11 July 2023 in IEEE Transactions on Neural Networks and Learning Systems
Extracting invariant representations in unlabeled electrocardiogram (ECG) signals is a challenge for deep neural networks (DNNs). Contrastive learning is a promising method for unsupervised learning. However, it should improve its robustness to noise and learn the spatiotemporal and semantic representations of categories, just like cardiologists. This article proposes a patient-level adversarial spatiotemporal contrastive learning (ASTCL) framework, which includes ECG augmentations, an adversarial module, and a spatiotemporal contrastive module. Based on the ECG noise attributes, two distinct but effective ECG augmentations, ECG noise enhancement, and ECG noise denoising, are introduced. These methods are beneficial for ASTCL to enhance the robustness of the DNN to noise. This article proposes a self-supervised task to increase the antiperturbation ability. This task is represented as a game between the discriminator and encoder in the adversarial module, which pulls the extracted representations into the shared distribution between the positive pairs to discard the perturbation representations and learn the invariant representations. The spatiotemporal contrastive module combines spatiotemporal prediction and patient discrimination to learn the spatiotemporal and semantic representations of categories. To learn category representations effectively, this article only uses patient-level positive pairs and alternately uses the predictor and the stop-gradient to avoid model collapse. To verify the effectiveness of the proposed method, various groups of experiments are conducted on four ECG benchmark datasets and one clinical dataset compared with the state-of-the-art methods. Experimental results showed that the proposed method outperforms the state-of-the-art methods.
2 August 2023 in IEEE Transactions on Instrumentation and Measurement
Worldwide, the automatic diagnosis (i.e., detection and localization) of myocardial infarction (MI) remains a challenging problem. Clinically, MI is mainly detected and located according to the changes and differences between each electrocardiogram (ECG) lead. Therefore, to capture the features of different MI locations with their corresponding leads, we proposes an end-to-end spatial-temporal MI network (Spa-Tem MI) for detecting and locating MI. To facilitate the extraction of spatial and temporal features, we proposed a width-concat method to convert 1-D multi-lead ECG signals into a 2-D ECG matrix. To learn the spatial and temporal features of MI, the Spa-Tem MI consists of a spatial feature extraction (SFE) module and a temporal feature extraction (TFE) module. The SFE module aims to obtain the spatial differences between leads via dense blocks. Through the self-attention mechanism, the TFE module can capture the global temporal changes of leads. The proposed method was validated with the real MI dataset HeMI, and the dataset was divided by the inter-patient paradigm. The precision, recall, F1 score, and Hamming loss of the results were 94.07%, 95.45%, 93.81%, and 0.03, respectively. In addition, the model was also verified with the public dataset PTB-XL. Compared with that of state-of-the-art methods, the proposed method achieved superior performance.
31 December 2025 in Expert Systems with Applications