Zongyun Gu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Zongyun Gu has a total of 34 co-authors. Here are the top co-authors by number of shared publications:
- Dengdi Sun
School of Artificial Intelligence, Anhui University, Hefei, Anhui, China
4 shared publications - Changxu Dong
School of Artificial Intelligence, Anhui University, Hefei 230601, P. R. China
3 shared publications - Yan Li
Department of Joint Surgery, Hefei First People’s Hospital, Hefei, China
2 shared publications - Zijian Wang
School of Computer Science and Information Engineering and the Intelligent Interconnected Systems Laboratory of Anhui Province, Hefei University of Technology, Hefei, Anhui, China
2 shared publications - Bin Luo
State Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China
1 shared publication - Bin Luo
Sichuan University
1 shared publication
Latest Publications
- 21 May 2026
- Published by World Scientific Pub Co Pte Ltd in International Journal of Neural Systems
- Vol. 36 (11) , 2650041
Adaptive Multiscale Spatiotemporal Mixing Network for Multiview Seizure Detection
Epileptic seizure detection from Electroencephalography (EEG) signals is challenging due to their complex temporal dynamics and intricate inter-channel dependencies. A key difficulty lies in capturing multiscale temporal features that span different time ranges. Seizure-related EEG patterns include rapidly varying micro-scale features and longer-duration macro-scale features, and modeling them jointly often leads to feature interference, hindering accurate temporal representation. In addition, many existing approaches fail to effectively capture spatial relationships between brain regions, further limiting detection performance. To cope with issues above, we design the Adaptive Multiscale Spatiotemporal Mixing Network (AMSMN). The framework first decomposes EEG signals into macro- and micro-scale sequences, which are processed independently across multiple temporal resolutions to reduce cross-scale interference and better represent temporal dynamics. A spatial attention mechanism then fuses the decomposed features, ensuring that important inter-channel information is preserved. Finally, an Informer-based sparse attention layer captures long-range dependencies, allowing the model to focus on the most relevant global interactions across brain regions. We carry out experiments on two available databases demonstrate that AMSMN consistently achieves superior performance to prior methods in both patient-specific and cross-patient settings. The results confirm that the proposed framework improves seizure detection accuracy, robustness, and generalization by effectively integrating multiscale temporal modeling with global spatial dependency extraction. This work advances EEG-based seizure detection by enabling precise multiscale temporal analysis and efficient global dependency modeling, offering strong performance and generalizability for clinical applications.
- 21 May 2026
- Published by World Scientific Pub Co Pte Ltd in International Journal of Neural Systems
- Vol. 36 (11) , 2650041
Adaptive Multiscale Spatiotemporal Mixing Network for Multiview Seizure Detection
Epileptic seizure detection from Electroencephalography (EEG) signals is challenging due to their complex temporal dynamics and intricate inter-channel dependencies. A key difficulty lies in capturing multiscale temporal features that span different time ranges. Seizure-related EEG patterns include rapidly varying micro-scale features and longer-duration macro-scale features, and modeling them jointly often leads to feature interference, hindering accurate temporal representation. In addition, many existing approaches fail to effectively capture spatial relationships between brain regions, further limiting detection performance. To cope with issues above, we design the Adaptive Multiscale Spatiotemporal Mixing Network (AMSMN). The framework first decomposes EEG signals into macro- and micro-scale sequences, which are processed independently across multiple temporal resolutions to reduce cross-scale interference and better represent temporal dynamics. A spatial attention mechanism then fuses the decomposed features, ensuring that important inter-channel information is preserved. Finally, an Informer-based sparse attention layer captures long-range dependencies, allowing the model to focus on the most relevant global interactions across brain regions. We carry out experiments on two available databases demonstrate that AMSMN consistently achieves superior performance to prior methods in both patient-specific and cross-patient settings. The results confirm that the proposed framework improves seizure detection accuracy, robustness, and generalization by effectively integrating multiscale temporal modeling with global spatial dependency extraction. This work advances EEG-based seizure detection by enabling precise multiscale temporal analysis and efficient global dependency modeling, offering strong performance and generalizability for clinical applications.
- 30 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. PP (21682194) , 1-14
Attention-based Multimodal Spatiotemporal Enhanced Interaction Network For Major Depressive Disorder Detection
Although deep learning models have shown promising results in detecting major depressive disorder (MDD), two main limitations remain: insufficient exploitation of interactive information across multimodal brain networks and a lack of adaptive mechanisms for capturing crucial spatiotemporal dependencies among brain regions. To address these challenges, we propose the Attention-based Multimodal Spatiotemporal Enhanced Interaction Network (AM-SEIN) for MDD detection. Specifically, to tackle the first challenge, we integrate structural information from 3D structural magnetic resonance imaging (sMRI) with functional temporal data from functional magnetic resonance imaging (fMRI). Additionally, we design the Cross-Modal Interaction Network (CMIN) and fusion layer to enhance mutual information aggregation and facilitate interactive guidance between the two modalities. For the second challenge, we develop an attention-based adaptive spatiotemporal feature-extracting architecture for both modalities, incorporating the fMRI-based Adaptive Spatiotemporal Fusion (fASF) and the sMRI-based Regional-Level Content-Dependent (sRLCD) modules. This approach enables the effective encoding of inter-regional interactions relevant to MDD detection. Finally, the proposed AM-SEIN is evaluated on the Rest-meta-MDD(RMM) and Rest-meta-MDD-V2(RMM-V2) datasets, achieving state-of-the-art performance.
- 30 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. PP (21682194) , 1-14
Attention-based Multimodal Spatiotemporal Enhanced Interaction Network For Major Depressive Disorder Detection
Although deep learning models have shown promising results in detecting major depressive disorder (MDD), two main limitations remain: insufficient exploitation of interactive information across multimodal brain networks and a lack of adaptive mechanisms for capturing crucial spatiotemporal dependencies among brain regions. To address these challenges, we propose the Attention-based Multimodal Spatiotemporal Enhanced Interaction Network (AM-SEIN) for MDD detection. Specifically, to tackle the first challenge, we integrate structural information from 3D structural magnetic resonance imaging (sMRI) with functional temporal data from functional magnetic resonance imaging (fMRI). Additionally, we design the Cross-Modal Interaction Network (CMIN) and fusion layer to enhance mutual information aggregation and facilitate interactive guidance between the two modalities. For the second challenge, we develop an attention-based adaptive spatiotemporal feature-extracting architecture for both modalities, incorporating the fMRI-based Adaptive Spatiotemporal Fusion (fASF) and the sMRI-based Regional-Level Content-Dependent (sRLCD) modules. This approach enables the effective encoding of inter-regional interactions relevant to MDD detection. Finally, the proposed AM-SEIN is evaluated on the Rest-meta-MDD(RMM) and Rest-meta-MDD-V2(RMM-V2) datasets, achieving state-of-the-art performance.
- 7 October 2025
- Published by World Scientific Pub Co Pte Ltd in International Journal of Neural Systems
- Vol. 36 (01) , 2550056
A Unified Hypergraph-Mamba Framework for Adaptive Electroencephalogram Modeling in Multi-view Seizure Prediction
Seizure prediction from Electroencephalogram (EEG) signals is a critical task for proactive intervention in epilepsy management. Existing models often struggle to capture high-order inter-channel dependencies dynamically and adapt to the spectral variations preceding seizure onset, especially in cross-patient scenarios. To address these issues, a novel Unified Hypergraph-Mamba (UHM) framework, which for the first time integrates hypergraph-based spatial modeling with Mamba-based adaptive spectral modeling. Specifically, a hypergraph attention mechanism is designed to capture high-order spatial interactions among EEG channels, enabling dynamic representation of inter-channel dependencies. Concurrently, an adaptive spectral modeling module based on the Mamba architecture selectively emphasizes frequency components most indicative of preictal states. Together, these components form a unified architecture capable of jointly modeling spatiotemporal EEG dynamics. Extensive experiments conducted on both patient-specific and cross-patient settings demonstrate that our model consistently outperforms state-of-the-art baselines, achieving superior sensitivity and AUC.
- 7 October 2025
- Published by World Scientific Pub Co Pte Ltd in International Journal of Neural Systems
- Vol. 36 (01) , 2550056
A Unified Hypergraph-Mamba Framework for Adaptive Electroencephalogram Modeling in Multi-view Seizure Prediction
Seizure prediction from Electroencephalogram (EEG) signals is a critical task for proactive intervention in epilepsy management. Existing models often struggle to capture high-order inter-channel dependencies dynamically and adapt to the spectral variations preceding seizure onset, especially in cross-patient scenarios. To address these issues, a novel Unified Hypergraph-Mamba (UHM) framework, which for the first time integrates hypergraph-based spatial modeling with Mamba-based adaptive spectral modeling. Specifically, a hypergraph attention mechanism is designed to capture high-order spatial interactions among EEG channels, enabling dynamic representation of inter-channel dependencies. Concurrently, an adaptive spectral modeling module based on the Mamba architecture selectively emphasizes frequency components most indicative of preictal states. Together, these components form a unified architecture capable of jointly modeling spatiotemporal EEG dynamics. Extensive experiments conducted on both patient-specific and cross-patient settings demonstrate that our model consistently outperforms state-of-the-art baselines, achieving superior sensitivity and AUC.
- Open Access
- 24 July 2025
- Published by Frontiers Media SA in Frontiers in Medicine
- Vol. 12, 1600736
Med-DGTN: Dynamic Graph Transformer with Adaptive Wavelet Fusion for multi-label medical image classification
Introduction: Multi-label classification of medical imaging data aims to enable simultaneous identification and diagnosis of multiple diseases, delivering comprehensive clinical decision support for complex conditions. Current methodologies demonstrate limitations in capturing disease co-occurrence patterns and preserving subtle pathological signatures. To address these challenges, we propose Med-DGTN, a dynamically integrated framework designed to advance multi-label classification performance in clinical imaging analytics.Methods: The proposed Med-DGTN (Dynamic Graph Transformer Network with Adaptive Wavelet Fusion) introduces three key innovations: (1) A cross-modal alignment mechanism integrating convolutional visual patterns with graph-based semantic dependencies through conditionally reweighted adjacency matrices; (2) Wavelet-transform-enhanced dense blocks (WTDense) employing multi-frequency decomposition to amplify low-frequency pathological biomarkers; (3) An adaptive fusion architecture optimizing multi-scale feature hierarchies across spatial and spectral domains.Results: Validated on two public medical imaging benchmarks, Med-DGTN demonstrates superior performance across modalities: (1) Achieving a mean average precision (mAP) of 70.65% on the retinal imaging dataset (MuReD2022), surpassing previous state-of-the-art methods by 2.68 percentage points. (2) On the chest X-ray dataset (ChestXray14), Med-DGTN achieves an average Area Under the Curve (AUC) of 0.841. It outperforms prior state-of-the-art methods in 5 of 14 disease categories.Discussion: This investigation establishes that joint modeling of dynamic disease correlations and wavelet-optimized feature representation significantly enhances multi-label diagnostic capabilities. Med-DGTN’s architecture demonstrates clinical translatability by revealing disease interaction patterns through interpretable graph structures, potentially informing precision diagnostics in multi-morbidity scenarios.
- 24 July 2025
- Published by Frontiers Media SA in Frontiers in Medicine
- Vol. 12, 1600736
Open AccessMed-DGTN: Dynamic Graph Transformer with Adaptive Wavelet Fusion for multi-label medical image classification
Introduction: Multi-label classification of medical imaging data aims to enable simultaneous identification and diagnosis of multiple diseases, delivering comprehensive clinical decision support for complex conditions. Current methodologies demonstrate limitations in capturing disease co-occurrence patterns and preserving subtle pathological signatures. To address these challenges, we propose Med-DGTN, a dynamically integrated framework designed to advance multi-label classification performance in clinical imaging analytics.Methods: The proposed Med-DGTN (Dynamic Graph Transformer Network with Adaptive Wavelet Fusion) introduces three key innovations: (1) A cross-modal alignment mechanism integrating convolutional visual patterns with graph-based semantic dependencies through conditionally reweighted adjacency matrices; (2) Wavelet-transform-enhanced dense blocks (WTDense) employing multi-frequency decomposition to amplify low-frequency pathological biomarkers; (3) An adaptive fusion architecture optimizing multi-scale feature hierarchies across spatial and spectral domains.Results: Validated on two public medical imaging benchmarks, Med-DGTN demonstrates superior performance across modalities: (1) Achieving a mean average precision (mAP) of 70.65% on the retinal imaging dataset (MuReD2022), surpassing previous state-of-the-art methods by 2.68 percentage points. (2) On the chest X-ray dataset (ChestXray14), Med-DGTN achieves an average Area Under the Curve (AUC) of 0.841. It outperforms prior state-of-the-art methods in 5 of 14 disease categories.Discussion: This investigation establishes that joint modeling of dynamic disease correlations and wavelet-optimized feature representation significantly enhances multi-label diagnostic capabilities. Med-DGTN’s architecture demonstrates clinical translatability by revealing disease interaction patterns through interpretable graph structures, potentially informing precision diagnostics in multi-morbidity scenarios.
- Open Access
- 3 June 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-16
Enhanced nuclear information fusion and visual transformer for pathological breast cancer image classification
Breast cancer poses a significant threat to women’s health. Early diagnosis using pathological images is crucial for effective treatment planning. However, the low resolution of pathological images poses significant challenges for the extraction of valid information, while their high complexity greatly increases the difficulty of image analysis. To address these challenges, this paper introduces an innovative classification method for breast cancer histopathological images, combining enhanced nuclear information with an Enhanced Vision Transformer (EVT) model using wavelet position embedding. The quintessence of the proposed method resides in its capacity to efficiently extract both biological and foundational image features from pathological images. This is accomplished by initially enhancing nuclear information through the application of segmentation models and sophisticated image processing techniques. Subsequently, wavelet positional embedding within the EVT model is leveraged to precisely capture key information embedded within the images. Experimental outcomes have demonstrated that our method attains an accuracy rate of 94.61% and an AUC value of 99.07% on the BreaKHis dataset, significantly outperforming other baseline network models in terms of classification efficacy. Furthermore, through visual representation, this study underscores the significance of nuclear information enhancement and wavelet position transformation in the EVT model, thereby further confirming the effectiveness and effectiveness of the method we proposed.
- 3 June 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-16
Open AccessEnhanced nuclear information fusion and visual transformer for pathological breast cancer image classification
Breast cancer poses a significant threat to women’s health. Early diagnosis using pathological images is crucial for effective treatment planning. However, the low resolution of pathological images poses significant challenges for the extraction of valid information, while their high complexity greatly increases the difficulty of image analysis. To address these challenges, this paper introduces an innovative classification method for breast cancer histopathological images, combining enhanced nuclear information with an Enhanced Vision Transformer (EVT) model using wavelet position embedding. The quintessence of the proposed method resides in its capacity to efficiently extract both biological and foundational image features from pathological images. This is accomplished by initially enhancing nuclear information through the application of segmentation models and sophisticated image processing techniques. Subsequently, wavelet positional embedding within the EVT model is leveraged to precisely capture key information embedded within the images. Experimental outcomes have demonstrated that our method attains an accuracy rate of 94.61% and an AUC value of 99.07% on the BreaKHis dataset, significantly outperforming other baseline network models in terms of classification efficacy. Furthermore, through visual representation, this study underscores the significance of nuclear information enhancement and wavelet position transformation in the EVT model, thereby further confirming the effectiveness and effectiveness of the method we proposed.
- Open Access
- 16 August 2024
- Published by Wolters Kluwer Health in Medicine
- Vol. 103 (33) , e39370
Research on privacy protection in the context of healthcare data based on knowledge map
With the rapid development of emerging information technologies such as artificial intelligence, cloud computing, and the Internet of Things, the world has entered the era of big data. In the face of growing medical big data, research on the privacy protection of personal information has attracted more and more attention, but few studies have analyzed and forecasted the research hotspots and future development trends on the privacy protection. Presently, to systematically and comprehensively summarize the relevant privacy protection literature in the context of big healthcare data, a bibliometric analysis was conducted to clarify the spatial and temporal distribution and research hotspots of privacy protection using the information visualization software CiteSpace. The literature papers related to privacy protection in the Web of Science were collected from 2012 to 2023. Through analysis of the time, author and countries distribution of relevant publications, we found that after 2013, research on the privacy protection has received increasing attention and the core institution of privacy protection research is the university, but the countries show weak cooperation. Additionally, keywords like privacy, big data, internet, challenge, care, and information have high centralities and frequency, indicating the research hotspots and research trends in the field of the privacy protection. All the findings will provide a comprehensive privacy protection research knowledge structure for scholars in the field of privacy protection research under the background of health big data, which can help them quickly grasp the research hotspots and choose future research projects.
- 16 August 2024
- Published by Wolters Kluwer Health in Medicine
- Vol. 103 (33) , e39370
Open AccessResearch on privacy protection in the context of healthcare data based on knowledge map
With the rapid development of emerging information technologies such as artificial intelligence, cloud computing, and the Internet of Things, the world has entered the era of big data. In the face of growing medical big data, research on the privacy protection of personal information has attracted more and more attention, but few studies have analyzed and forecasted the research hotspots and future development trends on the privacy protection. Presently, to systematically and comprehensively summarize the relevant privacy protection literature in the context of big healthcare data, a bibliometric analysis was conducted to clarify the spatial and temporal distribution and research hotspots of privacy protection using the information visualization software CiteSpace. The literature papers related to privacy protection in the Web of Science were collected from 2012 to 2023. Through analysis of the time, author and countries distribution of relevant publications, we found that after 2013, research on the privacy protection has received increasing attention and the core institution of privacy protection research is the university, but the countries show weak cooperation. Additionally, keywords like privacy, big data, internet, challenge, care, and information have high centralities and frequency, indicating the research hotspots and research trends in the field of the privacy protection. All the findings will provide a comprehensive privacy protection research knowledge structure for scholars in the field of privacy protection research under the background of health big data, which can help them quickly grasp the research hotspots and choose future research projects.
- Top Cited PapersOpen Access
- 3 January 2023
- Published by Wiley in Computational Intelligence and Neuroscience
- Vol. 2023 (1) , 1305583
Classification of Diabetic Retinopathy Severity in Fundus Images Using the Vision Transformer and Residual Attention
Diabetic retinopathy (DR) is a common retinal vascular disease, which can cause severe visual impairment. It is of great clinical significance to use fundus images for intelligent diagnosis of DR. In this paper, an intelligent DR classification model of fundus images is proposed. This method can detect all the five stages of DR, including of no DR, mild, moderate, severe, and proliferative. This model is composed of two key modules. FEB, feature extraction block, is mainly used for feature extraction of fundus images, and GPB, grading prediction block, is used to classify the five stages of DR. The transformer in the FEB has more fine‐grained attention that can pay more attention to retinal hemorrhage and exudate areas. The residual attention in the GPB can effectively capture different spatial regions occupied by different classes of objects. Comprehensive experiments on DDR datasets well demonstrate the superiority of our method, and compared with the benchmark method, our method has achieved competitive performance.
- 3 January 2023
- Published by Wiley in Computational Intelligence and Neuroscience
- Vol. 2023 (1) , 1305583
Top Cited PapersOpen AccessClassification of Diabetic Retinopathy Severity in Fundus Images Using the Vision Transformer and Residual Attention
Diabetic retinopathy (DR) is a common retinal vascular disease, which can cause severe visual impairment. It is of great clinical significance to use fundus images for intelligent diagnosis of DR. In this paper, an intelligent DR classification model of fundus images is proposed. This method can detect all the five stages of DR, including of no DR, mild, moderate, severe, and proliferative. This model is composed of two key modules. FEB, feature extraction block, is mainly used for feature extraction of fundus images, and GPB, grading prediction block, is used to classify the five stages of DR. The transformer in the FEB has more fine‐grained attention that can pay more attention to retinal hemorrhage and exudate areas. The residual attention in the GPB can effectively capture different spatial regions occupied by different classes of objects. Comprehensive experiments on DDR datasets well demonstrate the superiority of our method, and compared with the benchmark method, our method has achieved competitive performance.
- Open Access
- 1 January 2022
- Published by Wiley in Complexity
- Vol. 2022 (1)
An Efficient CNN for Hand X‐Ray Overall Scoring of Rheumatoid Arthritis
Rheumatoid arthritis (RA) is a progressive systemic autoimmune disease characterized by inflammation of the joints and surrounding tissues, which seriously affects the life of patients. The Sharp/van der Heijde method has been widely used in clinical evaluation for the RA disease. However, this manual method is time‐consuming and laborious. Even if two radiologists evaluate a specific location, their subjective evaluation may lead to low inter‐rater reliability. Here, we developed an efficient model powered by deep convolutional neural networks to solve these problems and automated the overall scoring on hand X‐rays. The depthwise separable (Dwise) convolution technique is used based on ResNet‐50 due to the high resolution of hand X‐rays. An inverted residual block is introduced to devise a ResNet‐Dwise50 model to enhance the efficiency of the model. The model was trained and tested using bilateral posteroanterior (two‐handed, side by side) images of 3818 patients. The experiment results show the ResNet‐Dwise50 model achieved an MAE of 14.90 and RMSE of 22.01 while ensuring high efficiency. There was no statistical difference between the average scores given by two experienced radiologists and predicted scores from our model.
- 1 January 2022
- Published by Wiley in Complexity
- Vol. 2022 (1)
Open AccessAn Efficient CNN for Hand X‐Ray Overall Scoring of Rheumatoid Arthritis
Rheumatoid arthritis (RA) is a progressive systemic autoimmune disease characterized by inflammation of the joints and surrounding tissues, which seriously affects the life of patients. The Sharp/van der Heijde method has been widely used in clinical evaluation for the RA disease. However, this manual method is time‐consuming and laborious. Even if two radiologists evaluate a specific location, their subjective evaluation may lead to low inter‐rater reliability. Here, we developed an efficient model powered by deep convolutional neural networks to solve these problems and automated the overall scoring on hand X‐rays. The depthwise separable (Dwise) convolution technique is used based on ResNet‐50 due to the high resolution of hand X‐rays. An inverted residual block is introduced to devise a ResNet‐Dwise50 model to enhance the efficiency of the model. The model was trained and tested using bilateral posteroanterior (two‐handed, side by side) images of 3818 patients. The experiment results show the ResNet‐Dwise50 model achieved an MAE of 14.90 and RMSE of 22.01 while ensuring high efficiency. There was no statistical difference between the average scores given by two experienced radiologists and predicted scores from our model.