Zhiyuan Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zhiyuan Li has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
Chengliang Liu
5 shared publications- Yanrui Jin
State Key Laboratory of Mechanical System and Vibration, School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China
5 shared publications - Jinlei Liu
School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, China
4 shared publications Yanrui Jin
4 shared publications- Chengjin Qin
Shanghai Jiao Tong University
3 shared publications - Chengliang Liu
Shanghai Jiao Tong University
3 shared publications
Latest Publications
- 21 October 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 131
Lead-adaptive self-supervised ECG representation model for multi-scenario ECG monitoring
- 21 October 2026
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 131
Lead-adaptive self-supervised ECG representation model for multi-scenario ECG monitoring
- Open Access
- 1 January 2026
- Published by Science Exploration Press in BME Horizon
- Vol. 4 (4) , 202619
AI-ECG for wearable monitoring: From arrhythmia diagnosis to early warning and multi-disease prediction
Wearable electrocardiography (ECG) is shifting cardiovascular monitoring from episodic in-hospital testing to continuous out-of-hospital assessment. Artificial intelligence-enabled ECG (AI-ECG) provides a new pathway for extracting clinically useful information from out-of-hospital wearable recordings. This review is organized around the wearable AI-ECG monitoring pipeline, summarizing key advances in model development, clinical application, and real-world validation, while emphasizing the special requirements that wearable scenarios impose on algorithm design and clinical translation. At present, arrhythmia screening remains the most mature application of AI-ECG, with deep learning models achieving cardiologist-level or clinician-comparable performance in several well-defined tasks. With the growing availability of long-term continuous recordings, research is further extending from post-event recognition to pre-event warning, particularly for high-risk events such as acute atrial fibrillation and malignant ventricular arrhythmias. Related methods are evolving from traditional feature-based machine learning toward deep learning and foundation models that can exploit waveform morphology, rhythm dynamics, and long-range temporal information. Beyond rhythm disorders, AI-ECG is also being explored for structural cardiac abnormalities, metabolic disorders, and broader systemic risk prediction, suggesting a potential role for ECG as a digital biomarker platform. However, several barriers continue to limit clinical translation, including limited cross-device and cross-population generalizability, insufficient interpretability, and the lack of prospective real-world validation. Future progress will likely depend on standardized data systems, artifact-aware modeling, cross-device validation, foundation models, longitudinal risk modeling, and intelligent systems designed for clinical workflows. Overall, wearable AI-ECG is evolving from passive abnormality detection toward continuous, proactive, and personalized health risk assessment.
- 1 January 2026
- Published by Science Exploration Press in BME Horizon
- Vol. 4 (4) , 202619
Open AccessAI-ECG for wearable monitoring: From arrhythmia diagnosis to early warning and multi-disease prediction
Wearable electrocardiography (ECG) is shifting cardiovascular monitoring from episodic in-hospital testing to continuous out-of-hospital assessment. Artificial intelligence-enabled ECG (AI-ECG) provides a new pathway for extracting clinically useful information from out-of-hospital wearable recordings. This review is organized around the wearable AI-ECG monitoring pipeline, summarizing key advances in model development, clinical application, and real-world validation, while emphasizing the special requirements that wearable scenarios impose on algorithm design and clinical translation. At present, arrhythmia screening remains the most mature application of AI-ECG, with deep learning models achieving cardiologist-level or clinician-comparable performance in several well-defined tasks. With the growing availability of long-term continuous recordings, research is further extending from post-event recognition to pre-event warning, particularly for high-risk events such as acute atrial fibrillation and malignant ventricular arrhythmias. Related methods are evolving from traditional feature-based machine learning toward deep learning and foundation models that can exploit waveform morphology, rhythm dynamics, and long-range temporal information. Beyond rhythm disorders, AI-ECG is also being explored for structural cardiac abnormalities, metabolic disorders, and broader systemic risk prediction, suggesting a potential role for ECG as a digital biomarker platform. However, several barriers continue to limit clinical translation, including limited cross-device and cross-population generalizability, insufficient interpretability, and the lack of prospective real-world validation. Future progress will likely depend on standardized data systems, artifact-aware modeling, cross-device validation, foundation models, longitudinal risk modeling, and intelligent systems designed for clinical workflows. Overall, wearable AI-ECG is evolving from passive abnormality detection toward continuous, proactive, and personalized health risk assessment.
- 2 September 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 298
Spatial-temporal hierarchical decoupled masked autoencoder: A self-supervised learning framework for electrocardiogram
- 2 September 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 298
Spatial-temporal hierarchical decoupled masked autoencoder: A self-supervised learning framework for electrocardiogram
- 18 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-15
EDDM: A Novel ECG Denoising Method Using Dual-Path Diffusion Model
Background: Electrocardiogram (ECG), as a low-cost and non-invasive measurement tool, plays a crucial role in detecting arrhythmias. However, the ECG measurement process is highly vulnerable to factors such as baseline wandering, electrode motion, and muscle artifacts, leading to significant degradation in signal quality. Consequently, exploring denoising methods to obtain clean ECGs has emerged as an essential and challenging task. Methods: We propose a novel denoising method for ECG signals called the ECG Denoising Diffusion Model (EDDM). Unlike traditional generative methods, EDDM employs two distinct pathways in the forward stage: ECG noise diffusion and Gaussian white noise diffusion. This unique approach allows for directed diffusion from the clean ECG domain to the measured ECG domain. In the reverse process, we design a U-Net model that combines deep aggregation pyramid pooling layers to accurately capture the multi-scale information of the ECG for predicting the two types of noise during the diffusion process. Therefore, EDDM follows an interpretable denoising path and directly obtains the clean ECG from the noisy ECG instead of using a generative approach, resulting in more intuitive and stable denoising outcomes. Conclusion: EDDM has achieved state-of-the-art (SOTA) results on five metrics in benchmark experiments, with a significant improvement of 4%-10% in the percentage root-meansquare difference (PRD) compared to the existing SOTA model. Analysis shows that EDDM interpretably achieves the diffusion of clean ECG into noisy ECG, rather than white noise. Furthermore, under extreme noise conditions, EDDM achieves more stable and higher-performance denoising results compared to the SOTA method, demonstrating the enormous potential of EDDM in wearable ECG measurement systems.
- 18 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-15
EDDM: A Novel ECG Denoising Method Using Dual-Path Diffusion Model
Background: Electrocardiogram (ECG), as a low-cost and non-invasive measurement tool, plays a crucial role in detecting arrhythmias. However, the ECG measurement process is highly vulnerable to factors such as baseline wandering, electrode motion, and muscle artifacts, leading to significant degradation in signal quality. Consequently, exploring denoising methods to obtain clean ECGs has emerged as an essential and challenging task. Methods: We propose a novel denoising method for ECG signals called the ECG Denoising Diffusion Model (EDDM). Unlike traditional generative methods, EDDM employs two distinct pathways in the forward stage: ECG noise diffusion and Gaussian white noise diffusion. This unique approach allows for directed diffusion from the clean ECG domain to the measured ECG domain. In the reverse process, we design a U-Net model that combines deep aggregation pyramid pooling layers to accurately capture the multi-scale information of the ECG for predicting the two types of noise during the diffusion process. Therefore, EDDM follows an interpretable denoising path and directly obtains the clean ECG from the noisy ECG instead of using a generative approach, resulting in more intuitive and stable denoising outcomes. Conclusion: EDDM has achieved state-of-the-art (SOTA) results on five metrics in benchmark experiments, with a significant improvement of 4%-10% in the percentage root-meansquare difference (PRD) compared to the existing SOTA model. Analysis shows that EDDM interpretably achieves the diffusion of clean ECG into noisy ECG, rather than white noise. Furthermore, under extreme noise conditions, EDDM achieves more stable and higher-performance denoising results compared to the SOTA method, demonstrating the enormous potential of EDDM in wearable ECG measurement systems.
- 20 July 2024
- Published by Elsevier in Applied Soft Computing
- Vol. 164
A self-supervised framework for computer-aided arrhythmia diagnosis
- 20 July 2024
- Published by Elsevier in Applied Soft Computing
- Vol. 164
A self-supervised framework for computer-aided arrhythmia diagnosis
- 17 May 2022
- Published by Elsevier in Knowledge-Based Systems
- Vol. 250
An efficient neural network-based method for patient-specific information involved arrhythmia detection
- 17 May 2022
- Published by Elsevier in Knowledge-Based Systems
- Vol. 250
An efficient neural network-based method for patient-specific information involved arrhythmia detection
- 1 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-11
A Novel Interpretable Method Based on Dual-Level Attentional Deep Neural Network for Actual Multilabel Arrhythmia Detection
Arrhythmia accounts for more than 80% of sudden cardiac death, and its incidence rate has increased rapidly recently. Nowadays, many studies have applied artificial intelligence (AI) methods to arrhythmia detection. Deep learning approaches can improve model performance but lack the capability to mine the relationship between the electrocardiogram (ECG) samples and final results. Furthermore, most arrhythmia diagnosis methods aim at only multiclassification problems. However, in many cases, an ECG recording may have more than one arrhythmia disease. In this article, we develop an accurate and interpretable model for multilabel ECG signals, which is called dual-level attentional convolutional long short-term memory neural network (DLA-CLSTM). Convolutional layers, bidirectional long short-term memory layers, and attention mechanisms are used for extracting beat- and rhythm-level information. Notably, the attention mechanism helps the model be more interpretable. In addition, we design a weighted multilabel classifier to address multilabel and imbalance problems. Finally, the MIT-BIH Arrhythmia Database and the 1st China Physiological Signal Challenge dataset are utilized to evaluate DLA-CLSTM. Compared with existing methods, DLA-CLSTM can improve the accuracy by 22.50% and the F1-macro by 20.51% on average. Further analysis shows that the areas focused by DLA-CLSTM match the clinical judgment basis to some extent. Collectively, DLA-CLSTM has good potential for assisting clinicians in real applications.
- 1 January 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 71 (00189456) , 1-11
A Novel Interpretable Method Based on Dual-Level Attentional Deep Neural Network for Actual Multilabel Arrhythmia Detection
Arrhythmia accounts for more than 80% of sudden cardiac death, and its incidence rate has increased rapidly recently. Nowadays, many studies have applied artificial intelligence (AI) methods to arrhythmia detection. Deep learning approaches can improve model performance but lack the capability to mine the relationship between the electrocardiogram (ECG) samples and final results. Furthermore, most arrhythmia diagnosis methods aim at only multiclassification problems. However, in many cases, an ECG recording may have more than one arrhythmia disease. In this article, we develop an accurate and interpretable model for multilabel ECG signals, which is called dual-level attentional convolutional long short-term memory neural network (DLA-CLSTM). Convolutional layers, bidirectional long short-term memory layers, and attention mechanisms are used for extracting beat- and rhythm-level information. Notably, the attention mechanism helps the model be more interpretable. In addition, we design a weighted multilabel classifier to address multilabel and imbalance problems. Finally, the MIT-BIH Arrhythmia Database and the 1st China Physiological Signal Challenge dataset are utilized to evaluate DLA-CLSTM. Compared with existing methods, DLA-CLSTM can improve the accuracy by 22.50% and the F1-macro by 20.51% on average. Further analysis shows that the areas focused by DLA-CLSTM match the clinical judgment basis to some extent. Collectively, DLA-CLSTM has good potential for assisting clinicians in real applications.
- 1 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-12
A Novel Incremental and Interactive Method for Actual Heartbeat Classification With Limited Additional Labeled Samples
In the actual scenario, it is difficult for an automatic diagnosis system to detect each heartbeat when facing new patients. In this article, a novel active dual-scale residual convolutional long short-term memory neural network is proposed to classify heartbeats interactively and automatically. First, each heartbeat and electrocardiogram segment including current heartbeat are combined to build dual-scale input information. In addition, convolution layers, residual block, and bidirectional long short memory are integrated to extract features from dual-scale inputs. Furthermore, active learning approach with random breaking-ties selection strategy is introduced to choose the most representative unlabeled samples for labeling and fine-tune the trained model with these representative labeled samples. The experimental results show that the proposed method uses only 5.14% of the additional data set and improves the model accuracy and macro F1 score by 13.23% and 35.62%, respectively. The proposed structure has good representative ability and our selection strategy can reduce the workload of experts and improve the performance greatly. Thus, the proposed interactive and incremental method has good potential in real applications.
- 1 January 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-12
A Novel Incremental and Interactive Method for Actual Heartbeat Classification With Limited Additional Labeled Samples
In the actual scenario, it is difficult for an automatic diagnosis system to detect each heartbeat when facing new patients. In this article, a novel active dual-scale residual convolutional long short-term memory neural network is proposed to classify heartbeats interactively and automatically. First, each heartbeat and electrocardiogram segment including current heartbeat are combined to build dual-scale input information. In addition, convolution layers, residual block, and bidirectional long short memory are integrated to extract features from dual-scale inputs. Furthermore, active learning approach with random breaking-ties selection strategy is introduced to choose the most representative unlabeled samples for labeling and fine-tune the trained model with these representative labeled samples. The experimental results show that the proposed method uses only 5.14% of the additional data set and improves the model accuracy and macro F1 score by 13.23% and 35.62%, respectively. The proposed structure has good representative ability and our selection strategy can reduce the workload of experts and improve the performance greatly. Thus, the proposed interactive and incremental method has good potential in real applications.