Research on a phonocardiogram and electrocardiogram signal classification algorithm based on global group coordinate attention mechanism and multi-scale feature fusion

Purpose In recent years, the incidence of cardiovascular disease has continued to rise, and early screening and prevention are especially critical. Phonocardiography (PCG) and electrocardiography (ECG), as simple, cost-effective and non-invasive tests, are important tools for clinical analysis. However, it is difficult to fully reflect the complexity of the cardiovascular system using PCG or ECG tests alone. Combining the multimodal signals of PCG and ECG can provide complementary information to improve the detection accuracy. Therefore, the purpose of this paper is to propose a multimodal signal classification method based on continuous wavelet transform and improved ResNet18. Design/methodology/approach The classification method is based on the ResNet18 backbone, and the ResNet18 network is improved by embedding the global grouped coordinate attention mechanism module and the improved bidirectional feature pyramid network. Firstly, a data acquisition system was built using a MEMS-integrated PCG-ECG sensor to construct a private data set. Second is the time-frequency transformation of PCG and ECG synchronized signals on public and private data sets using continuous wavelet transform. Finally, the time-frequency images are categorized. Findings The global grouped coordinate attention mechanism and bidirectional feature pyramid network modules proposed in this paper significantly enhance the model’s performance. On public data sets, the method achieves precision, sensitivity, specificity, accuracy and F1 score of 97.96%, 98.51%, 97.58%, 98.08% and 98.23%, respectively, which represent improvements of 3.54%, 3.92%, 4.18%, 4.03% and 3.72% compared to ResNet18. Additionally, it demonstrates a clear advantage over existing mainstream algorithms. On private data sets, the method’s five metrics are 98.15%, 98.76%, 98.08%, 98.42% and 98.45%, further validating the model’s generalization ability. Originality/value The method proposed in this paper not only improves the accuracy and efficiency of the test but also provides an effective solution for early screening and prevention of cardiovascular diseases.