Zhenming Yuan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Interests
Zhenming Yuan has a total of 44 co-authors. Here are the top co-authors by number of shared publications:
- Jian Zhang
Peking University, Shenzhen, China
4 shared publications - Sha Lu
Department of Obstetrics and Gynecology, Hangzhou Women’s Hospital (Hangzhou Maternity and Child Health Care Hospital), Hangzhou, People’s Republic of China
3 shared publications - Wensheng Hu
Department of Obstetrics, Women’s Hospital, School of Medicine, Zhejiang University, Hangzhou, China
3 shared publications - Xiaoyan Sun
School of Information Science and Technology, Hangzhou Normal University, Hangzhou, Zhejiang, China
3 shared publications - Jun Yu
Harbin Institute of Technology
2 shared publications Kai Yu
Department of Information Science and Technology, Hangzhou Normal University, Hangzhou, China
2 shared publications
Latest Publications
- Open Access
- 29 October 2026
- Published by MDPI AG in Bioengineering
- Vol. 13 (11) , 1281
Task-Specific Multimodal Fusion of MRI Radiomics, Patient-Specific Biomechanics, and Clinical Data for Noninvasive Liver Fibrosis Assessment
Accurate noninvasive assessment of liver fibrosis requires complementary characterization of tissue structure, biomechanical properties, and systemic disease status. In this retrospective single-center study, we developed a Task-Specific Multimodal Fusion (TSMF) framework integrating MRI radiomics, patient-specific Finite Element Modeling (FEM)-derived biomechanical features, clinical variables, and elastography-derived stiffness information. Data from 90 patients were used for pathology-based fibrosis classification, while subsets of patients with available FibroScan and Magnetic Resonance Elastography (MRE) measurements were further analyzed for continuous estimation of MRE-derived liver stiffness. For fibrosis classification, a clinical auxiliary gated-correction strategy built on a radiomics–FEM backbone achieved an AUROC of 0.818 ± 0.069, outperforming the best dual-modal model (0.794 ± 0.060) and direct tri-modal fusion (0.762 ± 0.096). For MRE stiffness estimation, FibroScan alone provided a strong baseline (R2 = 0.710 ± 0.084; mean absolute error [MAE] = 0.691 ± 0.133 kPa), while prediction-level stacking consistently improved performance over the corresponding direct-fusion approaches. Among the evaluated combinations, stacking of FibroScan and FEM features achieved an R2 of 0.723 ± 0.076 and an MAE of 0.665 ± 0.129 kPa, compared with an R2 of 0.689 ± 0.098 and an MAE of 0.723 ± 0.146 kPa using direct fusion. Notably, incorporating additional modalities did not necessarily improve performance, suggesting that effective multimodal modeling depends more on information complementarity and task-appropriate fusion strategies than on the number of modalities alone. These findings support a task-specific multimodal integration for noninvasive liver fibrosis assessment. The FibroScan-anchored regression framework demonstrated the feasibility of predicting MRE-derived liver stiffness within the study cohort; however, its clinical agreement, generalizability, and potential applicability as an alternative to direct MRE measurement require further validation.
- 29 October 2026
- Published by MDPI AG in Bioengineering
- Vol. 13 (11) , 1281
Open AccessTask-Specific Multimodal Fusion of MRI Radiomics, Patient-Specific Biomechanics, and Clinical Data for Noninvasive Liver Fibrosis Assessment
Accurate noninvasive assessment of liver fibrosis requires complementary characterization of tissue structure, biomechanical properties, and systemic disease status. In this retrospective single-center study, we developed a Task-Specific Multimodal Fusion (TSMF) framework integrating MRI radiomics, patient-specific Finite Element Modeling (FEM)-derived biomechanical features, clinical variables, and elastography-derived stiffness information. Data from 90 patients were used for pathology-based fibrosis classification, while subsets of patients with available FibroScan and Magnetic Resonance Elastography (MRE) measurements were further analyzed for continuous estimation of MRE-derived liver stiffness. For fibrosis classification, a clinical auxiliary gated-correction strategy built on a radiomics–FEM backbone achieved an AUROC of 0.818 ± 0.069, outperforming the best dual-modal model (0.794 ± 0.060) and direct tri-modal fusion (0.762 ± 0.096). For MRE stiffness estimation, FibroScan alone provided a strong baseline (R2 = 0.710 ± 0.084; mean absolute error [MAE] = 0.691 ± 0.133 kPa), while prediction-level stacking consistently improved performance over the corresponding direct-fusion approaches. Among the evaluated combinations, stacking of FibroScan and FEM features achieved an R2 of 0.723 ± 0.076 and an MAE of 0.665 ± 0.129 kPa, compared with an R2 of 0.689 ± 0.098 and an MAE of 0.723 ± 0.146 kPa using direct fusion. Notably, incorporating additional modalities did not necessarily improve performance, suggesting that effective multimodal modeling depends more on information complementarity and task-appropriate fusion strategies than on the number of modalities alone. These findings support a task-specific multimodal integration for noninvasive liver fibrosis assessment. The FibroScan-anchored regression framework demonstrated the feasibility of predicting MRE-derived liver stiffness within the study cohort; however, its clinical agreement, generalizability, and potential applicability as an alternative to direct MRE measurement require further validation.
- 20 December 2025
- Published by Elsevier in Pattern Recognition
- Vol. 174
Divide-and-conquer towards optimal adaptation of pre-trained model to medical tasks
- 20 December 2025
- Published by Elsevier in Pattern Recognition
- Vol. 174
Divide-and-conquer towards optimal adaptation of pre-trained model to medical tasks
- 25 July 2025
- Published by Springer Nature
- p. 260-272
BlastocystMask: An Instance Segmentation of Internal Structure in Human Blastocyst Images
- 25 July 2025
- Published by Springer Nature
- p. 260-272
BlastocystMask: An Instance Segmentation of Internal Structure in Human Blastocyst Images
- 2 May 2025
- Published by Elsevier in Pattern Recognition
- Vol. 167
Self-adaptive image-text fusion for medical image classification
- 2 May 2025
- Published by Elsevier in Pattern Recognition
- Vol. 167
Self-adaptive image-text fusion for medical image classification
- Open Access
- 12 March 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 21 (4) , 1-20
Semi-Supervised RGB-D Hand Gesture Recognition via Mutual Learning of Self-Supervised Models
Human hand gesture recognition is important to human–computer interaction. Gesture recognition based on RGB and Depth (RGB-D) data exploits both RGB and depth images to provide comprehensive results. However, the research under scenario with insufficient annotated data is not adequate. In view of the problem, our insight is to perform self-supervised learning with respect to each modality, transfer the learned information to modality-specific classifiers, and then fuse their results for final decision. To this end, we propose a semi-supervised hand gesture recognition method known as Mutual Learning of Rotation-Aware Gesture Predictors (MLRAGP), which exploits unlabeled training RGB and depth images via self-supervised learning and achieves multi-modal decision fusion through deep mutual learning. For each modality, we rotate both labeled and unlabeled images to fixed angles and train an angle predictor to predict the angles, then we use the feature extraction part of the angle predictor to construct the category predictor and train it through labeled data. We subsequently fuse the category predictors about both modalities by impelling each of them to simulate the probability estimation produced by the other, and making the prediction of labeled images to approach the ground truth annotation. During the training of category predictor and mutual learning, the parameters of feature extractors can be slighted fine-tuned to avoid under-fitting. Experimental results on NTU-Microsoft Kinect Hand Gesture dataset and Washington RGB-D dataset demonstrate the superiority of this framework to existing methods.
- 12 March 2025
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 21 (4) , 1-20
Open AccessSemi-Supervised RGB-D Hand Gesture Recognition via Mutual Learning of Self-Supervised Models
Human hand gesture recognition is important to human–computer interaction. Gesture recognition based on RGB and Depth (RGB-D) data exploits both RGB and depth images to provide comprehensive results. However, the research under scenario with insufficient annotated data is not adequate. In view of the problem, our insight is to perform self-supervised learning with respect to each modality, transfer the learned information to modality-specific classifiers, and then fuse their results for final decision. To this end, we propose a semi-supervised hand gesture recognition method known as Mutual Learning of Rotation-Aware Gesture Predictors (MLRAGP), which exploits unlabeled training RGB and depth images via self-supervised learning and achieves multi-modal decision fusion through deep mutual learning. For each modality, we rotate both labeled and unlabeled images to fixed angles and train an angle predictor to predict the angles, then we use the feature extraction part of the angle predictor to construct the category predictor and train it through labeled data. We subsequently fuse the category predictors about both modalities by impelling each of them to simulate the probability estimation produced by the other, and making the prediction of labeled images to approach the ground truth annotation. During the training of category predictor and mutual learning, the parameters of feature extractors can be slighted fine-tuned to avoid under-fitting. Experimental results on NTU-Microsoft Kinect Hand Gesture dataset and Washington RGB-D dataset demonstrate the superiority of this framework to existing methods.
- Open Access
- 8 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 25372-25385
Intelligent Recommendation for Departments Based on Medical Knowledge Graph
The clinical sub-speciality departments are increasing. It is usually difficult for patients without medical education to choose a suitable department when they need to make an online registration with doctor. A novel approach is presented for departments recommendation in this paper. The ICD codes and symptoms of medical health data are used to describe the patients’ characteristics, diseases, and departments in the quantitative relation. The knowledge graph is built with the relations to automatically recommend departments. The MacBERT-BiLSTM-CRF medical entity recognition model (MacNER) is proposed to identify the patient’s symptoms, parts, treatment methods and drug entities to structure a knowledge graph. The knowledge graph is used to provide knowledge for the intelligent department recommendation. The identified entities are mapped to nodes in the graph with the entity mapping method. Finally, an algorithm named Feature Rate Multiply Converse Disease Rate (FRMCDR) is proposed. The most appropriate department can be recommended by fusing the patient’s chief complaint and past medical history. The experiment shows that our method obtained an 88.77% precision rate in the recommendation.
- 8 March 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 25372-25385
Open AccessIntelligent Recommendation for Departments Based on Medical Knowledge Graph
The clinical sub-speciality departments are increasing. It is usually difficult for patients without medical education to choose a suitable department when they need to make an online registration with doctor. A novel approach is presented for departments recommendation in this paper. The ICD codes and symptoms of medical health data are used to describe the patients’ characteristics, diseases, and departments in the quantitative relation. The knowledge graph is built with the relations to automatically recommend departments. The MacBERT-BiLSTM-CRF medical entity recognition model (MacNER) is proposed to identify the patient’s symptoms, parts, treatment methods and drug entities to structure a knowledge graph. The knowledge graph is used to provide knowledge for the intelligent department recommendation. The identified entities are mapped to nodes in the graph with the entity mapping method. Finally, an algorithm named Feature Rate Multiply Converse Disease Rate (FRMCDR) is proposed. The most appropriate department can be recommended by fusing the patient’s chief complaint and past medical history. The experiment shows that our method obtained an 88.77% precision rate in the recommendation.
- Open Access
- 17 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 112898-112908
Deep Learning-Based Auricular Point Localization for Auriculotherapy
Auriculotherapy is one of the main forms of treatment in Traditional Chinese Medicine, whose potential as an alternative medicine for both health evaluation and disease treatment has been reported in many cases. However, its efficacy highly relies on the accurate localization of auricular points, which are not easy to be remembered due to their complexity. To explore an efficient way of locating auricular points, this study proposed a deep learning-based method of automatically locating auricular points from auricular images. A self-collected dataset named EID was created for TCM auriculotherapy research, with 91 auriculotherapy-related landmark points manually annotated according to the Chinese national standardization. A deep neural network structure was trained for landmark detection, and a direction normalization module was proposed to compensate for the detection error caused by the difference between the left and right ears. The trained model was validated on dataset EID. An average NME of 0.0514±0.0023 was achieved, which outperformed similar works. In addition, a certain auricular area corresponding to the digestive system was segmented based on the localized landmarks, and the results were tested in real-time video streaming. The proposed work for both auricular landmark and area identification can be widely used in auriculotherapy education and applications.
- 17 October 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 10, 112898-112908
Open AccessDeep Learning-Based Auricular Point Localization for Auriculotherapy
Auriculotherapy is one of the main forms of treatment in Traditional Chinese Medicine, whose potential as an alternative medicine for both health evaluation and disease treatment has been reported in many cases. However, its efficacy highly relies on the accurate localization of auricular points, which are not easy to be remembered due to their complexity. To explore an efficient way of locating auricular points, this study proposed a deep learning-based method of automatically locating auricular points from auricular images. A self-collected dataset named EID was created for TCM auriculotherapy research, with 91 auriculotherapy-related landmark points manually annotated according to the Chinese national standardization. A deep neural network structure was trained for landmark detection, and a direction normalization module was proposed to compensate for the detection error caused by the difference between the left and right ears. The trained model was validated on dataset EID. An average NME of 0.0514±0.0023 was achieved, which outperformed similar works. In addition, a certain auricular area corresponding to the digestive system was segmented based on the localized landmarks, and the results were tested in real-time video streaming. The proposed work for both auricular landmark and area identification can be widely used in auriculotherapy education and applications.
- Open Access
- 13 June 2022
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 10 (6) , e33835
The Prediction of Preterm Birth Using Time-Series Technology-Based Machine Learning: Retrospective Cohort Study
Background: Globally, the preterm birth rate has tended to increase over time. Ultrasonography cervical-length assessment is considered to be the most effective screening method for preterm birth, but routine, universal cervical-length screening remains controversial because of its cost. Objective: We used obstetric data to analyze and assess the risk of preterm birth. A machine learning model based on time-series technology was used to analyze regular, repeated obstetric examination records during pregnancy to improve the performance of the preterm birth screening model. Methods: This study attempts to use continuous electronic medical record (EMR) data from pregnant women to construct a preterm birth prediction classifier based on long short-term memory (LSTM) networks. Clinical data were collected from 5187 pregnant Chinese women who gave birth with natural vaginal delivery. The data included more than 25,000 obstetric EMRs from the early trimester to 28 weeks of gestation. The area under the curve (AUC), accuracy, sensitivity, and specificity were used to assess the performance of the prediction model. Results: Compared with a traditional cross-sectional study, the LSTM model in this time-series study had better overall prediction ability and a lower misdiagnosis rate at the same detection rate. Accuracy was 0.739, sensitivity was 0.407, specificity was 0.982, and the AUC was 0.651. Important-feature identification indicated that blood pressure, blood glucose, lipids, uric acid, and other metabolic factors were important factors related to preterm birth. Conclusions: The results of this study will be helpful to the formulation of guidelines for the prevention and treatment of preterm birth, and will help clinicians make correct decisions during obstetric examinations. The time-series model has advantages for preterm birth prediction.
- 13 June 2022
- Published by JMIR Publications Inc. in JMIR Medical Informatics
- Vol. 10 (6) , e33835
Open AccessThe Prediction of Preterm Birth Using Time-Series Technology-Based Machine Learning: Retrospective Cohort Study
Background: Globally, the preterm birth rate has tended to increase over time. Ultrasonography cervical-length assessment is considered to be the most effective screening method for preterm birth, but routine, universal cervical-length screening remains controversial because of its cost. Objective: We used obstetric data to analyze and assess the risk of preterm birth. A machine learning model based on time-series technology was used to analyze regular, repeated obstetric examination records during pregnancy to improve the performance of the preterm birth screening model. Methods: This study attempts to use continuous electronic medical record (EMR) data from pregnant women to construct a preterm birth prediction classifier based on long short-term memory (LSTM) networks. Clinical data were collected from 5187 pregnant Chinese women who gave birth with natural vaginal delivery. The data included more than 25,000 obstetric EMRs from the early trimester to 28 weeks of gestation. The area under the curve (AUC), accuracy, sensitivity, and specificity were used to assess the performance of the prediction model. Results: Compared with a traditional cross-sectional study, the LSTM model in this time-series study had better overall prediction ability and a lower misdiagnosis rate at the same detection rate. Accuracy was 0.739, sensitivity was 0.407, specificity was 0.982, and the AUC was 0.651. Important-feature identification indicated that blood pressure, blood glucose, lipids, uric acid, and other metabolic factors were important factors related to preterm birth. Conclusions: The results of this study will be helpful to the formulation of guidelines for the prevention and treatment of preterm birth, and will help clinicians make correct decisions during obstetric examinations. The time-series model has advantages for preterm birth prediction.
- Open Access
- 3 March 2022
- Published by Frontiers Media SA in Frontiers in Public Health
- Vol. 9, 835960
Improving Alzheimer's Disease Detection for Speech Based on Feature Purification Network
Alzheimer's disease (AD) is a neurodegenerative disease involving the decline of cognitive ability with illness progresses. At present, the diagnosis of AD mainly depends on the interviews between patients and doctors, which is slow, expensive, and subjective, so it is not a better solution to recognize AD using the currently available neuropsychological examinations and clinical diagnostic criteria. A recent study has indicated the potential of language analysis for AD diagnosis. In this study, we proposed a novel feature purification network that can improve the representation learning of transformer model further. Though transformer has made great progress in generating discriminative features because of its long-distance reasoning ability, there is still room for improvement. There exist many common features that are not indicative of any specific class, and we rule out the influence of common features from traditional features extracted by transformer encoder and can get more discriminative features for classification. We apply this method to improve transformer's performance on three public dementia datasets and get improved classification results markedly. Specifically, the method on Pitt datasets gets state-of-the-art (SOTA) result.
- 3 March 2022
- Published by Frontiers Media SA in Frontiers in Public Health
- Vol. 9, 835960
Open AccessImproving Alzheimer's Disease Detection for Speech Based on Feature Purification Network
Alzheimer's disease (AD) is a neurodegenerative disease involving the decline of cognitive ability with illness progresses. At present, the diagnosis of AD mainly depends on the interviews between patients and doctors, which is slow, expensive, and subjective, so it is not a better solution to recognize AD using the currently available neuropsychological examinations and clinical diagnostic criteria. A recent study has indicated the potential of language analysis for AD diagnosis. In this study, we proposed a novel feature purification network that can improve the representation learning of transformer model further. Though transformer has made great progress in generating discriminative features because of its long-distance reasoning ability, there is still room for improvement. There exist many common features that are not indicative of any specific class, and we rule out the influence of common features from traditional features extracted by transformer encoder and can get more discriminative features for classification. We apply this method to improve transformer's performance on three public dementia datasets and get improved classification results markedly. Specifically, the method on Pitt datasets gets state-of-the-art (SOTA) result.
- 28 November 2021
- Published by Springer Nature in Journal of Shanghai Jiaotong University (Science)
- Vol. 27 (2) , 160-167
Spontaneous Language Analysis in Alzheimer’s Disease: Evaluation of Natural Language Processing Technique for Analyzing Lexical Performance
- 28 November 2021
- Published by Springer Nature in Journal of Shanghai Jiaotong University (Science)
- Vol. 27 (2) , 160-167
Spontaneous Language Analysis in Alzheimer’s Disease: Evaluation of Natural Language Processing Technique for Analyzing Lexical Performance