The heart sound reflects the movement status of the cardiovascular system and contains the early pathological information of cardiovascular diseases. Automatic heart sound diagnosis plays an essential role in the early detection of cardiovascular diseases. In this study, we aim to develop a novel end-to-end heart sound abnormality detection and classification method, which can be adapted to different heart sound diagnosis tasks. Specifically, we developed a Multi-feature Decision Fusion Network (MDFNet) composed of a Multi-dimensional Feature Extraction (MFE) module and a Multi-dimensional Decision Fusion (MDF) module. The MFE module extracted spatial features, multi-level temporal features and spatial-temporal fusion features to learn heart sound characteristics from multiple perspectives. Through deep supervision and decision fusion, the MDF module made the multi-dimensional features extracted by the MFE module more discriminative, and fused the decision results of multi-dimensional features to integrate complementary information. Furthermore, attention modules were embedded in the MDFNet to emphasize the fundamental heart sounds containing effective feature information. Finally, we proposed an efficient data augmentation method to circumvent the diagnosis performance degradation caused by the lack of cardiac cycle segmentation in other end-to-end methods. The developed method achieved an overall accuracy of 94.44% and a F1-score of 86.90% on the binary classification task and a F1-score of 99.30% on the five-classification task. Our method outperformed other state-of-the-art methods and had good clinical application prospects.

BACKGROUND:Early diagnosis of atrial fibrillation (AF) is important for preventing stroke and other complications. Predicting AF risk in advance can improve early diagnostic efficiency. Deep learning has been used for disease risk prediction; however, it lacks adherence to evidence-based medicine standards. Identifying the underlying mechanisms behind disease risk prediction is important and required. METHODS:We developed an explainable deep learning model called HBBI-AI to predict AF risk using only heart beat-to-beat intervals (HBBIs) during sinus rhythm. We proposed a possible AF mechanism based on the model's explainability and verified this conjecture using confirmed AF risk factors while also examining new AF risk factors. Finally, we investigated the changes in clinicians' ability to predict AF risk using only HBBIs before and after learning the model's explainability. FINDINGS:HBBI-AI consistently performed well across large in-house and external public datasets. HBBIs with large changes or extreme stability were critical predictors for increased AF risk, and the underlying cause was autonomic imbalance. We verified various AF risk factors and discovered that autonomic imbalance was associated with all these factors. Finally, cardiologists effectively understood and learned from these findings to improve their abilities in AF risk prediction. CONCLUSIONS:HBBI-AI effectively predicted AF risk using only HBBI information through evaluating autonomic imbalance. Autonomic imbalance may play an important role in many risk factors of AF rather than in a limited number of risk factors. FUNDING:This study was supported in part by the National Key R&D Program and the National Natural Science Foundation of China.

Abstract Aims As the demand for atrial fibrillation (AF) screening increases, clinicians spend a significant amount of time identifying AF signals from massive amounts of data obtained during long-term dynamic electrocardiogram (ECG) monitoring. The identification of AF signals is subjective and depends on the experience of clinicians. However, experienced cardiologists are scarce. This study aimed to apply a deep learning-based algorithm to fully automate primary screening of patients with AF using 24-h Holter monitoring. Methods and results A deep learning model was developed to automatically detect AF episodes using RR intervals and was trained and evaluated on 23 621 (2297 AF and 21 324 non-AF) 24-h Holter recordings from 23 452 patients. Based on the AF episode detection results, patients with AF were automatically identified using the criterion of at least one AF episode lasting 6 min or longer. Performance was assessed on an independent real-world hospital-scenario test set (19 227 recordings) and a community-scenario test set (1299 recordings). For the two test sets, the model obtained high performance for the identification of patients with AF (sensitivity: 0.995 and 1.000; specificity: 0.985 and 0.997, respectively). Moreover, it obtained good and consistent performance (sensitivity: 1.000; specificity: 0.972) for an external public data set. Conclusion Using the criterion of at least one AF episode of 6 min or longer, the deep learning model can fully automatically screen patients for AF with high accuracy from long-term Holter monitoring data. This method may serve as a powerful and cost-effective tool for primary screening for AF.

Paroxysmal atrial fibrillation (AF) is generally diagnosed by long-term dynamic electrocardiogram (ECG) monitoring. Identifying AF episodes from long-term ECG data can place a heavy burden on clinicians. Many machine-learning-based automatic AF detection methods have been proposed to solve this issue. However, these methods require numerous annotated data to train the model, and the annotation of AF in long-term ECG is extremely time-consuming. Reducing the demand for labeled data can effectively improve the clinical practicability of automatic AF detection methods. In this study, we developed a novel semi-supervised learning method that generated modified low-entropy labels of unlabeled samples for training a deep learning model to automatically detect paroxysmal AF in 24 h Holter monitoring data. Our method employed a 1D CNN-LSTM neural network with RR intervals as input and used few labeled training data with numerous unlabeled data for training the neural network. This method was evaluated using a 24 h Holter monitoring dataset collected from 1000 paroxysmal AF patients. Using labeled samples from only 10 patients for model training, our method achieved a sensitivity of 97.8%, specificity of 97.9%, and accuracy of 97.9% in five-fold cross-validation. Compared to the supervised learning method with complete labeled samples, the detection accuracy of our method was only 0.5% lower, while the workload of data annotation was significantly reduced by more than 98%. In general, this is the first study to apply semi-supervised learning techniques for automatic AF detection using ECG. Our method can effectively reduce the demand for AF data annotations and can improve the clinical practicability of automatic AF detection.

SummaryBackgroundWith the increasing demand for atrial fibrillation (AF) screening, clinicians spend a significant amount of time in identifying the AF signals from massive electrocardiogram (ECG) data in long-term dynamic ECG monitoring. In this study, we aim to reduce clinicians’ workload and promote AF screening by using artificial intelligence (AI) to automatically detect AF episodes and identify AF patients in 24 h Holter recording.MethodsWe used a total of 22 979 Holter recordings (24 h) from 22 757 adult patients and established accurate annotations for AF by cardiologists. First, a randomized clinical cohort of 3 000 recordings (1 500 AF and 1 500 non-AF) from 3000 patients recorded between April 2012 and May 2020 was collected and randomly divided into training, validation and test sets (10:1:4). Then, a deep-learning-based AI model was developed to automatically detect AF episode using RR intervals and was tested with the test set. Based on AF episode detection results, AF patients were automatically identified by using a criterion of at least one AF episode of 6 min or longer. Finally, the clinical effectiveness of the model was verified with an independent real-world test set including 19 979 recordings (1 006 AF and 18 973 non-AF) from 19 757 consecutive patients recorded between June 2020 and January 2021.FindingsOur model achieved high performance for AF episode detection in both test sets (sensitivity: 0.992 and 0.972; specificity: 0.997 and 0.997, respectively). It also achieved high performance for AF patient identification in both test sets (sensitivity:0.993 and 0.994; specificity: 0.990 and 0.973, respectively). Moreover, it obtained superior and consistent performance in an external public database.InterpretationOur AI model can automatically identify AF in long-term ECG recording with high accuracy. This cost-effective strategy may promote AF screening by improving diagnostic effectiveness and reducing clinical workload.Research in contextEvidence before this studyWe searched Google Scholar and PubMed for research articles on artificial intelligence-based diagnosis of atrial fibrillation (AF) published in English between Jan 1, 2016 and Aug 1, 2021, using the search terms “deep learning” OR “deep neural network” OR “machine learning” OR “artificial intelligence” AND “atrial fibrillation”. We found that most of the previous deep learning models in AF detection were trained and validated on benchmark datasets (such as the PhysioNet database, the Massachusetts Institute of Technology Beth Israel Hospital AF database or Long-Term AF database), in which there were less than 100 patients or the recordings contained only short ECG segments (30-60s). Our search did not identify any articles that explored deep neural networks for AF detection in large real-world dataset of 24 h Holter recording, nor did we find articles that can automatically identify patients with AF in 24 h Holter recording.Added value of this studyFirst, long-term Holter monitoring is the main method of AF screening, however, most previous studies of automatic AF detection mainly tested on short ECG recordings. This work focused on 24 h Holter recording data and achieved high accuracy in detecting AF episodes. Second, AF episodes detection did not automatically transform to AF patient identification in 24 h Holter recording, since at present, there is no well-recognized criterion for automatically identifying AF patient. Therefore, we established a criterion to identify AF patients by use of at least one AF episode of 6 min or longer, as this condition led to significantly increased risk of thromboembolism. Using this criterion, our method identified AF patients with high accuracy. Finally, and more importantly, our model was trained on a randomized clinical dataset and tested on an independent real-world clinical dataset to show great potential in clinical application. We did not exclude rare or special cases in the real-world dataset so as not to inflate our AF detection performance. To the best of our knowledge, this is the first study to automatically identifies both AF episodes and AF patients in 24 h Holter recording of large real-world clinical dataset.Implications of all the available evidenceOur deep learning model automatically identified AF patient with high accuracy in 24 h Holter recording and was verified in real-world data, therefore, it can be embedded into the Holter analysis system and deployed at the clinical level to assist the decision making of Holter analysis system and clinicians. This approach can help improve the efficiency of AF screening and reduce the cost for AF diagnosis. In addition, our RR-interval-based model achieved comparable or better performance than the raw-ECG-based method, and can be widely applied to medical devices that can collect heartbeat information, including not only the multi-lead and single-lead Holter devices, but also other wearable devices that can reliably measure the heartbeat signals.

Background: For the nonstationarity of neural recordings in intracortical brain–machine interfaces, daily retraining in a supervised manner is always required to maintain the performance of the decoder. This problem can be improved by using a reinforcement learning (RL) based self-recalibrating decoder. However, quickly exploring new knowledge while maintaining a good performance remains a challenge in RL-based decoders. Methods: To solve this problem, we proposed an attention-gated RL-based algorithm combining transfer learning, mini-batch, and weight updating schemes to accelerate the weight updating and avoid over-fitting. The proposed algorithm was tested on intracortical neural data recorded from two monkeys to decode their reaching positions and grasping gestures. Results: The decoding results showed that our proposed algorithm achieved an approximate 20% increase in classification accuracy compared to that obtained by the non-retrained classifier and even achieved better classification accuracy than the daily retraining classifier. Moreover, compared with a conventional RL method, our algorithm improved the accuracy by approximately 10% and the online weight updating speed by approximately 70 times. Conclusions: This paper proposed a self-recalibrating decoder which achieved a good and robust decoding performance with fast weight updating and might facilitate its application in wearable device and clinical practice.

Discovering major penalties against investment targets is critical to due diligence and risk control in financial markets, but extracting and evaluating penalty events from millions of announcements released by thousands of administrative agencies at all levels is also very challenging. In this paper, we developed the first system for Chinese administrative penalty event extraction, which was formulated as question answering tasks, and constructed corresponding benchmark data for training and evaluation. We also applied the Retrospective Reader mechanism to replace hand-crafted heuristic rules for no-answer scenarios and devised new strategies for generating informative questions. Empirical results on the benchmark demonstrated that our method outperforms the previous state-of-the-art methods. Recently, our system has been successfully deployed in a due diligence project for a top tier Chinese investment bank.