findings

Over a median follow-up of 14 months, 494 patients had MACE. The feature selection strategy preserved only about 18% of the features that were consistent across the various strata (S_tr). The M_ecg model, comprising of eRiS alone was found to be significantly associated with MACE and had good discrimination of MACE (C-Index: 0.7, p = <2e-16). eRiS could predict time-to MACE (C-Index: 0.6, p = <2e-16 across all S_v). The M_ecg+cac model was associated with MACE (C-index: 0.71). Model comparison showed that M_ecg+cac was superior to M_ecg (p = 1.8e-10) or M_cac (p < 2.2e-16) alone. The M_nom, comprising of eRiS, CAC, age and sex was associated with MACE (C-index 0.71). eRiS had the most significant contribution, followed by CAC score and other clinical variables. Further, M_nom was able to identify unique patient risk-groups based on eRiS, CAC and clinical variables.

Conclusion

The use of ECG features in conjunction with CAC may allow for improved prognostication and identification of populations at risk. Future directions will involve prospective validation of the risk score and the nomogram across diverse populations with a heterogeneity of treatment effects.

Predicting Overall Survival in Patients with Nonmetastatic Gastric Signet Ring Cell Carcinoma: A Machine Learning Approach

Background and Aims

Accurate prediction is essential for the survival of patients with nonmetastatic gastric signet ring cell carcinoma (GSRC) and medical decision-making. Current models rely on prespecified variables, limiting their performance and not being suitable for individual patients. Our study is aimed at developing a more precise model for predicting 1-, 3-, and 5-year overall survival (OS) in patients with nonmetastatic GSRC based on a machine learning approach.

Methods

We selected 2127 GSRC patients diagnosed from 2004 to 2014 from the Surveillance, Epidemiology, and End Results (SEER) database and then randomly partitioned them into a training and validation cohort. We compared the performance of several machine learning-based models and finally chose the eXtreme gradient boosting (XGBoost) model as the optimal method to predict the OS in patients with nonmetastatic GSRC. The model was assessed using the receiver operating characteristic curve (ROC).

Results

In the training cohort, for predicting OS rates at 1-, 3-, and 5-year, the AUCs of the XGBoost model were 0.842, 0.831, and 0.838, respectively, while in the testing cohort, the AUCs of 1-, 3-, and 5-year OS rates were 0.749, 0.823, and 0.829, respectively. Besides, the XGBoost model also performed better when compared with the American Joint Committee on Cancer (AJCC) stage. The performance for this model was stably maintained when stratified by age and ethnicity.

Conclusion

The XGBoost-based model accurately predicts the 1-, 3-, and 5-year OS in patients with nonmetastatic GSRC. Machine learning is a promising way to predict the survival outcomes of tumor patients.

Left Ventricular Diastolic Dysfunction Screening by a Smartphone-Case Based on Single Lead ECG

Aims:

To investigate the potential of a signal processed by smartphone-case based on single lead electrocardiogram (ECG) for left ventricular diastolic dysfunction (LVDD) determination as a screening method.

Methods and Results:

We included 446 subjects for sample learning and 259 patients for sample test aged 39 to 74 years for testing with 2D-echocardiography, tissue Doppler imaging and ECG using a smartphone-case based single lead ECG monitor for the assessment of LVDD. Spectral analysis of ECG signals (spECG) has been used in combination with advanced signal processing and artificial intelligence methods. Wavelengths slope, time intervals between waves, amplitudes at different points of the ECG complexes, energy of the ECG signal and asymmetry indices were analyzed. The QTc interval indicated significant diastolic dysfunction with a sensitivity of 78% and a specificity of 65%, a Tpeak parameter >590 ms with 63% and 58%, a T value off >695 ms with 63% and 74%, and QRSfi > 674 ms with 74% and 57%, respectively. A combination of the threshold values from all 4 parameters increased sensitivity to 86% and specificity to 70%, respectively (OR 11.7 [2.7-50.9], P < .001). Algorithm approbation have shown: Sensitivity—95.6%, Specificity—97.7%, Diagnostic accuracy—96.5% and Repeatability—98.8%.

Conclusion:

Our results indicate a great potential of a smartphone-case based on single lead ECG as novel screening tool for LVDD if spECG is used in combination with advanced signal processing and machine learning technologies.

rECHOmmend: An ECG-based Machine-learning Approach for Identifying Patients at High-risk of Undiagnosed Structural Heart Disease Detectable by Echocardiography

Background:

Timely diagnosis of structural heart disease improves patient outcomes, yet many remain underdiagnosed. While population screening with echocardiography is impractical, ECG-based prediction models can help target high-risk patients. We developed a novel ECG-based machine learning approach to predict multiple structural heart conditions, hypothesizing that a composite model would yield higher prevalence and positive predictive values to facilitate meaningful recommendations for echocardiography.

Methods:

Using 2 232 130 ECGs linked to electronic health records and echocardiography reports from 484 765 adults between 1984 to 2021, we trained machine learning models to predict the presence or absence of any of 7 echocardiography-confirmed diseases within 1 year. This composite label included the following: moderate or severe valvular disease (aortic/mitral stenosis or regurgitation, tricuspid regurgitation), reduced ejection fraction <50%, or interventricular septal thickness >15 mm. We tested various combinations of input features (demographics, laboratory values, structured ECG data, ECG traces) and evaluated model performance using 5-fold cross-validation, multisite validation trained on 1 site and tested on 10 independent sites, and simulated retrospective deployment trained on pre-2010 data and deployed in 2010.

Results:

Our composite rECHOmmend model used age, sex, and ECG traces and had a 0.91 area under the receiver operating characteristic curve and a 42% positive predictive value at 90% sensitivity, with a composite label prevalence of 17.9%. Individual disease models had area under the receiver operating characteristic curves from 0.86 to 0.93 and lower positive predictive values from 1% to 31%. Area under the receiver operating characteristic curves for models using different input features ranged from 0.80 to 0.93, increasing with additional features. Multisite validation showed similar results to cross-validation, with an aggregate area under the receiver operating characteristic curve of 0.91 across our independent test set of 10 clinical sites after training on a separate site. Our simulated retrospective deployment showed that for ECGs acquired in patients without preexisting structural heart disease in the year 2010, 11% were classified as high risk and 41% (4.5% of total patients) developed true echocardiography-confirmed disease within 1 year.

Conclusions:

An ECG-based machine learning model using a composite end point can identify a high-risk population for having undiagnosed, clinically significant structural heart disease while outperforming single-disease models and improving practical utility with higher positive predictive values. This approach can facilitate targeted screening with echocardiography to improve underdiagnosis of structural heart disease.

Can echocardiographic assessment of diastolic function be automated?

Echocardiographic evaluation of left ventricular diastolic function relies on a multi-pronged algorithm, which incorporates Doppler-based and volumetric parameters. Integration of clinical data in diastolic assessment is recommended, though not clearly outlined. We sought to develop an automated tool for diastolic function, compare its performance to human-generated diagnoses and identify the common sources of error. Our software tool is based on the 2016 diastolic guidelines algorithm, which uses 8 parameters as input, with 10 conditions as the logic and 5 possible outputs as final diagnoses. Initially, we prospectively studied 563 patients whose diastolic function was independently evaluated by an expert echocardiographer and by the automated tool. Incongruent cases were further analyzed, after which features of myocardial disease were integrated into a refined version of the software that was tested in an independent cohort of 1106 patients. In the initial analysis, 202/563 grades (36%) were incongruent between the automated and human reads, with the highest rate of discordance for mild and indeterminate categories. In 17% of cases, human diagnoses differed from that dictated by the algorithm due to integration of clinical factors. Follow-up analysis using the refined automated tool did not improve the discordance rate (440/1106; 40%). There was more discordance in cases of: age > 40 years, impaired mitral inflow patterns (E/A < 0.8) and reduced mitral e' values. Further analysis revealed differences in how readers interpreted the interaction between these factors and diastolic function, which could not be incorporated into the automated tool. In conclusion, although assessment of diastolic function relies on an algorithm that can be automated, this algorithm does not include clear guidance on how to incorporate age, or age-related changes in Doppler-based parameters, often resulting in discordant diagnoses. Standardized interpretation of these factors is needed to improve the reproducibility of diastolic function grading by human readers and the accuracy of the automated classification.

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.

Machine Learning to Calculate Heparin Dose in COVID-19 Patients with Active Cancer

To realize a machine learning (ML) model to estimate the dose of low molecular weight heparin to be administered, preventing thromboembolism events in COVID-19 patients with active cancer. Methods: We used a dataset comprising 131 patients with active cancer and COVID-19. We considered five ML models: logistic regression, decision tree, random forest, support vector machine and Gaussian naive Bayes. We decided to implement the logistic regression model for our study. A model with 19 variables was analyzed. Data were randomly split into training (70%) and testing (30%) sets. Model performance was assessed by confusion matrix metrics on the testing data for each model as positive predictive value, sensitivity and F1-score. Results: We showed that the five selected models outperformed classical statistical methods of predictive validity and logistic regression was the most effective, being able to classify with an accuracy of 81%. The most relevant result was finding a patient-proof where python function was able to obtain the exact dose of low weight molecular heparin to be administered and thereby to prevent the occurrence of VTE. Conclusions: The world of machine learning and artificial intelligence is constantly developing. The identification of a specific LMWH dose for preventing VTE in very high-risk populations, such as the COVID-19 and active cancer population, might improve with the use of new training ML-based algorithms. Larger studies are needed to confirm our exploratory results.

Electrocardiographic Features of Left Ventricular Diastolic Dysfunction and Heart Failure With Preserved Ejection Fraction: A Systematic Review

Background: Electrocardiographic features are well-known for heart failure with reduced ejection fraction (HFrEF), but not for left ventricular diastolic dysfunction (LVDD) and heart failure with preserved ejection fraction (HFpEF). As ECG features could help to identify high-risk individuals in primary care, we systematically reviewed the literature for ECG features diagnosing women and men suspected of LVDD and HFpEF. Methods and Results: Among the 7,127 records identified, only 10 studies reported diagnostic measures, of which 9 studied LVDD. For LVDD, the most promising features were T-end-P/(PQ*age), which is the electrocardiographic equivalent of the passive-to-active filling (AUC: 0.91-0.96), and repolarization times (QTc interval ≥ 350 ms, AUC: 0.85). For HFpEF, the Cornell product ≥ 1,800 mm*ms showed poor sensitivity of 40% (AUC: 0.62). No studies presented results stratified by sex. Conclusion: Electrocardiographic features are not widely evaluated in diagnostic studies for LVDD and HFpEF. Only for LVDD, two ECG features related to the diastolic interval, and repolarization measures showed diagnostic potential. To improve diagnosis and care for women and men suspected of heart failure, reporting of sex-specific data on ECG features is encouraged.

Using Deep-Learning Algorithms to Simultaneously Identify Right and Left Ventricular Dysfunction From the Electrocardiogram

OBJECTIVES

This study sought to develop DL models capable of comprehensively quantifying left and right ventricular dysfunction from ECG data in a large, diverse population.

BACKGROUND

Rapid evaluation of left and right ventricular function using deep learning (DL) on electrocardiograms (ECGs) can assist diagnostic workflow. However, DL tools to estimate right- ventricular (RV) function do not exist, whereas those to estimate left ventricular (LV) function are restricted to quantification of very low LV function only.

METHODS

A multicenter study was conducted with data from 5 New York City hospitals: 4 for internal testing and 1 serving as external validation. We created novel DL models to classify left ventricular ejection fraction (LVEF) into categories derived from the latest universal definition of heart failure, estimate LVEF through regression, and predict a composite outcome of either RV systolic dysfunction or RV dilation.

RESULTS

We obtained echocardiogram LVEF estimates for 147,636 patients paired to 715,890 ECGs. We used natural language processing (NLP) to extract RV size and systolic function information from 404,502 echocardiogram reports paired to 761,510 ECGs for 148,227 patients. For LVEF classification in internal testing, area under curve (AUC) at detection of LVEF ≤40%, 40% < LVEF ≤50%, and LVEF >50% was 0.94 (95% CI: 0.94-0.94), 0.82 (95% CI: 0.81-0.83), and 0.89 (95% CI: 0.89-0.89), respectively. For external validation, these results were 0.94 (95% CI: 0.94-0.95), 0.73 (95% CI: 0.72-0.74), and 0.87 (95% CI: 0.87-0.88). For regression, the mean absolute error was 5.84% (95% CI: 5.82%-5.85%) for internal testing and 6.14% (95% CI: 6.13%-6.16%) in external validation. For prediction of the composite RV outcome, AUC was 0.84 (95% CI: 0.84-0.84) in both internal testing and external validation.

CONCLUSIONS

DL on ECG data can be used to create inexpensive screening, diagnostic, and predictive tools for both LV and RV dysfunction. Such tools may bridge the applicability of ECGs and echocardiography and enable prioritization of patients for further interventions for either sided failure progressing to biventricular disease.

Spatio-temporal hybrid neural networks reduce erroneous human "judgement calls" in the diagnosis of Takotsubo syndrome

BACKGROUND

We investigate whether deep learning (DL) neural networks can reduce erroneous human "judgment calls" on bedside echocardiograms and help distinguish Takotsubo syndrome (TTS) from anterior wall ST segment elevation myocardial infarction (STEMI).

METHODS

We developed a single-channel (DCNN[2D SCI]), a multi-channel (DCNN[2D MCI]), and a 3-dimensional (DCNN[2D+t]) deep convolution neural network, and a recurrent neural network (RNN) based on 17,280 still-frame images and 540 videos from 2-dimensional echocardiograms in 10 years (1 January 2008 to 1 January 2018) retrospective cohort in University of Iowa (UI) and eight other medical centers. Echocardiograms from 450 UI patients were randomly divided into training and testing sets for internal training, testing, and model construction. Echocardiograms of 90 patients from the other medical centers were used for external validation to evaluate the model generalizability. A total of 49 board-certified human readers performed human-side classification on the same echocardiography dataset to compare the diagnostic performance and help data visualization.

FINDINGS

The DCNN (2D SCI), DCNN (2D MCI), DCNN(2D+t), and RNN models established based on UI dataset for TTS versus STEMI prediction showed mean diagnostic accuracy 73%, 75%, 80%, and 75% respectively, and mean diagnostic accuracy of 74%, 74%, 77%, and 73%, respectively, on the external validation. DCNN(2D+t) (area under the curve [AUC] 0·787 vs. 0·699, P = 0·015) and RNN models (AUC 0·774 vs. 0·699, P = 0·033) outperformed human readers in differentiating TTS and STEMI by reducing human erroneous judgement calls on TTS.

INTERPRETATION

Spatio-temporal hybrid DL neural networks reduce erroneous human "judgement calls" in distinguishing TTS from anterior wall STEMI based on bedside echocardiographic videos.

FUNDING

University of Iowa Obermann Center for Advanced Studies Interdisciplinary Research Grant, and Institute for Clinical and Translational Science Grant. National Institutes of Health Award (1R01EB025018-01).

Automated ECG Interpretation—A Brief History from High Expectations to Deepest Networks

This article traces the development of automated electrocardiography from its beginnings in Washington, DC around 1960 through to its current widespread application worldwide. Changes in the methodology of recording ECGs in analogue form using sizeable equipment through to digital recording, even in wearables, are included. Methods of analysis are considered from single lead to three leads to twelve leads. Some of the influential figures are mentioned while work undertaken locally is used to outline the progress of the technique mirrored in other centres. Applications of artificial intelligence are also considered so that the reader can find out how the field has been constantly evolving over the past 50 years.

Validation of the 2016 ASE/EACVI Guideline for Diastolic Dysfunction in Patients With Unexplained Dyspnea and a Preserved Left Ventricular Ejection Fraction

Background

Echocardiography is considered the cornerstone of the diagnostic workup of heart failure with preserved ejection fraction. Thus far, validation of the 2016 American Society of Echocardiography/European Association of Cardiovascular Imaging (ASE/EACVI) echo‐algorithm for evaluation of diastolic (dys)function in a patient suspected of heart failure with preserved ejection fraction has been limited.

Methods and Results

The diagnostic performance of the 2016 ASE/EACVI algorithm was assessed in 204 patients evaluated for unexplained dyspnea or pulmonary hypertension with echocardiogram and right heart catheterization. Invasively measured pulmonary capillary wedge pressure (PCWP) was used as the gold standard. In addition, the diagnostic performance of H₂FPEF score and NT‐proBNP (N‐terminal pro‐B‐type natriuretic peptide) were evaluated. There was a poor correlation between indexed left atrial volume, E/e′ (septal and average) or early mitral inflow (E), and PCWP (r=0.25–0.30, P values all <0.01). No correlation was found in our cohort between e′ (septal or lateral) or tricuspid valve regurgitation and PCWP. The correlation between diastolic function grades of the ASE/EACVI algorithm and PCWP was poor (r=0.17, P<0.05). The ASE/EACVI algorithm had a sensitivity and specificity of 35% and 87%, respectively; an accuracy of 67% and an area under the curve of 0.56. Moreover, in 30% of cases the algorithm was not applicable or indeterminate. H₂FPEF score had a modest correlation with PCWP (r=0.44, P<0.0001), and accuracy was 73%; NT‐proBNP correlated weakly with PCWP (r=0.24, P<0.001), and accuracy was 57%.

Conclusions

The 2016 ASE/EACVI algorithm for the assessment of diastolic function has a limited diagnostic accuracy in patients evaluated for unexplained dyspnea and/or pulmonary hypertension, and especially sensitivity to detect diastolic dysfunction was low.

Deep learning detects heart failure with preserved ejection fraction using a baseline electrocardiogram

Aims

Heart failure with preserved ejection fraction (HFpEF) is a rapidly growing global health problem. To date, diagnosis of HFpEF is based on clinical, invasive, and laboratory examinations. Electrocardiographic findings may vary, and there are no known typical ECG features for HFpEF.

Methods and results

This study included two patient cohorts. In the derivation cohort, we included n = 1884 patients who presented with exertional dyspnoea or equivalent and preserved ejection fraction (≥50%) and clinical suspicion for coronary artery disease. The ECGs were divided in segments, yielding a total of 77 558 samples. We trained a convolutional neural network (CNN) to classify HFpEF and control patients according to European Society of Cardiology (ESC) criteria. An external group of 203 volunteers in a prospective heart failure screening programme served as a validation cohort of the CNN. The external validation of the CNN yielded an area under the curve of 0.80 [95% confidence interval (CI) 0.74-0.86] for detection of HFpEF according to ESC criteria, with a sensitivity of 0.99 (95% CI 0.98-0.99) and a specificity of 0.60 (95% CI 0.56-0.64), with a positive predictive value of 0.68 (95%CI 0.64-0.72) and a negative predictive value of 0.98 (95% CI 0.95-0.99).

Conclusion

In this study, we report the first deep learning-enabled CNN for identifying patients with HFpEF according to ESC criteria including NT-proBNP measurements in the diagnostic algorithm among patients at risk. The suitability of the CNN was validated on an external validation cohort of patients at risk for developing heart failure, showing a convincing screening performance.

The Use of Machine Learning for the Care of Hypertension and Heart Failure

Machine learning (ML) is a branch of artificial intelligence that combines computer science, statistics, and decision theory to learn complex patterns from voluminous data. In the last decade, accumulating evidence has shown the utility of ML for prediction, diagnosis, and classification of hypertension and heart failure (HF). In addition, ML-enabled image analysis has potential value in assessing cardiac structure and function in an accurate, scalable, and efficient way. Considering the high burden of hypertension and HF in China and worldwide, ML may help address these challenges from different aspects. Indeed, prior studies have shown that ML can enhance each stage of patient care, from research and development, to daily clinical practice and population health. Through reviewing the published literature, the aims of the current systemic review are to summarize the utilities of ML for the care of those with hypertension and HF.

Deep learning analysis of resting electrocardiograms for the detection of myocardial dysfunction, hypertrophy, and ischaemia: a systematic review

The aim of this review was to assess the evidence for deep learning (DL) analysis of resting electrocardiograms (ECGs) to predict structural cardiac pathologies such as left ventricular (LV) systolic dysfunction, myocardial hypertrophy, and ischaemic heart disease. A systematic literature search was conducted to identify published original articles on end-to-end DL analysis of resting ECG signals for the detection of structural cardiac pathologies. Studies were excluded if the ECG was acquired by ambulatory, stress, intracardiac, or implantable devices, and if the pathology of interest was arrhythmic in nature. After duplicate reviewers screened search results, 12 articles met the inclusion criteria and were included. Three articles used DL to detect LV systolic dysfunction, achieving an area under the curve (AUC) of 0.89-0.93 and an accuracy of 98%. One study used DL to detect LV hypertrophy, achieving an AUC of 0.87 and an accuracy of 87%. Six articles used DL to detect acute myocardial infarction, achieving an AUC of 0.88-1.00 and an accuracy of 83-99.9%. Two articles used DL to detect stable ischaemic heart disease, achieving an accuracy of 95-99.9%. Deep learning models, particularly those that used convolutional neural networks, outperformed rules-based models and other machine learning models. Deep learning is a promising technique to analyse resting ECG signals for the detection of structural cardiac pathologies, which has clinical applicability for more effective screening of asymptomatic populations and expedited diagnostic work-up of symptomatic patients at risk for cardiovascular disease.

Patient Phenotypes and SGLT-2 Inhibition in Type 2 Diabetes: Insights From the EMPA-REG OUTCOME Trial

OBJECTIVES

Using latent class analysis (LCA) of EMPA-REG OUTCOME (BI 10773 [Empagliflozin] Cardiovascular Outcome Event Trial in Type 2 Diabetes Mellitus Patients), this study identified distinct phenotypes in subjects with type 2 diabetes (T2D) and cardiovascular (CV) disease and explored treatment effects across phenotypes.

BACKGROUND

In the EMPA-REG OUTCOME trial, empagliflozin reduced risk of CV death or hospitalization for heart failure (HHF) by 34% in subjects with T2D and CV disease. Among such subjects, there has been limited evaluation of clinical phenotypes.

METHODS

Overall, 7,020 participants were treated with empagliflozin 25 mg, 10 mg, or placebo. For this post hoc analysis, participants were randomly separated into training (two-thirds of patients) and validation (remaining one-third) sets. LCA identified 3 phenotype groups (n = 6,639 with complete data). The phenotype association with CV death or HHF and empagliflozin treatment effect across groups was explored by Cox regression (in training and validation sets).

RESULTS

In the training set, phenotype group 1 (n = 1,463; 33.1%) included younger patients with shorter T2D duration and the highest estimated glomerular filtration rate (eGFR). Phenotype group 2 (n = 1,172; 26.5%) included more women with non-coronary artery disease. Phenotype group 3 (n = 1,785; 40.4%) included older patients with advanced coronary disease and the lowest eGFR. The risk of CV death varied across phenotypes (group 2 vs. 1: hazard ratio [HR]; 1.83; 95% confidence interval [CI]: 1.23 to 2.71; group 3 vs. 1: HR: 1.86; 95% CI: 1.30 to 2.67) with similar patterns for CV death or HHF. Consistent treatment effects of empagliflozin were seen across phenotypes in the training and validation sets (interaction p > 0.30).

CONCLUSIONS

Among participants with T2D, this study identified 3 phenotypes with varying CV risk. The treatment effect across phenotypes reaffirms the robustness of CV death or HHF reduction with empagliflozin. (BI 10773 [Empagliflozin] Cardiovascular Outcome Event Trial in Type 2 Diabetes Mellitus Patients [EMPA-REG OUTCOME]; NCT01131676).

Machine Learning of ECG Waveforms to Improve Selection for Testing for Asymptomatic Left Ventricular Dysfunction Prompt

OBJECTIVES

To identify whether machine learning from processing of continuous wave transforms (CWTs) to provide an "energy waveform" electrocardiogram (ewECG) could be integrated with echocardiographic assessment of subclinical systolic and diastolic left ventricular dysfunction (LVD).

BACKGROUND

Asymptomatic LVD has management implications, but routine echocardiography is not undertaken in subjects at risk of heart failure. Signal processing of the surface ECG with the use of CWT can identify abnormal myocardial relaxation.

METHODS

EwECG and echocardiography were undertaken in 398 participants at risk of heart failure (HF). Reduced global longitudinal strain (GLS ≤16%)), diastolic abnormalities (E/e' >15, left atrial enlargement with E/e' >10 or impaired relaxation) or LV hypertrophy defined LVD. EwECG feature selection and supervised machine-learning by random forest (RF) classifier was undertaken with 643 CWT-derived features and the Atherosclerosis Risk in Communities (ARIC) heart failure risk score.

RESULTS

The ARIC score and 18 CWT features were selected to build a RF predictive model for LVD in a training dataset (n = 287; 60% female, median age 71 [interquartile range: 68 to 74] years). Model performance was tested in an independent group (n = 111; 49% female, median age 61 years [59 to 66 years]), demonstrating 85% sensitivity and 72% specificity (area under the receiver-operating characteristic curve [AUC]: 0.83; 95% confidence interval [CI]: 0.74 to 0.92). With ARIC score removed, sensitivity was 88% and specificity, 70% (AUC: 0.78; 95% CI: 0.70 to 0.86). RF models for reduced GLS and diastolic abnormalities including similar features had sensitivities that were unsuitable for screening. Conventional candidates for LVD screening (ARIC score, N-terminal pro-B-type natriuretic peptide, and standard automated ECG analysis) had inferior discriminative ability. Integration of ewECG in screening of people at risk of HF would reduce need for echocardiography by 45% while missing 12% of LVD cases.

CONCLUSIONS

Machine learning applied to ewECG is a sensitive screening test for LVD, and its integration into screening of patients at risk for HF would reduce the number of echocardiograms by almost one-half.

Deep-Learning Models for the Echocardiographic Assessment of Diastolic Dysfunction

OBJECTIVES

The authors explored a deep neural network (DeepNN) model that integrates multidimensional echocardiographic data to identify distinct patient subgroups with heart failure with preserved ejection fraction (HFpEF).

BACKGROUND

The clinical algorithms for phenotyping the severity of diastolic dysfunction in HFpEF remain imprecise.

METHODS

The authors developed a DeepNN model to predict high- and low-risk phenogroups in a derivation cohort (n = 1,242). Model performance was first validated in 2 external cohorts to identify elevated left ventricular filling pressure (n = 84) and assess its prognostic value (n = 219) in patients with varying degrees of systolic and diastolic dysfunction. In 3 National Heart, Lung, and Blood Institute-funded HFpEF trials, the clinical significance of the model was further validated by assessing the relationships of the phenogroups with adverse clinical outcomes (TOPCAT [Aldosterone Antagonist Therapy for Adults With Heart Failure and Preserved Systolic Function] trial, n = 518), cardiac biomarkers, and exercise parameters (NEAT-HFpEF [Nitrate's Effect on Activity Tolerance in Heart Failure With Preserved Ejection Fraction] and RELAX-HF [Evaluating the Effectiveness of Sildenafil at Improving Health Outcomes and Exercise Ability in People With Diastolic Heart Failure] pooled cohort, n = 346).

RESULTS

The DeepNN model showed higher area under the receiver-operating characteristic curve than 2016 American Society of Echocardiography guideline grades for predicting elevated left ventricular filling pressure (0.88 vs. 0.67; p = 0.01). The high-risk (vs. low-risk) phenogroup showed higher rates of heart failure hospitalization and/or death, even after adjusting for global left ventricular and atrial longitudinal strain (hazard ratio [HR]: 3.96; 95% confidence interval [CI]: 1.24 to 12.67; p = 0.021). Similarly, in the TOPCAT cohort, the high-risk (vs. low-risk) phenogroup showed higher rates of heart failure hospitalization or cardiac death (HR: 1.92; 95% CI: 1.16 to 3.22; p = 0.01) and higher event-free survival with spironolactone therapy (HR: 0.65; 95% CI: 0.46 to 0.90; p = 0.01). In the pooled RELAX-HF/NEAT-HFpEF cohort, the high-risk (vs. low-risk) phenogroup had a higher burden of chronic myocardial injury (p < 0.001), neurohormonal activation (p < 0.001), and lower exercise capacity (p = 0.001).

CONCLUSIONS

This publicly available DeepNN classifier can characterize the severity of diastolic dysfunction and identify a specific subgroup of patients with HFpEF who have elevated left ventricular filling pressures, biomarkers of myocardial injury and stress, and adverse events and those who are more likely to respond to spironolactone.

Machine learning: at the heart of failure diagnosis

PURPOSE OF REVIEW

Refinement in machine learning (ML) techniques and approaches has rapidly expanded artificial intelligence applications for the diagnosis and classification of heart failure (HF). This review is designed to provide the clinician with the basics of ML, as well as this technologies future utility in HF diagnosis and the potential impact on patient outcomes.

RECENT FINDINGS

Recent studies applying ML methods to unique data sets available from electrocardiography, vectorcardiography, echocardiography, and electronic health records show significant promise for improving diagnosis, enhancing detection, and advancing treatment of HF. Innovations in both supervised and unsupervised methods have heightened the diagnostic accuracy of models developed to identify the presence of HF and further augmentation of model capabilities are likely utilizing ensembles of ML algorithms derived from different techniques.

SUMMARY

This article is an overview of recent applications of ML to achieve improved diagnosis of HF and the resultant implications for patient management.

Detection of abnormal left ventricular geometry in patients without cardiovascular disease through machine learning: An ECG-based approach

Cardiac remodeling is recognized as an important aspect of cardiovascular disease (CVD) progression. Machine learning (ML) techniques were applied to basic clinical parameters and electrocardiographic features, in order to detect abnormal left ventricular geometry (LVG) even before the onset of left ventricular hypertrophy (LVH), in a population without established CVD. The authors enrolled 528 patients with and without essential hypertension, but no other indications of CVD. All patients underwent a full echocardiographic evaluation and were classified into 3 groups; normal geometry (NG), concentric remodeling without LVH (CR), and LVH. Abnormal LVG was identified as increased relative wall thickness (RWT) and/or left ventricular mass index (LVMi). The authors trained supervised ML models to classify patients with abnormal LVG and calculated SHAP values to perform feature importance and interaction analysis. Hypertension, age, body mass index over the Sokolow‐Lyon voltage, QRS‐T angle, and QTc duration were some of the most important features. Our model was able to distinguish NG from CR+LVH combined, with 87% accuracy on an unseen test set, 75% specificity, 97% sensitivity, and area under the receiver operating curve (AUC/ROC) equal to 0.91. The authors also trained our model to classify NG and CR (NG + CR) against those with LVH, with 89% test set accuracy, 93% specificity, 67% sensitivity, and an AUC/ROC value of 0.89, for a 0.4 decision threshold. Our ML algorithm effectively detects abnormal LVG even at early stages. Innovative solutions are needed to improve risk stratification of patients without established CVD, and ML may enable progress in this direction.

Artificial intelligence assessment for early detection of heart failure with preserved ejection fraction based on electrocardiographic features

Aims

Although heart failure with preserved ejection fraction (HFpEF) is a rapidly emerging global health problem, an adequate tool to screen it reliably and economically does not exist. We developed an interpretable deep learning model (DLM) using electrocardiography (ECG) and validated its performance.

Methods and results

This retrospective cohort study included two hospitals. 34 103 patients who underwent echocardiography and ECG within 1 week and indicated normal left ventricular systolic function were included in this study. A DLM based on an ensemble neural network was developed using 32 671 ECGs of 20 169 patients. The internal validation included 1979 ECGs of 1979 patients. Furthermore, we conducted an external validation with 11 955 ECGs of 11 955 patients from another hospital. The endpoint was to detect HFpEF. During the internal and external validation, the area under the receiver operating characteristic curves of a DLM using 12-lead ECG for detecting HFpEF were 0.866 (95% confidence interval 0.850-0.883) and 0.869 (0.860-0.877), respectively. In the 1412 individuals without HFpEF at initial echocardiography, patients whose DLM was defined as having a higher risk had a significantly higher chance of developing HFpEF than those in the low-risk group (33.6% vs. 8.4%, P < 0.001). Sensitivity map showed that the DLM focused on the QRS complex and T-wave.

Conclusion

The DLM demonstrated high performance for HFpEF detection using not only a 12-lead ECG but also 6- single-lead ECG. These results suggest that HFpEF can be screened using conventional ECG devices and diverse life-type ECG machines employing the DLM, thereby preventing disease progression.

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 &lt; 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.