Multimodality Risk Assessment of Patients with Ischemic Heart Disease Using Deep Learning Models Applied to Electrocardiograms and Chest X-rays.

Comprehensive management approaches for patients with ischemic heart disease (IHD) are important aids for prognostication and treatment planning. While single-modality deep neural networks (DNNs) have shown promising performance for detecting cardiac abnormalities, the potential benefits of using DNNs for multimodality risk assessment in patients with IHD have not been reported. The purpose of this study was to investigate the effectiveness of multimodality risk assessment in patients with IHD using a DNN that utilizes 12-lead electrocardiograms (ECGs) and chest X-rays (CXRs), with the prediction of major adverse cardiovascular events (MACEs) being of particular concern.DNN models were applied to detection of left ventricular systolic dysfunction (LVSD) on ECGs and identification of cardiomegaly findings on CXRs. A total of 2107 patients who underwent elective percutaneous coronary intervention were categorized into 4 groups according to the models' outputs: Dual-modality high-risk (n = 105), ECG high-risk (n = 181), CXR high-risk (n = 392), and No-risk (n = 1,429).A total of 342 MACEs were observed. The incidence of a MACE was the highest in the Dual-modality high-risk group (P < 0.001). Multivariate Cox hazards analysis for predicting MACE revealed that the Dual-modality high-risk group had a significantly higher risk of MACE than the No-risk group (hazard ratio (HR): 2.370, P < 0.001), the ECG high-risk group (HR: 1.906, P = 0.010), and the CXR high-risk group (HR: 1.624, P = 0.018), after controlling for confounding factors.The results suggest the usefulness of multimodality risk assessment using DNN models applied to 12-lead ECG and CXR data from patients with IHD.

Deep Learning Models for Predicting Left Heart Abnormalities From Single-Lead Electrocardiogram for the Development of Wearable Devices.

BACKGROUND: Left heart abnormalities are risk factors for heart failure. However, echocardiography is not always available. Electrocardiograms (ECGs), which are now available from wearable devices, have the potential to detect these abnormalities. Nevertheless, whether a model can detect left heart abnormalities from single Lead I ECG data remains unclear.Methods and Results: We developed Lead I ECG models to detect low ejection fraction (EF), wall motion abnormality, left ventricular hypertrophy (LVH), left ventricular dilatation, and left atrial dilatation. We used a dataset comprising 229,439 paired sets of ECG and echocardiography data from 8 facilities, and validated the model using external verification with data from 2 facilities. The area under the receiver operating characteristic curves of our model was 0.913 for low EF, 0.832 for wall motion abnormality, 0.797 for LVH, 0.838 for left ventricular dilatation, and 0.802 for left atrial dilatation. In interpretation tests with 12 cardiologists, the accuracy of the model was 78.3% for low EF and 68.3% for LVH. Compared with cardiologists who read the 12-lead ECGs, the model's performance was superior for LVH and similar for low EF. CONCLUSIONS: From a multicenter study dataset, we developed models to predict left heart abnormalities using Lead I on the ECG. The Lead I ECG models show superior or equivalent performance to cardiologists using 12-lead ECGs.

Clinical valve thrombosis and arterial embolism in a cancer patient after transcatheter aortic valve replacement.

The number of cancer patients with severe aortic stenosis and atrial fibrillation (AF) is increasing in the aging population. Transcatheter aortic valve replacement (TAVR) is an established treatment option for severe aortic stenosis with high surgical risk, including individuals with cancer. Antithrombotic therapy should be considered for post-TAVR or AF patients. However, antithrombotic management in cancer patients remains challenging due to the increased risk of both thromboembolism and bleeding. We present a case of clinical valve thrombosis and arterial embolism after transcatheter aortic valve replacement in an elderly patient with a history of metastatic pancreatic cancer and permanent atrial fibrillation under treatment of single antiplatelet therapy. Warfarin treatment after successful surgical thrombectomy to the occluded arteries improved clinical valve thrombosis, although the long-term outcome remains unclear. This case demonstrates that novel management algorithms for thromboembolism and bleeding in elderly cancer patients with AF and valvular heart disease are urgently needed.

An explainable artificial intelligence-enabled electrocardiogram analysis model for the classification of reduced left ventricular function.

Aims: The black box nature of artificial intelligence (AI) hinders the development of interpretable AI models that are applicable in clinical practice. We aimed to develop an AI model for classifying patients of reduced left ventricular ejection fraction (LVEF) from 12-lead electrocardiograms (ECG) with the decision-interpretability. Methods and results: We acquired paired ECG and echocardiography datasets from the central and co-operative institutions. For the central institution dataset, a random forest model was trained to identify patients with reduced LVEF among 29 907 ECGs. Shapley additive explanations were applied to 7196 ECGs. To extract the model's decision criteria, the calculated Shapley additive explanations values were clustered for 192 non-paced rhythm patients in which reduced LVEF was predicted. Although the extracted criteria were different for each cluster, these criteria generally comprised a combination of six ECG findings: negative T-wave inversion in I/V5-6 leads, low voltage in I/II/V4-6 leads, Q wave in V3-6 leads, ventricular activation time prolongation in I/V5-6 leads, S-wave prolongation in V2-3 leads, and corrected QT interval prolongation. Similarly, for the co-operative institution dataset, the extracted criteria comprised a combination of the same six ECG findings. Furthermore, the accuracy of seven cardiologists' ECG readings improved significantly after watching a video explaining the interpretation of these criteria (before, 62.9% ± 3.9% vs. after, 73.9% ± 2.4%; P = 0.02). Conclusion: We visually interpreted the model's decision criteria to evaluate its validity, thereby developing a model that provided the decision-interpretability required for clinical application.

Age prediction from coronary angiography using a deep neural network: Age as a potential label to extract prognosis-related imaging features.

Coronary angiography (CAG) is still considered the reference standard for coronary artery assessment, especially in the treatment of acute coronary syndrome (ACS). Although aging causes changes in coronary arteries, the age-related imaging features on CAG and their prognostic relevance have not been fully characterized. We hypothesized that a deep neural network (DNN) model could be trained to estimate vascular age only using CAG and that this age prediction from CAG could show significant associations with clinical outcomes of ACS. A DNN was trained to estimate vascular age using ten separate frames from each of 5,923 CAG videos from 572 patients. It was then tested on 1,437 CAG videos from 144 patients. Subsequently, 298 ACS patients who underwent percutaneous coronary intervention (PCI) were analysed to assess whether predicted age by DNN was associated with clinical outcomes. Age predicted as a continuous variable showed mean absolute error of 4 years with R squared of 0.72 (r = 0.856). Among the ACS patients stratified by predicted age from CAG images before PCI, major adverse cardiovascular events (MACE) were more frequently observed in the older vascular age group than in the younger vascular age group (p = 0.017). Furthermore, after controlling for actual age, gender, peak creatine kinase, and history of heart failure, the older vascular age group independently suffered from more MACE (hazard ratio 2.14, 95% CI 1.07 to 4.29, p = 0.032). The vascular age estimated based on CAG imaging by DNN showed high predictive value. The age predicted from CAG images by DNN could have significant associations with clinical outcomes in patients with ACS.

Automatic Detection of Left Ventricular Dilatation and Hypertrophy from Electrocardiograms Using Deep Learning.

Left ventricular dilatation (LVD) and left ventricular hypertrophy (LVH) are risk factors for heart failure, and their detection improves heart failure screening. This study aimed to investigate the ability of deep learning to detect LVD and LVH from a 12-lead electrocardiogram (ECG). Using ECG and echocardiographic data, we developed deep learning and machine learning models to detect LVD and LVH. We also examined conventional ECG criteria for the diagnosis of LVH. We calculated the area under the receiver operating characteristic (AUROC) curve, sensitivity, specificity, and accuracy of each model and compared the performance of the models. We analyzed data for 18,954 patients (mean age (standard deviation): 64.2 (16.5) years, men: 56.7%). For the detection of LVD, the value (95% confidence interval) of the AUROC was 0.810 (0.801-0.819) for the deep learning model, and this was significantly higher than that of the logistic regression and random forest methods (P < 0.001). The AUROCs for the logistic regression and random forest methods (machine learning models) were 0.770 (0.761-0.779) and 0.757 (0.747-0.767), respectively. For the detection of LVH, the AUROC was 0.784 (0.777-0.791) for the deep learning model, and this was significantly higher than that of the logistic regression and random forest methods and conventional ECG criteria (P < 0.001). The AUROCs for the logistic regression and random forest methods were 0.758 (0.751-0.765) and 0.716 (0.708-0.724), respectively. This study suggests that deep learning is a useful method to detect LVD and LVH from 12-lead ECGs.

Deep learning model to detect significant aortic regurgitation using electrocardiography.

BACKGROUND: Aortic regurgitation (AR) is a common heart disease, with a relatively high prevalence of 4.9% in the Framingham Heart Study. Because the prevalence increases with advancing age, an upward shift in the age distribution may increase the burden of AR. To provide an effective screening method for AR, we developed a deep learning-based artificial intelligence algorithm for the diagnosis of significant AR using electrocardiography (ECG). METHODS: Our dataset comprised 29,859 paired data of ECG and echocardiography, including 412 AR cases, from January 2015 to December 2019. This dataset was divided into training, validation, and test datasets. We developed a multi-input neural network model, which comprised a two-dimensional convolutional neural network (2D-CNN) using raw ECG data and a fully connected deep neural network (FC-DNN) using ECG features, and compared its performance with the performances of a 2D-CNN model and other machine learning models. In addition, we used gradient-weighted class activation mapping (Grad-CAM) to identify which parts of ECG waveforms had the most effect on algorithm decision making. RESULTS: The area under the receiver operating characteristic curve of the multi-input model (0.802; 95% CI, 0.762-0.837) was significantly greater than that of the 2D-CNN model alone (0.734; 95% CI, 0.679-0.783; p<0.001) and those of other machine learning models. Grad-CAM demonstrated that the multi-input model tended to focus on the QRS complex in leads I and aVL when detecting AR. CONCLUSIONS: The multi-input deep learning model using 12-lead ECG data could detect significant AR with modest predictive value.

The Effectiveness of a Deep Learning Model to Detect Left Ventricular Systolic Dysfunction from Electrocardiograms.

Deep learning models can be applied to electrocardiograms (ECGs) to detect left ventricular (LV) dysfunction. We hypothesized that applying a deep learning model may improve the diagnostic accuracy of cardiologists in predicting LV dysfunction from ECGs. We acquired 37,103 paired ECG and echocardiography data records of patients who underwent echocardiography between January 2015 and December 2019. We trained a convolutional neural network to identify the data records of patients with LV dysfunction (ejection fraction < 40%) using a dataset of 23,801 ECGs. When tested on an independent set of 7,196 ECGs, we found the area under the receiver operating characteristic curve was 0.945 (95% confidence interval: 0.936-0.954). When 7 cardiologists interpreted 50 randomly selected ECGs from the test dataset of 7,196 ECGs, their accuracy for predicting LV dysfunction was 78.0% ± 6.0%. By referring to the model's output, the cardiologist accuracy improved to 88.0% ± 3.7%, which indicates that model support significantly improved the cardiologist diagnostic accuracy (P = 0.02). A sensitivity map demonstrated that the model focused on the QRS complex when detecting LV dysfunction on ECGs. We developed a deep learning model that can detect LV dysfunction on ECGs with high accuracy. Furthermore, we demonstrated that support from a deep learning model can help cardiologists to identify LV dysfunction on ECGs.

Outcomes of Patients With Acute Myocardial Infarction Who Recovered From Severe In-hospital Complications.

Acute myocardial infarction (AMI) would sometimes raise severe in-hospital complications such as cardiopulmonary arrest, shock, stroke, atrioventricular block, and respiratory failure. The purpose of this retrospective study was to compare the clinical outcomes of AMI patients who recovered from severe in-hospital complications with those who did not have in-hospital complications. We included 494 AMI patients, and divided those into the in-hospital complications group (n = 166) and noncomplications group (n = 328). The primary end point was the major adverse cardiovascular events (MACE) defined as the composite of all cause death, nonfatal myocardial infarction (MI), and readmission for heart failure within 1 year after the hospital discharge. A total of 50 postdischarge MACE were observed during the study period. MACE was more frequently observed in the in-hospital complications group (14.5%) than in the noncomplications group (7.9%) (p = 0.023). The presence of in-hospital complications was significantly associated with the MACE (Odds Ratio 1.889, 95% Confidence Interval 1.077 to 3.313, p = 0.026) after controlling age, gender, ST-elevation MI, and culprit of AMI. In conclusion, the MACE was significantly frequent in AMI patients who recovered from severe in-hospital complications and discharged to home, as compared with those who did not have in-hospital complications. AMI patients who recovered from complications could be recognized as a high risk group, and should be carefully managed after discharge to prevent cardiovascular events.

Further Validation of a Novel Acute Myocardial Infarction Risk Stratification (nARS) System for Patients with Acute Myocardial Infarction.

Recently, we developed a novel acute myocardial infarction (AMI) risk stratification system (nARS), which stratifies AMI patients into low- (L), intermediate- (I), and high- (H) risk groups. We have shown that the nARS shortened the length of intensive care unit (ICU) stay as well as that of hospitalization. However, the incidence of AMI-related adverse outcomes has not been fully investigated. The purpose of this study was to investigate the incidence of severe complications requiring ICU care among the 3 risk groups stratified by nARS. We retrospectively reviewed AMI patients between October 2016 and December 2018. A total of 592 patients were divided into the L- (n = 285), I- (n = 124), and H- (n = 183) risk groups. The primary endpoint was in-hospital complications requiring ICU care defined as death/cardiopulmonary arrest, shock, stroke, atrioventricular block, and respiratory failure. Among 592 patients, 239 (40.4%) developed at least 1 complication requiring ICU care, but only 28 (11.7%) developed complications in general wards. Complications requiring ICU care were most frequently observed in the H-risk group (68.9%), followed by the I-risk group (50.8%), and least in the L-risk group (17.5%) (P < 0.001). Complications requiring ICU care that occurred in the general wards were more frequently observed in the H-risk group (8.7%) compared to the I-risk (3.2%) and L-risk (2.8%) groups (P = 0.009). In conclusion, complications requiring ICU care rarely happened in the general wards, and were less in the I- and L-risk groups than in the H-risk group. These results validated the nARS, and might support the widespread use of nARS.