Xingyao Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xingyao Wang has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Chengyu Liu
Southeast University
3 shared publications - Jianqing Li
School of Instrument Science and Engineering, Southeast University, Nanjing, China
3 shared publications Chengyu Liu
2 shared publications- Chenxi Yang
School of Instrument Science and Engineering, Southeast University, Nanjing 210096, China
2 shared publications Hongxiang Gao
2 shared publicationsJianqing Li
2 shared publications
Latest Publications
- Open Access
- 15 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10 (2) , 1-32
M 3 PD Dataset: Enabling Dual-view Photoplethysmography on Smartphones in Lab and Clinical Settings
Portable physiological monitoring is essential for early detection and management of cardiovascular disease, but current methods often require specialized equipment that limits accessibility or imposes impractical postures that patients cannot maintain. Video-based photoplethysmography on smartphones offers a convenient non-invasive alternative, yet it still faces reliability challenges caused by motion artifacts, lighting variations, and single-view constraints. Few studies have demonstrated that this technology can be reliably applied to physiological monitoring of cardiovascular patients, and no widely used open datasets exist for researchers to examine its cross-device accuracy. To address these limitations, we introduce the M 3 PD dataset—the first publicly available dual-view mobile photoplethysmography dataset—comprising synchronized facial and fingertip videos captured simultaneously via front and rear smartphone cameras from 60 participants (including 47 cardiovascular patients). Building on this dual-view setting, we further propose the F 3 Mamba, which fuses the facial and fingertip views through Mamba-based temporal modeling. The model reduces heart-rate error by 21.9-30.2% over existing single-view baselines while showing enhanced robustness across challenging real-world scenarios, and further achieves a 4.4-16.4% MAE reduction over the best standard dual-view fusion baselines. Data and code are released at https://github.com/Health-HCI-Group/F3Mamba.
- 15 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies
- Vol. 10 (2) , 1-32
Open AccessM 3 PD Dataset: Enabling Dual-view Photoplethysmography on Smartphones in Lab and Clinical Settings
Portable physiological monitoring is essential for early detection and management of cardiovascular disease, but current methods often require specialized equipment that limits accessibility or imposes impractical postures that patients cannot maintain. Video-based photoplethysmography on smartphones offers a convenient non-invasive alternative, yet it still faces reliability challenges caused by motion artifacts, lighting variations, and single-view constraints. Few studies have demonstrated that this technology can be reliably applied to physiological monitoring of cardiovascular patients, and no widely used open datasets exist for researchers to examine its cross-device accuracy. To address these limitations, we introduce the M 3 PD dataset—the first publicly available dual-view mobile photoplethysmography dataset—comprising synchronized facial and fingertip videos captured simultaneously via front and rear smartphone cameras from 60 participants (including 47 cardiovascular patients). Building on this dual-view setting, we further propose the F 3 Mamba, which fuses the facial and fingertip views through Mamba-based temporal modeling. The model reduces heart-rate error by 21.9-30.2% over existing single-view baselines while showing enhanced robustness across challenging real-world scenarios, and further achieves a 4.4-16.4% MAE reduction over the best standard dual-view fusion baselines. Data and code are released at https://github.com/Health-HCI-Group/F3Mamba.
- Open Access
- 18 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 63-67
Automatic Cardiac Feature Point Annotation of Cardio-mechanical Signals from Valvular Heart Disease Patients Using Multi-modal Sensor Fusion
The study proposes an innovative ECG-free algorithm for the automatic annotation of cardiac feature points in seismo-cardiogram (SCG) signals from valvular heart disease (VHD) patients. The algorithm is based on multi-modality signals, obviating the need for manual SCG annotation. It consists of self-supervised learning and pseudo-label generation from ECG signals. Initially, the algorithm employs self-supervised learning to analyze the pseudo-periodic nature of SCG signals. Subsequently, pseudo-labels are established using synchronized ECG signals for model training. Finally, labeling tests are conducted with SCG signals alone. The training and testing datasets consist of samples from 100 VHD patients. Experiments demonstrate that the proposed algorithm achieves a precision of 98.94% and a recall of 97.44% on the test set, suggesting the feasibility of the proposed method. The presented framework would be essential for further fiducial point detection and analysis of cardio-mechanical signals, which might provide out-of-clinic evaluations of VHD in the future.
- 18 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 63-67
Open AccessAutomatic Cardiac Feature Point Annotation of Cardio-mechanical Signals from Valvular Heart Disease Patients Using Multi-modal Sensor Fusion
The study proposes an innovative ECG-free algorithm for the automatic annotation of cardiac feature points in seismo-cardiogram (SCG) signals from valvular heart disease (VHD) patients. The algorithm is based on multi-modality signals, obviating the need for manual SCG annotation. It consists of self-supervised learning and pseudo-label generation from ECG signals. Initially, the algorithm employs self-supervised learning to analyze the pseudo-periodic nature of SCG signals. Subsequently, pseudo-labels are established using synchronized ECG signals for model training. Finally, labeling tests are conducted with SCG signals alone. The training and testing datasets consist of samples from 100 VHD patients. Experiments demonstrate that the proposed algorithm achieves a precision of 98.94% and a recall of 97.44% on the test set, suggesting the feasibility of the proposed method. The presented framework would be essential for further fiducial point detection and analysis of cardio-mechanical signals, which might provide out-of-clinic evaluations of VHD in the future.
- 13 January 2023
- Published by Elsevier in Knowledge-Based Systems
- Vol. 263
Label decoupling strategy for 12-lead ECG classification
- 13 January 2023
- Published by Elsevier in Knowledge-Based Systems
- Vol. 263
Label decoupling strategy for 12-lead ECG classification
- Open Access
- 1 January 2023
- Published by Association for Computational Linguistics (ACL)
- p. 9845-9862
Making Pre-trained Language Models both Task-solvers and Self-calibrators
Yangyi Chen, Xingyao Wang, Heng Ji. Findings of the Association for Computational Linguistics: ACL 2023. 2023.
- 1 January 2023
- Published by Association for Computational Linguistics (ACL)
- p. 9845-9862
Open AccessMaking Pre-trained Language Models both Task-solvers and Self-calibrators
Yangyi Chen, Xingyao Wang, Heng Ji. Findings of the Association for Computational Linguistics: ACL 2023. 2023.
- Open Access
- 2 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
PN-QRS: An Uncertainty-aware QRS-complex Detection Method for Wearable ECGs
Long-term wearable ECG monitoring has increased the influence of ambiguous factors on QRS-complex detection. Reliable ECG information must be extracted from abundant noises and confusing artifacts. To address this issue, we suggest a positive-negative QRS-complex detector (PN-QRS), a QRS-complex detection model that accounts for uncertainty. Specifically, we relate non-ECG screening and artifact recognition to epistemic and aleatoric uncertainty, respectively. To disentangle the two uncertainties, we introduce the class-biased transformation based on the uniqueness of this problem. The performance of artifact and non-ECG detection and multi-lead QRS-complex localization is then validated. The results indicate that PN-QRS reduces ECG error detection rate by 22.24% and improves non-ECG screening accuracy by 5.9% compared to conventional signal quality assessment methods. And for identification of ambiguous beats, PN-QRS achieves a F1 score of 82.41% in a manually annotated ambiguous ECG dataset, while maintaining high precision for QRS-complex detection (99.38% F1 score). Above all, for multi-lead QRS-complex location, PN-QRS is approaching the performance upper bound through integrating the frame-level results on the lead with the minimum uncertainty. Our work suggests that the proposed PN-QRS has the potential to be employed as a QRS-complex detector in wearable ECG monitoring, with the capacity to remove invalid non-ECG episodes and identify QRS-like artifacts simultaneously.
- 2 November 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessPN-QRS: An Uncertainty-aware QRS-complex Detection Method for Wearable ECGs
Long-term wearable ECG monitoring has increased the influence of ambiguous factors on QRS-complex detection. Reliable ECG information must be extracted from abundant noises and confusing artifacts. To address this issue, we suggest a positive-negative QRS-complex detector (PN-QRS), a QRS-complex detection model that accounts for uncertainty. Specifically, we relate non-ECG screening and artifact recognition to epistemic and aleatoric uncertainty, respectively. To disentangle the two uncertainties, we introduce the class-biased transformation based on the uniqueness of this problem. The performance of artifact and non-ECG detection and multi-lead QRS-complex localization is then validated. The results indicate that PN-QRS reduces ECG error detection rate by 22.24% and improves non-ECG screening accuracy by 5.9% compared to conventional signal quality assessment methods. And for identification of ambiguous beats, PN-QRS achieves a F1 score of 82.41% in a manually annotated ambiguous ECG dataset, while maintaining high precision for QRS-complex detection (99.38% F1 score). Above all, for multi-lead QRS-complex location, PN-QRS is approaching the performance upper bound through integrating the frame-level results on the lead with the minimum uncertainty. Our work suggests that the proposed PN-QRS has the potential to be employed as a QRS-complex detector in wearable ECG monitoring, with the capacity to remove invalid non-ECG episodes and identify QRS-like artifacts simultaneously.
- 3 November 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-10
Sequence-to-Sequence Load Disaggregation Using Multiscale Residual Neural Network
With the increased demand on economy and efficiency of measurement technology, Non-Intrusive Load Monitoring (NILM) has received more and more attention as a costeffective way to monitor electricity and provide feedback to users. Deep neural networks have been showing a great potential in the field of load disaggregation. In this paper, firstly, a new convolutional model based on residual blocks is proposed to avoid the degradation problem which traditional networks more or less suffer from when network layers are increased in order to learn more complex features. Secondly, we propose dilated convolution to curtail the excessive quantity of model parameters and obtain bigger receptive field, and multi-scale structure to learn mixed data features in a more targeted way. Thirdly, we give details about generating training and test set under certain rules. Finally, the algorithm is tested on real-house public dataset, UK-DALE, with three existing neural networks. The results are compared and analysed, the proposed model shows improvements on F1 score, MAE as well as model complexity across different appliances.
- 3 November 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 70 (00189456) , 1-10
Sequence-to-Sequence Load Disaggregation Using Multiscale Residual Neural Network
With the increased demand on economy and efficiency of measurement technology, Non-Intrusive Load Monitoring (NILM) has received more and more attention as a costeffective way to monitor electricity and provide feedback to users. Deep neural networks have been showing a great potential in the field of load disaggregation. In this paper, firstly, a new convolutional model based on residual blocks is proposed to avoid the degradation problem which traditional networks more or less suffer from when network layers are increased in order to learn more complex features. Secondly, we propose dilated convolution to curtail the excessive quantity of model parameters and obtain bigger receptive field, and multi-scale structure to learn mixed data features in a more targeted way. Thirdly, we give details about generating training and test set under certain rules. Finally, the algorithm is tested on real-house public dataset, UK-DALE, with three existing neural networks. The results are compared and analysed, the proposed model shows improvements on F1 score, MAE as well as model complexity across different appliances.
- 15 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 413-418
An Octave Convolution Neural Network-based QRS Detector
Detection of QRS complex in electrocardiogram (ECG) is the most critical and basic step for automated cardiac diagnosis. Many automated QRS complex detection methods have been proposed, and several state-of-art approaches show acceptable detection accuracy. However, current methods cannot generalize to out-of-distribution data, especially inter-patient and low-quality wearable ECGs.In our work, a 20-layer convolutional neural network (CNN) using octave convolution was proposed to detect QRS complex in unknown and noisy ECGs. Octave convolution allows inter-frequency communication between high- and low-frequency bands in latent space. All extracted features are sent to fully connected classifier for element-wise QRS complex detection. Intra- and inter-independent databases testing was conducted to evaluate methods’ generalization capacity and noise immunity.Our experimental evaluations show that our method results in significant F1 gains on multi independent databases. The proposed method results in average 24.23% improvement comparing with P&T algorithm and 2.16% with SENet, which is the first-place-method in China Physiological signal Challenge (CPSC) 2019. Notably, the proposed method shows superior stability to SENet on the whole databases, especially on databases from out-of-distribution sources of the training set.
- 15 October 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 413-418
An Octave Convolution Neural Network-based QRS Detector
Detection of QRS complex in electrocardiogram (ECG) is the most critical and basic step for automated cardiac diagnosis. Many automated QRS complex detection methods have been proposed, and several state-of-art approaches show acceptable detection accuracy. However, current methods cannot generalize to out-of-distribution data, especially inter-patient and low-quality wearable ECGs.In our work, a 20-layer convolutional neural network (CNN) using octave convolution was proposed to detect QRS complex in unknown and noisy ECGs. Octave convolution allows inter-frequency communication between high- and low-frequency bands in latent space. All extracted features are sent to fully connected classifier for element-wise QRS complex detection. Intra- and inter-independent databases testing was conducted to evaluate methods’ generalization capacity and noise immunity.Our experimental evaluations show that our method results in significant F1 gains on multi independent databases. The proposed method results in average 24.23% improvement comparing with P&T algorithm and 2.16% with SENet, which is the first-place-method in China Physiological signal Challenge (CPSC) 2019. Notably, the proposed method shows superior stability to SENet on the whole databases, especially on databases from out-of-distribution sources of the training set.
- 20 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Engineering
- Vol. 68 (2) , 650-663
Temporal-Framing Adaptive Network for Heart Sound Segmentation Without Prior Knowledge of State Duration
Objective: This paper presents a novel heart sound segmentation algorithm based on Temporal-Framing Adaptive Network (TFAN), including state transition loss and dynamic inference. Methods: In contrast to previous state-of-the-art approaches, TFAN does not require any prior knowledge of the state duration of heart sounds and is therefore likely to generalize to non sinus rhythm. TFAN was trained on 50 recordings randomly chosen from Training set A of the 2016 PhysioNet/Computer in Cardiology Challenge and tested on the other 12 independent databases (2,099 recordings and 52,180 beats). And further testing of performance was conducted on databases with three levels of increasing difficulty (LEVEL-I, -II and -III). Results: TFAN achieved a superior F1 score for all 12 databases except for `Test-B', with an average of 96.72%, compared to 94.56% for logistic regression hidden semi-Markov model (LR-HSMM) and 94.18% for bidirectional gated recurrent neural network (BiGRNN). Moreover, TFAN achieved an overall F1 score of 99.21%, 94.17%, 91.31% on LEVEL-I, -II and -III databases respectively, compared to 98.37%, 87.56%, 78.46% for LR-HSMM and 99.01%, 92.63%, 88.45% for BiGRNN. Conclusion: TFAN therefore provides a substantial improvement on heart sound segmentation while using less parameters compared to BiGRNN. Significance: The proposed method is highly flexible and likely to apply to other non-stationary time series. Further work is required to understand to what extent this approach will provide improved diagnostic performance, although it is logical to assume superior segmentation will lead to improved diagnostics.
- 20 July 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Biomedical Engineering
- Vol. 68 (2) , 650-663
Temporal-Framing Adaptive Network for Heart Sound Segmentation Without Prior Knowledge of State Duration
Objective: This paper presents a novel heart sound segmentation algorithm based on Temporal-Framing Adaptive Network (TFAN), including state transition loss and dynamic inference. Methods: In contrast to previous state-of-the-art approaches, TFAN does not require any prior knowledge of the state duration of heart sounds and is therefore likely to generalize to non sinus rhythm. TFAN was trained on 50 recordings randomly chosen from Training set A of the 2016 PhysioNet/Computer in Cardiology Challenge and tested on the other 12 independent databases (2,099 recordings and 52,180 beats). And further testing of performance was conducted on databases with three levels of increasing difficulty (LEVEL-I, -II and -III). Results: TFAN achieved a superior F1 score for all 12 databases except for `Test-B', with an average of 96.72%, compared to 94.56% for logistic regression hidden semi-Markov model (LR-HSMM) and 94.18% for bidirectional gated recurrent neural network (BiGRNN). Moreover, TFAN achieved an overall F1 score of 99.21%, 94.17%, 91.31% on LEVEL-I, -II and -III databases respectively, compared to 98.37%, 87.56%, 78.46% for LR-HSMM and 99.01%, 92.63%, 88.45% for BiGRNN. Conclusion: TFAN therefore provides a substantial improvement on heart sound segmentation while using less parameters compared to BiGRNN. Significance: The proposed method is highly flexible and likely to apply to other non-stationary time series. Further work is required to understand to what extent this approach will provide improved diagnostic performance, although it is logical to assume superior segmentation will lead to improved diagnostics.