Deep learning-based regional ECG diagnosis platform.

OBJECTIVE: To enable the intelligent diagnosis of a variety of common Electrocardiogram (ECG), we investigate the deep learning-based ECG diagnosis system. METHODS: From January 2015 to December 2019, four consecutive years of 100,120 conventional 12-lead ECG data were collected in our hospital. Utilizing this dataset, we constructed a deep learning model designed to intelligently diagnose prevalent ECG anomalies by employing a multi-task learning framework. The system performance was evaluated using various metrics, including sensitivity, specificity, negative predictive value, positive predictive value, and so forth. Additionally, we employed an ECG intelligent diagnostic platform for clinical application to undertake real-time online analysis of 2500 conventional 12-lead ECG samples in June 2020, aiming to validate our model. At this stage, we compared the performance of our model against the traditional manual identification method. RESULTS: The efficacy of the ECG intelligent diagnostic model was notably high for common and straightforward ECG patterns, such as sinus rhythm (F1 = 98.01%), sinus tachycardia (F1 = 96.26%), sinus bradycardia (F1 = 94.88%), and a normal electrocardiogram (F1 = 91.71%), as well as for Premature Ventricular Contractions (F1 = 91.62%). Nevertheless, when diagnosing rarer and more intricate ECG anomalies, the system requires an increased number of samples to refine the deep learning models. During the validation stage, our model exhibited better efficiency in terms of accuracy, labor time and labor cost when compared to the manual identification approach. CONCLUSIONS: Our deep learning-driven intelligent ECG diagnostic model clearly demonstrates significant clinical utility. The integrated artificial intelligence diagnosis system not only has the potential to augment physicians in their diagnostic processes but also offers a viable avenue to reduce associated labor costs.

CNN-FWS: A Model for the Diagnosis of Normal and Abnormal ECG with Feature Adaptive.

(1) Background and objective: Cardiovascular disease is one of the most common causes of death in today's world. ECG is crucial in the early detection and prevention of cardiovascular disease. In this study, an improved deep learning method is proposed to diagnose abnormal and normal ECG accurately. (2) Methods: This paper proposes a CNN-FWS that combines three convolutional neural networks (CNN) and recursive feature elimination based on feature weights (FW-RFE), which diagnoses abnormal and normal ECG. F1 score and Recall are used to evaluate the performance. (3) Results: A total of 17,259 records were used in this study, which validated the diagnostic performance of CNN-FWS for normal and abnormal ECG signals in 12 leads. The experimental results show that the F1 score of CNN-FWS is 0.902, and the Recall of CNN-FWS is 0.889. (4) Conclusion: CNN-FWS absorbs the advantages of convolutional neural networks (CNN) to obtain three parts of different spatial information and enrich the learned features. CNN-FWS can select the most relevant features while eliminating unrelated and redundant features by FW-RFE, making the residual features more representative and effective. The method is an end-to-end modeling approach that enables an adaptive feature selection process without human intervention.

Explainable localization of premature ventricular contraction using deep learning-based semantic segmentation of 12-lead electrocardiogram.

Background: Predicting the origin of premature ventricular contraction (PVC) from the preoperative electrocardiogram (ECG) is important for catheter ablation therapies. We propose an explainable method that localizes PVC origin based on the semantic segmentation result of a 12-lead ECG using a deep neural network, considering suitable diagnosis support for clinical application. Methods: The deep learning-based semantic segmentation model was trained using 265 12-lead ECG recordings from 84 patients with frequent PVCs. The model classified each ECG sampling time into four categories: background (BG), sinus rhythm (SR), PVC originating from the left ventricular outflow tract (PVC-L), and PVC originating from the right ventricular outflow tract (PVC-R). Based on the ECG segmentation results, a rule-based algorithm classified ECG recordings into three categories: PVC-L, PVC-R, as well as Neutral, which is a group for the recordings requiring the physician's careful assessment before separating them into PVC-L and PVC-R. The proposed method was evaluated with a public dataset which was used in previous research. Results: The evaluation of the proposed method achieved neutral rate, accuracy, sensitivity, specificity, F1-score, and area under the curve of 0.098, 0.932, 0.963, 0.882, 0.945, and 0.852 on a private dataset, and 0.284, 0.916, 0.912, 0.930, 0.943, and 0.848 on a public dataset, respectively. These quantitative results indicated that the proposed method outperformed almost all previous studies, although a significant number of recordings resulted in requiring the physician's assessment. Conclusions: The feasibility of explainable localization of premature ventricular contraction was demonstrated using deep learning-based semantic segmentation of 12-lead ECG.Clinical trial registration: M26-148-8.

Impact of diagnostic checklists on the interpretation of normal and abnormal electrocardiograms.

OBJECTIVES: Checklists that aim to support clinicians' diagnostic reasoning processes are often recommended to prevent diagnostic errors. Evidence on checklist effectiveness is mixed and seems to depend on checklist type, case difficulty, and participants' expertise. Existing studies primarily use abnormal cases, leaving it unclear how the diagnosis of normal cases is affected by checklist use. We investigated how content-specific and debiasing checklists impacted performance for normal and abnormal cases in electrocardiogram (ECG) diagnosis. METHODS: In this randomized experiment, 42 first year general practice residents interpreted normal, simple abnormal, and complex abnormal ECGs without a checklist. One week later, they were randomly assigned to diagnose the ECGs again with either a debiasing or content-specific checklist. We measured residents' diagnostic accuracy, confidence, patient management, and time taken to diagnose. Additionally, confidence-accuracy calibration was assessed. RESULTS: Accuracy, confidence, and patient management were not significantly affected by checklist use. Time to diagnose decreased with a checklist (M=147s (77)) compared to without a checklist (M=189s (80), Z=-3.10, p=0.002). Additionally, residents' calibration improved when using a checklist (phase 1: R2=0.14, phase 2: R2=0.40). CONCLUSIONS: In both normal and abnormal cases, checklist use improved confidence-accuracy calibration, though accuracy and confidence were not significantly affected. Time to diagnose was reduced. Future research should evaluate this effect in more experienced GPs. Checklists appear promising for reducing overconfidence without negatively impacting normal or simple ECGs. Reducing overconfidence has the potential to improve diagnostic performance in the long term.

Arrhythmia Disease Diagnosis Based on ECG Time-Frequency Domain Fusion and Convolutional Neural Network.

Electrocardiogram (ECG) signals are often used to diagnose cardiac status. However, most of the existing ECG diagnostic methods only use the time-domain information, resulting in some obviously lesion information in frequency-domain of ECG signals are not being fully utilized. Therefore, we propose a method to fuse the time and frequency domain information in ECG signals by convolutional neural network (CNN). First, we adapt multi-scale wavelet decomposition to filter the ECG signal; Then, R-wave localization is used to segment each individual heartbeat cycle; And then, the frequency domain information of this heartbeat cycle is extracted via fast Fourier transform. Finally, the temporal information is spliced with the frequency domain information and input to the neural network for classification. The experimental results show that the proposed method has the highest recognition accuracy (99.43%) of ECG singles compared with state-of-the-art methods. Clinical and Translational Impact Statement- The proposed ECG classification method provides an effective solution for ECG interrogation to quickly diagnose the presence of arrhythmia in a patient from the ECG signal. It can increase the efficiency of the interrogating physician by aiding diagnosis.

Diagnosis of arrhythmias with few abnormal ECG samples using metric-based meta learning.

A major challenge in artificial intelligence based ECG diagnosis lies that it is difficult to obtain sufficient annotated training samples for each rhythm type, especially for rare diseases, which makes many approaches fail to achieve the desired performance with limited ECG records. In this paper, we propose a Meta Siamese Network (MSN) based on metric learning to achieve high accuracy for automatic ECG arrhythmias diagnosis with limited ECG records. First, the ECG signals from three different ECG datasets are preprocessed through resampling, wavelet denoising, R-wave localization, heartbeat segmentation and Z-score normalization. Then, an ECG dataset with limited records is constructed to verify the performance of the proposed model and explore variation of model performance with the sample size. Second, a metric-based meta-learning framework is proposed to address the challenge of few-shot learning for automatic ECG diagnosis of cardiac arrhythmia, and siamese network is employed to achieve arrhythmia diagnosis based on similarity metric. Finally, the N-way K-shot meta-testing strategy is proposed based on the siamese network with double inputs, and the experimental results demonstrate that the proposed strategy can effectively improve the robustness of the proposed model.

A Detector for Premature Atrial and Ventricular Complexes.

The relationship between premature atrial complexes (PACs) and atrial fibrillation (AF), stroke and myocardium degradation is unclear. Current PAC detectors are beat classifiers that attain low sensitivity on PAC detection. The lack of a proper PAC detector hinders the study of the implications of this event and its monitoring. In this work a PAC and ventricular detector is presented. Two PhysioNet open-source databases were used: the long-term ST database (LTSTDB) and the supraventricular arrhythmia database (SVDB). A combination of heart rate variability (HRV) and morphological features were used to classify beats. Morphological features were extracted from the ECG as well as on the 4th scale of the discrete wavelet transform (DWT). After feature selection, a random forest algorithm was trained for a binary classification of PAC (S) vs. others and for a multi-labels classification to discriminate between normal (N), S and ventricular (V) beats. The algorithm was tested in a 10-fold cross-validation following a patient-wise train-test division (i.e., no beats belonging to the same patient were included both in the test and train set). The resultant median sensitivity, specificity and positive predictive value (PPV) were 99.29, 99.54, and 100% for (N), 95.83, 99.39, and 35.68% for (S), 100, 99.90, and 79.63% for (V). The proposed method attains a greater PAC and ventricular beat sensitivity and PPV than the state-of-the-art classifiers.