Guoxin Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Guoxin Wang has a total of 34 co-authors. Here are the top co-authors by number of shared publications:
- Deepu John
University College Dublin, Dublin, Ireland
8 shared publications - Qingyuan Wang
University College Dublin, Dublin, Ireland
5 shared publications - Avishek Nag
School of Computer Science, University College Dublin, Dublin, Ireland
3 shared publications - Ganesh Neelakanta Iyer
National University of Singapore, Cluny Road, Singapore
3 shared publications - Barry Cardiff
University College Dublin
2 shared publications - Xiaolin Li
Beijing University of Technology, Beijing, China
2 shared publications
Latest Publications
- Open Access
- 9 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Translational Engineering in Health and Medicine
- Vol. 14, 285-297
Neural Architecture Search-Based ECG Classification for Low-Power Edge Deployment in IoT Healthcare Systems
Objective: To develop and validate a deployment-centric ECG arrhythmia classification framework for low-power wearable edge devices, with emphasis on efficient on-device screening and reduced communication burden in remote cardiac monitoring workflows, optimized for seamless integration into clinical remote monitoring workflows using resource-constrained wearable IoT edge devices. Methods and procedures: We design a clinically-oriented wearable-to-phone workflow in which the edge device performs on-device ECG inference and transmits actionable anomaly summaries to a smartphone via Bluetooth Low Energy (BLE). This approach reduces latency, energy cost, and privacy risks compared to continuous raw-signal transmission. A NAS process tailored to arrhythmia detection automatically explores candidate architectures under diverse compute and memory constraints to identify feasible models for embedded platforms. Multiple deployment-ready variants are generated via quantization and pruning, and implemented on an ARM Cortex-M33-based system. System efficiency is quantified through end-to-end measurements of energy consumption and data throughput, ensuring its readiness for long-term ambulatory monitoring. Results: The deployed models achieve real-time inference on an ARM Cortex-M33 with as little as 21.7 kB RAM and 265k MACs. After optimization, energy consumption is reduced to 5.51 mJ per inference. Compared with the original model, RAM usage is reduced by 53.6%, with only a 0.5% absolute decrease in accuracy (97.9% to 97.4%). Hardware-in-the-loop power profiling confirms significant efficiency gains, validating the clinical feasibility of the proposed selective offloading strategy. Conclusion: A NAS-guided, deployment-centric pipeline-spanning architecture search, compression/mapping, and embedded power profiling enables accurate, real-time ECG classification within stringent wearable constraints. The selective-offloading strategy bridges the gap between laboratory algorithms and practical cardiac surveillance by providing a high-fidelity, low-power, and clinically-aware solution for continuous arrhythmia screening. Clinical impact: This work supports the design of low-power wearable arrhythmia screening systems that can perform beat-level on-device inference and transmit compact abnormal-event summaries to a companion smartphone for downstream review. By reducing unnecessary wireless transmissions while maintaining timely anomaly reporting, the proposed framework may support battery-aware, privacy-enhancing remote cardiac monitoring in home and outpatient settings. The system is intended as a screening and triage aid rather than a standalone diagnostic tool. Clinical and Translational Impact Statement: An energy-efficient on-wearable ECG arrhythmia classifier sends BLE anomaly summaries to a phone, enabling privacy-preserving home monitoring and scalable integration into remote cardiac surveillance workflows.
- 9 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Translational Engineering in Health and Medicine
- Vol. 14, 285-297
Open AccessNeural Architecture Search-Based ECG Classification for Low-Power Edge Deployment in IoT Healthcare Systems
Objective: To develop and validate a deployment-centric ECG arrhythmia classification framework for low-power wearable edge devices, with emphasis on efficient on-device screening and reduced communication burden in remote cardiac monitoring workflows, optimized for seamless integration into clinical remote monitoring workflows using resource-constrained wearable IoT edge devices. Methods and procedures: We design a clinically-oriented wearable-to-phone workflow in which the edge device performs on-device ECG inference and transmits actionable anomaly summaries to a smartphone via Bluetooth Low Energy (BLE). This approach reduces latency, energy cost, and privacy risks compared to continuous raw-signal transmission. A NAS process tailored to arrhythmia detection automatically explores candidate architectures under diverse compute and memory constraints to identify feasible models for embedded platforms. Multiple deployment-ready variants are generated via quantization and pruning, and implemented on an ARM Cortex-M33-based system. System efficiency is quantified through end-to-end measurements of energy consumption and data throughput, ensuring its readiness for long-term ambulatory monitoring. Results: The deployed models achieve real-time inference on an ARM Cortex-M33 with as little as 21.7 kB RAM and 265k MACs. After optimization, energy consumption is reduced to 5.51 mJ per inference. Compared with the original model, RAM usage is reduced by 53.6%, with only a 0.5% absolute decrease in accuracy (97.9% to 97.4%). Hardware-in-the-loop power profiling confirms significant efficiency gains, validating the clinical feasibility of the proposed selective offloading strategy. Conclusion: A NAS-guided, deployment-centric pipeline-spanning architecture search, compression/mapping, and embedded power profiling enables accurate, real-time ECG classification within stringent wearable constraints. The selective-offloading strategy bridges the gap between laboratory algorithms and practical cardiac surveillance by providing a high-fidelity, low-power, and clinically-aware solution for continuous arrhythmia screening. Clinical impact: This work supports the design of low-power wearable arrhythmia screening systems that can perform beat-level on-device inference and transmit compact abnormal-event summaries to a companion smartphone for downstream review. By reducing unnecessary wireless transmissions while maintaining timely anomaly reporting, the proposed framework may support battery-aware, privacy-enhancing remote cardiac monitoring in home and outpatient settings. The system is intended as a screening and triage aid rather than a standalone diagnostic tool. Clinical and Translational Impact Statement: An energy-efficient on-wearable ECG arrhythmia classifier sends BLE anomaly summaries to a phone, enabling privacy-preserving home monitoring and scalable integration into remote cardiac surveillance workflows.
- Open Access
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-12
Dynamic Model Mixtures for Efficient AI Inference in Real-Time Biomedical Applications
Efficient and accurate biomedical signal analysis is critical for real-time clinical diagnostics and biometric identification, especially in wearable and edge-computing environments. However, most deep learning models struggle to balance predictive accuracy with computational cost, limiting their deployment on resource-constrained devices. To address this challenge, we propose a Dynamic Model Mixture (DMM) framework, which adaptively selects pre-trained models based on feature confidence to balance accuracy and computational complexity. Unlike conventional approaches, DMM dynamically adjusts inference paths, reducing unnecessary computations while maintaining high performance. Experimental results in ECG classification and identification tasks demonstrate that DMM achieves state-of-the-art accuracy while significantly reducing inference-time computational workload, measured in FLOPs. Specifically, DMM reduces FLOPs by 64% in arrhythmia classification (95.51% accuracy) and by 74% in ECG-based identification (97.94% accuracy), outperforming existing deep learning models when processing previously unseen data. Moreover, the incorporation of probability weighting further enhances model selection efficiency. DMM is scalable, requires no retraining of expert models, and is well-suited for real-time ECG applications on embedded systems and edge-cloud platforms.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-12
Open AccessDynamic Model Mixtures for Efficient AI Inference in Real-Time Biomedical Applications
Efficient and accurate biomedical signal analysis is critical for real-time clinical diagnostics and biometric identification, especially in wearable and edge-computing environments. However, most deep learning models struggle to balance predictive accuracy with computational cost, limiting their deployment on resource-constrained devices. To address this challenge, we propose a Dynamic Model Mixture (DMM) framework, which adaptively selects pre-trained models based on feature confidence to balance accuracy and computational complexity. Unlike conventional approaches, DMM dynamically adjusts inference paths, reducing unnecessary computations while maintaining high performance. Experimental results in ECG classification and identification tasks demonstrate that DMM achieves state-of-the-art accuracy while significantly reducing inference-time computational workload, measured in FLOPs. Specifically, DMM reduces FLOPs by 64% in arrhythmia classification (95.51% accuracy) and by 74% in ECG-based identification (97.94% accuracy), outperforming existing deep learning models when processing previously unseen data. Moreover, the incorporation of probability weighting further enhances model selection efficiency. DMM is scalable, requires no retraining of expert models, and is well-suited for real-time ECG applications on embedded systems and edge-cloud platforms.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 16642-16652
Optimal Brain Connection: Toward Efficient Structural Pruning
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28, 16642-16652
Optimal Brain Connection: Toward Efficient Structural Pruning
- Open Access
ORXE: Orchestrating Experts for Dynamically Configurable Efficiency
This paper presents ORXE, a modular and adaptable framework for achieving real-time configurable efficiency in AI models. By leveraging a collection of pre-trained experts with diverse computational costs and performance levels, ORXE dynamically adjusts inference pathways based on the complexity of input samples. Unlike conventional approaches that require complex metamodel training, ORXE achieves high efficiency and flexibility without complicating the development process. The proposed system utilizes a confidence-based gating mechanism to allocate appropriate computational resources for each input. ORXE also supports adjustments to the preference between inference cost and prediction performance across a wide range during runtime. We implemented a training-free ORXE system for image classification tasks, evaluating its efficiency and accuracy across various devices. The results demonstrate that ORXE achieves superior performance compared to individual experts and other dynamic models in most cases. This approach can be extended to other applications, providing a scalable solution for diverse real-world deployment scenarios.
Open AccessORXE: Orchestrating Experts for Dynamically Configurable Efficiency
This paper presents ORXE, a modular and adaptable framework for achieving real-time configurable efficiency in AI models. By leveraging a collection of pre-trained experts with diverse computational costs and performance levels, ORXE dynamically adjusts inference pathways based on the complexity of input samples. Unlike conventional approaches that require complex metamodel training, ORXE achieves high efficiency and flexibility without complicating the development process. The proposed system utilizes a confidence-based gating mechanism to allocate appropriate computational resources for each input. ORXE also supports adjustments to the preference between inference cost and prediction performance across a wide range during runtime. We implemented a training-free ORXE system for image classification tasks, evaluating its efficiency and accuracy across various devices. The results demonstrate that ORXE achieves superior performance compared to individual experts and other dynamic models in most cases. This approach can be extended to other applications, providing a scalable solution for diverse real-world deployment scenarios.
- Top Cited PapersOpen Access
- 9 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 28 (11) , 6606-6618
ECG Biometric Authentication Using Self-Supervised Learning for IoT Edge Sensors
Wearable Internet of Things (IoT) devices are gaining ground for continuous physiological data acquisition and health monitoring. These physiological signals can be used for security applications to achieve continuous authentication and user convenience due to passive data acquisition. This paper investigates an electrocardiogram (ECG) based biometric user authentication system using features derived from the Convolutional Neural Network (CNN) and self-supervised contrastive learning. Contrastive learning enables us to use large unlabeled datasets to train the model and establish its generalizability. We propose approaches enabling the CNN encoder to extract appropriate features that distinguish the user from other subjects. When evaluated using the PTB ECG database with 290 subjects, the proposed technique achieved an authentication accuracy of 99.15%. To test its generalizability, we applied the model to two new datasets, the MIT-BIH Arrhythmia Database and the ECG-ID Database, achieving over 98.5% accuracy without any modifications. Furthermore, we show that repeating the authentication step three times can increase accuracy to nearly 100% for both PTBDB and ECGIDDB. This paper also presents model optimizations for embedded device deployment, which makes the system more relevant to real-world scenarios. To deploy our model in IoT edge sensors, we optimized the model complexity by applying quantization and pruning. The optimized model achieves 98.67% accuracy on PTBDB, with 0.48% accuracy loss and 62.6% CPU cycles compared to the unoptimized model. An accuracy-vs-time-complexity tradeoff analysis is performed, and results are presented for different optimization levels.
- 9 September 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 28 (11) , 6606-6618
Top Cited PapersOpen AccessECG Biometric Authentication Using Self-Supervised Learning for IoT Edge Sensors
Wearable Internet of Things (IoT) devices are gaining ground for continuous physiological data acquisition and health monitoring. These physiological signals can be used for security applications to achieve continuous authentication and user convenience due to passive data acquisition. This paper investigates an electrocardiogram (ECG) based biometric user authentication system using features derived from the Convolutional Neural Network (CNN) and self-supervised contrastive learning. Contrastive learning enables us to use large unlabeled datasets to train the model and establish its generalizability. We propose approaches enabling the CNN encoder to extract appropriate features that distinguish the user from other subjects. When evaluated using the PTB ECG database with 290 subjects, the proposed technique achieved an authentication accuracy of 99.15%. To test its generalizability, we applied the model to two new datasets, the MIT-BIH Arrhythmia Database and the ECG-ID Database, achieving over 98.5% accuracy without any modifications. Furthermore, we show that repeating the authentication step three times can increase accuracy to nearly 100% for both PTBDB and ECGIDDB. This paper also presents model optimizations for embedded device deployment, which makes the system more relevant to real-world scenarios. To deploy our model in IoT edge sensors, we optimized the model complexity by applying quantization and pruning. The optimized model achieves 98.67% accuracy on PTBDB, with 0.48% accuracy loss and 62.6% CPU cycles compared to the unoptimized model. An accuracy-vs-time-complexity tradeoff analysis is performed, and results are presented for different optimization levels.
- Open Access
- 29 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (13) , 21776-21785
A Task-Generic High-Performance Unsupervised Pre-Training Framework for ECG
Electrocardiogram (ECG) feature extraction is becoming increasingly important in the biomedical field. In deep learning studies centred on ECG data, established methods mainly rely on labelled signals. However, obtaining high-quality labelled records from open-source datasets is challenging, whereas many unlabeled data remain underutilized. Unsupervised pre-training methods have shown significant efficacy in utilizing datasets and improving performance. Therefore, incorporating unsupervised pre-training paradigms into ECG feature extraction is a strategic choice. This paper presents an unsupervised pre-training framework based on the Masked Autoencoder for extracting ECG signal features. An encoder-decoder framework is designed to recover artificially masked ECGs. During the reconstruction process, the framework can learn knowledge from masked samples. Benefiting from the utilization of large-scale datasets and the reconstruction design, the framework is general and has a high level of performance across a wide range of tasks. In particular, the framework achieves an accuracy of 95.6% on the MITDB dataset for the ECG arrhythmia classification task and 98.8% on the ECGIDDB dataset for the human identification task. Our method introduces novel strategies for training on large-scale unlabeled data. Empirical evaluations, including multiple downstream tasks and comparisons with the state of the art, confirm the validity of our proposed approach.
- 29 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (13) , 21776-21785
Open AccessA Task-Generic High-Performance Unsupervised Pre-Training Framework for ECG
Electrocardiogram (ECG) feature extraction is becoming increasingly important in the biomedical field. In deep learning studies centred on ECG data, established methods mainly rely on labelled signals. However, obtaining high-quality labelled records from open-source datasets is challenging, whereas many unlabeled data remain underutilized. Unsupervised pre-training methods have shown significant efficacy in utilizing datasets and improving performance. Therefore, incorporating unsupervised pre-training paradigms into ECG feature extraction is a strategic choice. This paper presents an unsupervised pre-training framework based on the Masked Autoencoder for extracting ECG signal features. An encoder-decoder framework is designed to recover artificially masked ECGs. During the reconstruction process, the framework can learn knowledge from masked samples. Benefiting from the utilization of large-scale datasets and the reconstruction design, the framework is general and has a high level of performance across a wide range of tasks. In particular, the framework achieves an accuracy of 95.6% on the MITDB dataset for the ECG arrhythmia classification task and 98.8% on the ECGIDDB dataset for the human identification task. Our method introduces novel strategies for training on large-scale unlabeled data. Empirical evaluations, including multiple downstream tasks and comparisons with the state of the art, confirm the validity of our proposed approach.
- Open Access
Unsupervised Pre-Training Using Masked Autoencoders for ECG Analysis
Unsupervised learning methods have become increasingly important in deep learning due to their demonstrated large utilization of datasets and higher accuracy in computer vision and natural language processing tasks. There is a growing trend to extend unsupervised learning methods to other domains, which helps to utilize a large amount of unlabelled data. This paper proposes an unsupervised pre-training technique based on masked autoencoder (MAE) for electrocardiogram (ECG) signals. In addition, we propose a task-specific fine-tuning to form a complete framework for ECG analysis. The framework is high-level, universal, and not individually adapted to specific model architectures or tasks. Experiments are conducted using various model architectures and large-scale datasets, resulting in an accuracy of 94.39% on the MITDB dataset for ECG arrhythmia classification task. The result shows a better performance for the classification of previously unseen data for the proposed approach compared to fully supervised methods.
Open AccessUnsupervised Pre-Training Using Masked Autoencoders for ECG Analysis
Unsupervised learning methods have become increasingly important in deep learning due to their demonstrated large utilization of datasets and higher accuracy in computer vision and natural language processing tasks. There is a growing trend to extend unsupervised learning methods to other domains, which helps to utilize a large amount of unlabelled data. This paper proposes an unsupervised pre-training technique based on masked autoencoder (MAE) for electrocardiogram (ECG) signals. In addition, we propose a task-specific fine-tuning to form a complete framework for ECG analysis. The framework is high-level, universal, and not individually adapted to specific model architectures or tasks. Experiments are conducted using various model architectures and large-scale datasets, resulting in an accuracy of 94.39% on the MITDB dataset for ECG arrhythmia classification task. The result shows a better performance for the classification of previously unseen data for the proposed approach compared to fully supervised methods.
- Open Access
- 1 January 2018
- Published by Elsevier in Antiviral Research
- Vol. 149, 48-57
A virus-like particle of the hepatitis B virus preS antigen elicits robust neutralizing antibodies and T cell responses in mice
- 1 January 2018
- Published by Elsevier in Antiviral Research
- Vol. 149, 48-57
Open AccessA virus-like particle of the hepatitis B virus preS antigen elicits robust neutralizing antibodies and T cell responses in mice
- Open Access
- 1 January 2015
- Published by Wolters Kluwer Health in Endoscopic Ultrasound
- Vol. 4 (2) , 152-5
Endoscopic ultrasound-assisted cholecystogastrostomy by a novel fully covered metal stent for the treatment of gallbladder stones
An 85-year-old male patient with common bile duct stones and gallbladder stone was admitted to the hospital. Endoscopic ultrasound (EUS)-guided cholecystogastrostomy and the placement of a novel covered mental stent was performed after the endoscopic sphincter ectomy procedure. Two weeks later the stents were removed, and an endoscope was advanced into the gallbladder via the fistula, and cholecystolithotomy was performed. For weeks later gallbladder was assessed by abdominal ultrasound. EUS-guided cholecystogastrostomy with mental stent deployment was successfully performed. Two weeks after the procedure, the fistulas had formed, and the stent were removed. Endoscopic cholecystolithotomy was successfully performed through the fistula. The ultrasound exam of gallbladder 4 weeks later showed no stone remain and satisfactory function. The EUS-guided placement of a novel metal stent was a safe and simple approach to performing an endoscopic cholecystogastrostomy, which can subsequently allow procedures for treating biliary disease, including cholecystolithotomy.
- 1 January 2015
- Published by Wolters Kluwer Health in Endoscopic Ultrasound
- Vol. 4 (2) , 152-5
Open AccessEndoscopic ultrasound-assisted cholecystogastrostomy by a novel fully covered metal stent for the treatment of gallbladder stones
An 85-year-old male patient with common bile duct stones and gallbladder stone was admitted to the hospital. Endoscopic ultrasound (EUS)-guided cholecystogastrostomy and the placement of a novel covered mental stent was performed after the endoscopic sphincter ectomy procedure. Two weeks later the stents were removed, and an endoscope was advanced into the gallbladder via the fistula, and cholecystolithotomy was performed. For weeks later gallbladder was assessed by abdominal ultrasound. EUS-guided cholecystogastrostomy with mental stent deployment was successfully performed. Two weeks after the procedure, the fistulas had formed, and the stent were removed. Endoscopic cholecystolithotomy was successfully performed through the fistula. The ultrasound exam of gallbladder 4 weeks later showed no stone remain and satisfactory function. The EUS-guided placement of a novel metal stent was a safe and simple approach to performing an endoscopic cholecystogastrostomy, which can subsequently allow procedures for treating biliary disease, including cholecystolithotomy.