Xunhua Huang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xunhua Huang has a total of 46 co-authors. Here are the top co-authors by number of shared publications:
- Haoyi Fan
School of Computer and Artificial Intelligence, Zhengzhou University, Zhengzhou 450001, China
6 shared publications - Fengbin Zhang
Harbin University of Science and Technology
5 shared publications - Liang Xi
Harbin University of Science and Technology
5 shared publications - Ruidong Wang
Zhejiang Normal University
4 shared publications - Han Liu
School of Computer Science and Technology, Harbin University of Science and Technology, Harbin, China
3 shared publications Xinrui Qiu
Department of Traditional Chinese Medicine of Jiangbei Campus, The First Affiliated Hospital of Army Medical University, Chongqing, China
2 shared publications
Latest Publications
- 5 July 2026
- Published by American Association for the Advancement of Science (AAAS) in Intelligent Computing
- Vol. 5
L2L-ESN: Electrocardiogram Lead Synthesis for Assisting Myocardial Infarction Diagnosis
Wearable electrocardiogram (ECG) monitoring devices, important terminals in the Internet of Medical Things (IoMT), can provide real-time and rapid myocardial infarction (MI) warning services for large populations. This enables early medical intervention and considerably improves the survival rate of MI patients, thereby alleviating the increasing prevalence of cardiovascular diseases. However, the number of leads in wearable ECG devices is insufficient to accurately diagnose complex cardiac conditions. To address this issue, this study proposes the Lead-to-Lead ECG Synthesis Network (L2L-ESN), which is based on a conditional Wasserstein generative adversarial network. It utilizes multihead generation to learn the relationships among leads, synthesize complete 12-lead ECGs from specified leads, and automatically identify the location of MI, thereby maximizing the utilization of available resources and improving diagnostic accuracy under limited-lead constraints, ultimately enabling accurate early warning and timely treatment of MI through wearable devices. The proposed method was evaluated in an IoMT-oriented reduced-lead setting on 2 real-world ECG datasets using waveform similarity metrics, downstream diagnostic tests, qualitative visualization, and deployment profiling. The results demonstrate that L2L-ESN generates more accurate electrocardiographic signals for MI diagnosis, achieving a higher diagnostic accuracy.
- 5 July 2026
- Published by American Association for the Advancement of Science (AAAS) in Intelligent Computing
- Vol. 5
L2L-ESN: Electrocardiogram Lead Synthesis for Assisting Myocardial Infarction Diagnosis
Wearable electrocardiogram (ECG) monitoring devices, important terminals in the Internet of Medical Things (IoMT), can provide real-time and rapid myocardial infarction (MI) warning services for large populations. This enables early medical intervention and considerably improves the survival rate of MI patients, thereby alleviating the increasing prevalence of cardiovascular diseases. However, the number of leads in wearable ECG devices is insufficient to accurately diagnose complex cardiac conditions. To address this issue, this study proposes the Lead-to-Lead ECG Synthesis Network (L2L-ESN), which is based on a conditional Wasserstein generative adversarial network. It utilizes multihead generation to learn the relationships among leads, synthesize complete 12-lead ECGs from specified leads, and automatically identify the location of MI, thereby maximizing the utilization of available resources and improving diagnostic accuracy under limited-lead constraints, ultimately enabling accurate early warning and timely treatment of MI through wearable devices. The proposed method was evaluated in an IoMT-oriented reduced-lead setting on 2 real-world ECG datasets using waveform similarity metrics, downstream diagnostic tests, qualitative visualization, and deployment profiling. The results demonstrate that L2L-ESN generates more accurate electrocardiographic signals for MI diagnosis, achieving a higher diagnostic accuracy.
- Open Access
- 17 June 2026
- Published by Elsevier in SSRN Electronic Journal
EF-ConsNET: Physiologically Anchored Multimodal Deep Learning for Diagnosis and Severity Grading of Pediatric Myocarditis
Early diagnosis of pediatric myocarditis remains challenging due to non-specific clinical presentations and resource constraints in real-world practice that often lead to incomplete testing. While cardiac magnetic resonance imaging and endomyocardial biopsy serve as reference standards, their routine application is hindered by limited accessibility, high costs, and invasiveness. To address these issues, we propose EF-ConsNET, a robustness-oriented multimodal deep learning framework. This framework integrates digital radiography chest X-ray images, electrocardiogram signals, and structured laboratory and demographic features. Crucially, it utilizes echocardiographic left ventricular ejection fraction as a physiological supervision anchor to constrain multimodal representations at the physiological level and to enhance the consistency of interpretability related to cardiac function. We curated a two-center inpatient cohort comprising 5,147 cases from The First Affiliated Hospital of Zhengzhou University (700 myocarditis and 4,447 non-myocarditis) for internal training and validation, alongside 529 cases from West China Second University Hospital (197 myocarditis and 332 non-myocarditis) for cross-center external validation. In the internal cohort, EF-ConsNET achieved an AUROC of 0.926 and an AUPRC of 0.760 for myocarditis detection, as well as an AUROC of 0.911 and an AUPRC of 0.929 for severity grading (mild vs. severe). The performance remained stable under single-modality missingness simulations of up to 20%, maintaining an AUROC of at least 0.899 across various settings. Following cross-center validation, EF-ConsNET retained moderate transfer performance in the external cohort, achieving an AUROC of 0.812 and an AUPRC of 0.717 for myocarditis detection, and an AUROC of 0.765 and an AUPRC of 0.744 for severity grading. Mechanistic analyses suggest that EF supervision improves cross-modal alignment within the EF-related physiological space, while cardiopulmonary attention enhances interpretability by encouraging the image branch to focus on clinically relevant anatomical regions. Overall, EF-ConsNET provides an accessible, cost-effective, and physiologically grounded approach to support risk stratification and early clinical alerting for pediatric myocarditis.
- 17 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessEF-ConsNET: Physiologically Anchored Multimodal Deep Learning for Diagnosis and Severity Grading of Pediatric Myocarditis
Early diagnosis of pediatric myocarditis remains challenging due to non-specific clinical presentations and resource constraints in real-world practice that often lead to incomplete testing. While cardiac magnetic resonance imaging and endomyocardial biopsy serve as reference standards, their routine application is hindered by limited accessibility, high costs, and invasiveness. To address these issues, we propose EF-ConsNET, a robustness-oriented multimodal deep learning framework. This framework integrates digital radiography chest X-ray images, electrocardiogram signals, and structured laboratory and demographic features. Crucially, it utilizes echocardiographic left ventricular ejection fraction as a physiological supervision anchor to constrain multimodal representations at the physiological level and to enhance the consistency of interpretability related to cardiac function. We curated a two-center inpatient cohort comprising 5,147 cases from The First Affiliated Hospital of Zhengzhou University (700 myocarditis and 4,447 non-myocarditis) for internal training and validation, alongside 529 cases from West China Second University Hospital (197 myocarditis and 332 non-myocarditis) for cross-center external validation. In the internal cohort, EF-ConsNET achieved an AUROC of 0.926 and an AUPRC of 0.760 for myocarditis detection, as well as an AUROC of 0.911 and an AUPRC of 0.929 for severity grading (mild vs. severe). The performance remained stable under single-modality missingness simulations of up to 20%, maintaining an AUROC of at least 0.899 across various settings. Following cross-center validation, EF-ConsNET retained moderate transfer performance in the external cohort, achieving an AUROC of 0.812 and an AUPRC of 0.717 for myocarditis detection, and an AUROC of 0.765 and an AUPRC of 0.744 for severity grading. Mechanistic analyses suggest that EF supervision improves cross-modal alignment within the EF-related physiological space, while cardiopulmonary attention enhances interpretability by encouraging the image branch to focus on clinically relevant anatomical regions. Overall, EF-ConsNET provides an accessible, cost-effective, and physiologically grounded approach to support risk stratification and early clinical alerting for pediatric myocarditis.
- 24 March 2026
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 174
Antinoise Adaptive Time–Frequency Fusion for multivariate time series anomaly detection
- 24 March 2026
- Published by Elsevier in Engineering Applications of Artificial Intelligence
- Vol. 174
Antinoise Adaptive Time–Frequency Fusion for multivariate time series anomaly detection
- 25 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 18 (4) , 2158-2171
Joint Time-Frequency Pseudo Anomalies for Multimodal Electrocardiogram Quality Assessment in Healthcare Service Computing
Electrocardiogram(ECG) signal analysis is crucial in healthcare service computing. Ensuring accurate assessment of ECG signal quality is vital to prevent wastage of transmission bandwidth and ineffective analysis caused by noise. This enables the efficient utilization of service resources. However, existing ECG signal quality assessment(SQA) methods primarily focus on single-modal learning, overlooking the interrelation of ECG in a multimodal feature space and failing to effectively exploit available information for pattern mining. In this paper, we model the SQA for ECG as an anomaly detection problem and propose a multimodal unsupervised SQA method. It jointly explores the boundaries between high-quality ECG and noise in both the time and frequency domains by introducing time-frequency pseudo anomalies. Specifically, we first simulate real ECG noise from the time-domain using a combination of a series of noises and convert it to the frequency-domain to form time-frequency pseudo-anomalies. Next, we map the time-frequency pseudo anomalies onto hyperspheres and jointly refine the hyperspheres learned only from high-quality ECG samples in both feature spaces. Finally, the noise score is defined as the distance from the joint time-frequency features to the center of the hypersphere. Multiple experiments on various real-world ECG datasets validate the superior performance of our proposed method.
- 25 June 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Services Computing
- Vol. 18 (4) , 2158-2171
Joint Time-Frequency Pseudo Anomalies for Multimodal Electrocardiogram Quality Assessment in Healthcare Service Computing
Electrocardiogram(ECG) signal analysis is crucial in healthcare service computing. Ensuring accurate assessment of ECG signal quality is vital to prevent wastage of transmission bandwidth and ineffective analysis caused by noise. This enables the efficient utilization of service resources. However, existing ECG signal quality assessment(SQA) methods primarily focus on single-modal learning, overlooking the interrelation of ECG in a multimodal feature space and failing to effectively exploit available information for pattern mining. In this paper, we model the SQA for ECG as an anomaly detection problem and propose a multimodal unsupervised SQA method. It jointly explores the boundaries between high-quality ECG and noise in both the time and frequency domains by introducing time-frequency pseudo anomalies. Specifically, we first simulate real ECG noise from the time-domain using a combination of a series of noises and convert it to the frequency-domain to form time-frequency pseudo-anomalies. Next, we map the time-frequency pseudo anomalies onto hyperspheres and jointly refine the hyperspheres learned only from high-quality ECG samples in both feature spaces. Finally, the noise score is defined as the distance from the joint time-frequency features to the center of the hypersphere. Multiple experiments on various real-world ECG datasets validate the superior performance of our proposed method.
- 7 September 2024
- Published by Elsevier in Neural Networks
- Vol. 180, 106709
Bidirectional consistency with temporal-aware for semi-supervised time series classification
- 7 September 2024
- Published by Elsevier in Neural Networks
- Vol. 180, 106709
Bidirectional consistency with temporal-aware for semi-supervised time series classification
- Open Access
- 13 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 352-358
PIMGAN: Prior-induced Masked GAN for Myocardial Infarction Anomaly Detection
Myocardial infarction (MI) is a critical heart disease that occurs when the blood supply to the coronary arteries is drastically reduced or interrupted. Deep learning models of MI detection from electrocardiogram (ECG) play an important role in disease early diagnosis. However, most existing methods rely on numerous labeled training samples of MI ECG, which are scarce data in real scenarios. In this paper, we model MI detection as an anomaly detection problem that does not require any MI ECG samples for training. Moreover, inspired by the diagnostic prior knowledge from electrocardiologists, we propose a simple and effective method of unsupervised anomaly detection called PIMGAN, which imposes the model to focus on essential wave features of MI ECG such as Q wave, T wave, and ST segment. Specifically, PIMGAN is a prior-induced mask-based dual generative adversarial network (GAN), where one GAN learns the global feature from the original ECG, and another GAN learns the key MI wave feature from the masked ECG. Extensive experiments on multiple real-world datasets demonstrate the effectiveness of the proposed method compared with the current advanced anomaly detection baselines.
- 13 August 2024
- Published by Association for Computing Machinery (ACM)
- p. 352-358
Open AccessPIMGAN: Prior-induced Masked GAN for Myocardial Infarction Anomaly Detection
Myocardial infarction (MI) is a critical heart disease that occurs when the blood supply to the coronary arteries is drastically reduced or interrupted. Deep learning models of MI detection from electrocardiogram (ECG) play an important role in disease early diagnosis. However, most existing methods rely on numerous labeled training samples of MI ECG, which are scarce data in real scenarios. In this paper, we model MI detection as an anomaly detection problem that does not require any MI ECG samples for training. Moreover, inspired by the diagnostic prior knowledge from electrocardiologists, we propose a simple and effective method of unsupervised anomaly detection called PIMGAN, which imposes the model to focus on essential wave features of MI ECG such as Q wave, T wave, and ST segment. Specifically, PIMGAN is a prior-induced mask-based dual generative adversarial network (GAN), where one GAN learns the global feature from the original ECG, and another GAN learns the key MI wave feature from the masked ECG. Extensive experiments on multiple real-world datasets demonstrate the effectiveness of the proposed method compared with the current advanced anomaly detection baselines.
- 28 May 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 253
Self-supervised multi-transformation learning for time series anomaly detection
- 28 May 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 253
Self-supervised multi-transformation learning for time series anomaly detection
- 3 January 2024
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 170, 107928
Pseudo anomalies enhanced deep support vector data description for electrocardiogram quality assessment
- 3 January 2024
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 170, 107928
Pseudo anomalies enhanced deep support vector data description for electrocardiogram quality assessment
- 1 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-11
KalmanAE: Deep Embedding Optimized Kalman Filter for Time Series Anomaly Detection
The Kalman filter performs well in system state estimation by inferring a joint probability distribution over time variables, which has numerous technological applications in time series analysis. However, the complex Kalman filter parameter settings prevent it from optimally estimating the system state, and this suboptimal estimation makes it difficult to effectively distinguish between normal and anomalous system states in anomaly detection. In this paper, we propose a deep embedding-optimized Kalman filter for unsupervised time series anomaly detection, where the system state of a normal time series can be fitted by the embedding-optimized Kalman filter in an unsupervised manner and anomalies can be detected from data points that deviate from the normal system state. Specifically, we use an autoencoder-enhanced Kalman filter to capture the normal pattern of the time series, where the original time series signal is first fed into the Kalman filter, then the autoencoder encodes the filtered signal and reconstructs the original signal, and the learned embedding from encoder is used to sequentially optimize the Kalman filter. Finally, the optimized filter captures the normal pattern of the time series and the reconstruction error from the filtered signal can be measured to detect anomalies. The validity of the method is verified on real-world time series datasets.
- 1 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 72 (00189456) , 1-11
KalmanAE: Deep Embedding Optimized Kalman Filter for Time Series Anomaly Detection
The Kalman filter performs well in system state estimation by inferring a joint probability distribution over time variables, which has numerous technological applications in time series analysis. However, the complex Kalman filter parameter settings prevent it from optimally estimating the system state, and this suboptimal estimation makes it difficult to effectively distinguish between normal and anomalous system states in anomaly detection. In this paper, we propose a deep embedding-optimized Kalman filter for unsupervised time series anomaly detection, where the system state of a normal time series can be fitted by the embedding-optimized Kalman filter in an unsupervised manner and anomalies can be detected from data points that deviate from the normal system state. Specifically, we use an autoencoder-enhanced Kalman filter to capture the normal pattern of the time series, where the original time series signal is first fed into the Kalman filter, then the autoencoder encodes the filtered signal and reconstructs the original signal, and the learned embedding from encoder is used to sequentially optimize the Kalman filter. Finally, the optimized filter captures the normal pattern of the time series and the reconstruction error from the filtered signal can be measured to detect anomalies. The validity of the method is verified on real-world time series datasets.
- 27 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 21 (4) , 2888-2899
Adaptive-Correlation-Aware Unsupervised Deep Learning for Anomaly Detection in Cyber-Physical Systems
Cyber-Physical System needs high security to ensure the safe operation. Anomaly detection is one of the mainstream security technologies, the core of which is data analysis and learning. Unsupervised Deep-Learning-based Anomaly Detection Methods can be used in the scenarios that collects large amounts of unlabeled data and are more in line with the actual needs of CPS. However, the correlation among data did not attract enough attention to exploring their implicit relationship, and the adaptive training was deficient. Therefore, we propose an Adaptive-Correlation-aware Unsupervised Deep Learning (ACUDL) for anomaly detection in CPS. It constructs a directed graph structure to represent the implicit correlation among data and adaptively updates with dynamic graph; then, designs a dual-autoencoder to extract the original non-correlation, correlation, and reconstruction features, and builds an estimation network using the Gaussian mixture model (GMM) to estimate the anomaly energy. Experimental results on several CPS data scenarios show that ACUDL can be well adapted to many application scenarios with different data characteristics and achieves better overall results than some up-to-date DL-ADMs.
- 27 September 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Dependable and Secure Computing
- Vol. 21 (4) , 2888-2899
Adaptive-Correlation-Aware Unsupervised Deep Learning for Anomaly Detection in Cyber-Physical Systems
Cyber-Physical System needs high security to ensure the safe operation. Anomaly detection is one of the mainstream security technologies, the core of which is data analysis and learning. Unsupervised Deep-Learning-based Anomaly Detection Methods can be used in the scenarios that collects large amounts of unlabeled data and are more in line with the actual needs of CPS. However, the correlation among data did not attract enough attention to exploring their implicit relationship, and the adaptive training was deficient. Therefore, we propose an Adaptive-Correlation-aware Unsupervised Deep Learning (ACUDL) for anomaly detection in CPS. It constructs a directed graph structure to represent the implicit correlation among data and adaptively updates with dynamic graph; then, designs a dual-autoencoder to extract the original non-correlation, correlation, and reconstruction features, and builds an estimation network using the Gaussian mixture model (GMM) to estimate the anomaly energy. Experimental results on several CPS data scenarios show that ACUDL can be well adapted to many application scenarios with different data characteristics and achieves better overall results than some up-to-date DL-ADMs.