ObjectiveTo explore the application of Tsetlin Machine (TM) in heart beat classification. MethodsTM was used to classify the normal beats, premature ventricular contraction (PVC) and supraventricular premature beats (SPB) in the 2020 data set of China Physiological Signal Challenge. This data set consisted of the single-lead electrocardiogram data of 10 patients with arrhythmia. One patient with atrial fibrillation was excluded, and finally data of the other 9 patients were included in this study. The classification results were then analyzed. ResultsThe classification results showed that the average recognition accuracy of TM was 84.3%, and the basis of classification could be shown by the bit pattern interpretation diagram. ConclusionTM can explain the classification results when classifying heart beats. The reasonable interpretation of classification results can increase the reliability of the model and facilitate people's review and understanding.
Objective To propose a heart sound segmentation method based on multi-feature fusion network. Methods Data were obtained from the CinC/PhysioNet 2016 Challenge dataset (a total of 3 153 recordings from 764 patients, about 91.93% of whom were male, with an average age of 30.36 years). Firstly the features were extracted in time domain and time-frequency domain respectively, and reduced redundant features by feature dimensionality reduction. Then, we selected optimal features separately from the two feature spaces that performed best through feature selection. Next, the multi-feature fusion was completed through multi-scale dilated convolution, cooperative fusion, and channel attention mechanism. Finally, the fused features were fed into a bidirectional gated recurrent unit (BiGRU) network to heart sound segmentation results. Results The proposed method achieved precision, recall and F1 score of 96.70%, 96.99%, and 96.84% respectively. Conclusion The multi-feature fusion network proposed in this study has better heart sound segmentation performance, which can provide high-accuracy heart sound segmentation technology support for the design of automatic analysis of heart diseases based on heart sounds.
ObjectiveTo use machine learning technology to predict the recurrence of atrial fibrillation (AF) after radiofrequency ablation, and try to find the risk factors affecting postoperative recurrence. MethodsA total of 300 patients with valvular AF who underwent radiofrequency ablation in West China Hospital and its branch (Shangjin Hospital) from January 2017 to January 2021 were enrolled, including 129 males and 171 females with a mean age of 52.56 years. We built 5 machine learning models to predict AF recurrence, combined the 3 best performing models into a voting classifier, and made prediction again. Finally, risk factor analysis was performed using the SHApley Additive exPlanations method. ResultsThe voting classifier yielded a prediction accuracy rate of 75.0%, a recall rate of 61.0%, and an area under the receiver operating characteristic curve of 0.79. In addition, factors such as left atrial diameter, ejection fraction, and right atrial diameter were found to have an influence on postoperative recurrence. ConclusionMachine learning-based prediction of recurrence of valvular AF after radiofrequency ablation can provide a certain reference for the clinical diagnosis of AF, and reduce the risk to patients due to ineffective ablation. According to the risk factors found in the study, it can provide patients with more personalized treatment.
Objective To establish a machine learning based framework to rapidly screen out high-risk patients who may develop atrial fibrillation (AF) from patients with valvular heart disease and provide the information related to risk prediction to clinicians as clinical guidance for timely treatment decisions. Methods Clinical data were retrospectively collected from 1 740 patients with valvular heart disease at West China Hospital of Sichuan University and its branches, including 831 (47.76%) males and 909 (52.24%) females at an average age of 54 years. Based on these data, we built classical logistic regression, three standard machine learning models, and three integrated machine learning models for risk prediction and characterization analysis of AF. We compared the performance of machine learning models with classical logistic regression and selected the best two models, and applied the SHAP algorithm to provide interpretability at the population and single-unit levels. In addition, we provided visualization of feature analysis results. ResultsThe Stack model performed best among all models (AF detection rate 85.6%, F1 score 0.753), while XGBoost outperformed the standard machine learning models (AF detection rate 71.9%, F1 score 0.732), and both models performed significantly better than the logistic regression model (AF detection rate 65.2%, F1 score 0.689). SHAP algorithm showed that left atrial internal diameter, mitral E peak flow velocity (Emv), right atrial internal diameter output per beat, and cardiac function class were the most important features affecting AF prediction. Both the Stack model and XGBoost had excellent predictive ability and interpretability. ConclusionThe Stack model has the highest AF detection performance and comprehensive performance. The Stack model loaded with the SHAP algorithm can be used to screen high-risk patients for AF and reveal the corresponding risk characteristics. Our framework can be used to guide clinical intervention and monitoring of AF.
ObjectiveTo evaluate the use of machine learning algorithms for the prediction and characterization of cardiac thrombosis in patients with valvular heart disease and atrial fibrillation. MethodsThis article collected data of patients with valvular disease and atrial fibrillation from West China Hospital of Sichuan University and its branches from 2016 to 2021. From a total of 2 515 patients who underwent valve surgery, 886 patients with valvular disease and atrial fibrillation were included in the study, including 545 (61.5%) males and 341 (38.5%) females, with a mean age of 55.62±9.26 years, and 192 patients had intraoperatively confirmed cardiac thrombosis. We used five supervised machine learning algorithms to predict thrombosis in patients. Based on the clinical data of the patients (33 features after feature screening), the 10-fold nested cross-validation method was used to evaluate the predictive effect of the model through evaluation indicators such as area under the curve, F1 score and Matthews correlation coefficient. Finally, the SHAP interpretation method was used to interpret the model, and the characteristics of the model were analyzed using a patient as an example. ResultsThe final experiment showed that the random forest classifier had the best comprehensive evaluation indicators, the area under the receiver operating characteristic curve was 0.748±0.043, and the accuracy rate reached 79.2%. Interpretation and analysis of the model showed that factors such as stroke volume, peak mitral E-wave velocity and tricuspid pressure gradient were important factors influencing the prediction. ConclusionThe random forest model achieves the best predictive performance and is expected to be used by clinicians as an aided decision-making tool for screening high-embolic risk patients with valvular atrial fibrillation.
Objective To recognize the different phases of Korotkoff sounds through deep learning technology, so as to improve the accuracy of blood pressure measurement in different populations. Methods A classification model of the Korotkoff sounds phases was designed, which fused attention mechanism (Attention), residual network (ResNet) and bidirectional long short-term memory (BiLSTM). First, a single Korotkoff sound signal was extracted from the whole Korotkoff sounds signals beat by beat, and each Korotkoff sound signal was converted into a Mel spectrogram. Then, the local feature extraction of Mel spectrogram was processed by using the Attention mechanism and ResNet network, and BiLSTM network was used to deal with the temporal relations between features, and full-connection layer network was applied in reducing the dimension of features. Finally, the classification was completed by SoftMax function. The dataset used in this study was collected from 44 volunteers (24 females, 20 males with an average age of 36 years), and the model performance was verified using 10-fold cross-validation. Results The classification accuracy of the established model for the 5 types of Korotkoff sounds phases was 93.4%, which was higher than that of other models. Conclusion This study proves that the deep learning method can accurately classify Korotkoff sounds phases, which lays a strong technical foundation for the subsequent design of automatic blood pressure measurement methods based on the classification of the Korotkoff sounds phases.
Currently, in precision cardiac surgery, there are still some pressing issues that need to be addressed. For example, cardiopulmonary bypass remains a critical factor in precise surgical treatment, and many core aspects still rely on the experience and subjective judgment of cardiopulmonary bypass specialists and surgeons, lacking precise data feedback. With the increasing elderly population and rising surgical complexity, precise feedback during cardiopulmonary bypass becomes crucial for improving surgical success rates and facilitating high-complexity procedures. Overcoming these key challenges requires not only a solid medical background but also close collaboration among multiple interdisciplinary fields. Establishing a multidisciplinary team encompassing professionals from the medical, information, software, and related industries can provide high-quality solutions to these challenges. This article shows several patents from a collaborative medical and electronic information team, illustrating how to identify unresolved technical issues and find corresponding solutions in the field of precision cardiac surgery while sharing experiences in applying for invention patents.