Yu Shen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yu Shen has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Meiyun Wang
Department of Radiology, Zhengzhou University People’s Hospital & Henan Provincial People’s Hospital, Zhengzhou, China
2 shared publications - Qinglei Zhou
School of Computer and Artificial Intelligence, ZhengZhou University, Zhengzhou, China
2 shared publications - Bin Yan
Henan Key Laboratory of Imaging and Intelligent Processing, Information Engineering University, Zhengzhou, China
1 shared publication - Chen-Hong Li
Department of Neurology, Henan University People's Hospital, Zhengzhou, China
1 shared publication - Hongna Tan
Department of Radiology, People’s Hospital of Zhengzhou University, Zhengzhou, Henan, People’s Republic of China
1 shared publication - Jiewen Zhang
Department of Neurology, Henan Provincial People’s Hospital, People’s Hospital of Zhengzhou University, Zhengzhou, China
1 shared publication
Latest Publications
- Open Access
- 7 August 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e92183
A Bilingual Benchmark for Evaluating Diagnostic Performance of Multimodal Large Language Models in Radiology (RadM-Bench): Evaluation Development and Validation
Background: Multimodal large language models are increasingly used in radiological diagnosis, but their performance has not been systematically evaluated across volumetric (3D) imaging, real-world clinical versus public teaching cases, and bilingual contexts. Objective: The aim of the study is to develop a bilingual radiology benchmark and characterize the diagnostic performance of state-of-the-art multimodal large language models across input modality, clinical setting (public teaching vs routine clinical), disease rarity, and clinical-history language and to disentangle linguistic from clinical-content effects through a cross-linguistic control experiment. Methods: We constructed RadM-Bench, comprising 720 cases evenly distributed across 9 radiological subspecialties: 360 English public teaching cases enriched in rare diseases (RadEdu) and 360 Chinese routine clinical cases (RealClin). In total, 4 proprietary models (GPT-4o, O3, Gemini-2-Flash, and Gemini-2.5-Flash-Thinking) and 6 open-source models (Qwen2.5-VL-72B/7B, InternVL3-78B/8B, Llama-4-Scout-17B-16E, and MedGemma-4B) were evaluated under 4 input conditions: clinical history alone, history with radiologist-selected 2D key images, and history with volumetric data sampled at 2 and 10 frames per second (fps). Each response was scored on a 4-tier 0‐3 diagnostic-quality rubric by 2 board-certified radiologists blinded to model identity. Mean scores with bias-corrected and accelerated bootstrap 95% CIs are reported. To disentangle language from clinical content, all 360 RealClin histories were translated into English and re-evaluated, with paired comparisons by Wilcoxon signed-rank tests and Benjamini-Hochberg false-discovery-rate correction. Results: Mean performance remained below 1.5 on the 0‐3 scale for all 10 models on both datasets. Adding radiologist-selected 2D key images to clinical history improved performance in all 10 models (+19.8% to +139.2%). In RealClin at fps=10, all 8 evaluable models scored lower with volumetric input than with the 2D-image baseline (−5.3% to −31.4%); MedGemma-4B and Llama-4-Scout-17B-16E could only be evaluated at fps=2 due to context-window and graphics processing unit–memory constraints. At fps=2, a total of 8 out of 10 models declined (−6.8% to −28.6%), while Qwen2.5-VL-7B and InternVL3-8B showed marginal improvements (+2.4% and +2.1%). Cross-dataset transfer diverged by model category: proprietary models declined from RadEdu to RealClin (eg, O3 with images: 1.14 to 0.79), whereas Chinese-centric open-source models improved (eg, InternVL3-78B: 0.48 to 0.75). The rare-disease premium observed in 9 out of 10 models in RadEdu reversed in RealClin, where common-disease scores exceeded rare-disease scores in 7 out of 10 models under history-only input. Translating RealClin histories into English produced a numerical decrease in mean score for all 10 models, which were statistically significant in 9 out of 10 models after false discovery rate correction, excluding a Chinese-language penalty. Conclusions: Within the scope of this benchmark, multimodal inputs improved performance over clinical history alone, but performance gaps remain in volumetric data processing and cross-context generalization, with mean diagnostic performance across the 10 evaluated models remaining below clinically actionable levels on both datasets.
- 7 August 2026
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 28, e92183
Open AccessA Bilingual Benchmark for Evaluating Diagnostic Performance of Multimodal Large Language Models in Radiology (RadM-Bench): Evaluation Development and Validation
Background: Multimodal large language models are increasingly used in radiological diagnosis, but their performance has not been systematically evaluated across volumetric (3D) imaging, real-world clinical versus public teaching cases, and bilingual contexts. Objective: The aim of the study is to develop a bilingual radiology benchmark and characterize the diagnostic performance of state-of-the-art multimodal large language models across input modality, clinical setting (public teaching vs routine clinical), disease rarity, and clinical-history language and to disentangle linguistic from clinical-content effects through a cross-linguistic control experiment. Methods: We constructed RadM-Bench, comprising 720 cases evenly distributed across 9 radiological subspecialties: 360 English public teaching cases enriched in rare diseases (RadEdu) and 360 Chinese routine clinical cases (RealClin). In total, 4 proprietary models (GPT-4o, O3, Gemini-2-Flash, and Gemini-2.5-Flash-Thinking) and 6 open-source models (Qwen2.5-VL-72B/7B, InternVL3-78B/8B, Llama-4-Scout-17B-16E, and MedGemma-4B) were evaluated under 4 input conditions: clinical history alone, history with radiologist-selected 2D key images, and history with volumetric data sampled at 2 and 10 frames per second (fps). Each response was scored on a 4-tier 0‐3 diagnostic-quality rubric by 2 board-certified radiologists blinded to model identity. Mean scores with bias-corrected and accelerated bootstrap 95% CIs are reported. To disentangle language from clinical content, all 360 RealClin histories were translated into English and re-evaluated, with paired comparisons by Wilcoxon signed-rank tests and Benjamini-Hochberg false-discovery-rate correction. Results: Mean performance remained below 1.5 on the 0‐3 scale for all 10 models on both datasets. Adding radiologist-selected 2D key images to clinical history improved performance in all 10 models (+19.8% to +139.2%). In RealClin at fps=10, all 8 evaluable models scored lower with volumetric input than with the 2D-image baseline (−5.3% to −31.4%); MedGemma-4B and Llama-4-Scout-17B-16E could only be evaluated at fps=2 due to context-window and graphics processing unit–memory constraints. At fps=2, a total of 8 out of 10 models declined (−6.8% to −28.6%), while Qwen2.5-VL-7B and InternVL3-8B showed marginal improvements (+2.4% and +2.1%). Cross-dataset transfer diverged by model category: proprietary models declined from RadEdu to RealClin (eg, O3 with images: 1.14 to 0.79), whereas Chinese-centric open-source models improved (eg, InternVL3-78B: 0.48 to 0.75). The rare-disease premium observed in 9 out of 10 models in RadEdu reversed in RealClin, where common-disease scores exceeded rare-disease scores in 7 out of 10 models under history-only input. Translating RealClin histories into English produced a numerical decrease in mean score for all 10 models, which were statistically significant in 9 out of 10 models after false discovery rate correction, excluding a Chinese-language penalty. Conclusions: Within the scope of this benchmark, multimodal inputs improved performance over clinical history alone, but performance gaps remain in volumetric data processing and cross-context generalization, with mean diagnostic performance across the 10 evaluated models remaining below clinically actionable levels on both datasets.
- Open Access
- 1 February 2026
- Published by AME Publishing Company in Quantitative Imaging in Medicine and Surgery
- Vol. 16 (2) , 137
Structural and Statistical Knowledge-Enhanced Attention Network for early Parkinson’s disease diagnosis
Background: Early diagnosis of Parkinson’s disease (PD) is crucial for prompt treatment and improved clinical outcomes, but accurate early diagnosis remains challenging. Deep learning (DL) methods have demonstrated significant potential for diagnosing neurological diseases using magnetic resonance imaging (MRI). However, most existing models use generic pattern recognition approaches without incorporating the distinctive characteristics of neuroimaging data and domain-specific knowledge. These limitations restrict both diagnostic accuracy (ACC) and clinical interpretability. This diagnostic ACC study aims to develop a specialized DL framework that effectively integrates neuroimaging domain knowledge to enhance both diagnostic performance and clinical interpretability for early PD detection. Methods: In this study, we propose a Structural and Statistical Knowledge-Enhanced Attention Network (SSKEA-Net) for early PD diagnosis, consisting of two innovative cascaded modules: the Gray-White Interactive Modulation (GWIM) module, which utilizes a structurally specific gray-white matter separation mechanism to modulate channel-wise attention and further enhances tissue-specific features; and the Statistical Prior-Guided Attention (SPGA) module, which incorporates voxel-level statistically significant maps as spatial attention weights to guide feature extraction toward disease-related brain regions. By incorporating the domain knowledge-guided explicit feature enhancement strategy, SSKEA-Net effectively reduces feature interference between gray and white matter, significantly enhancing both model performance and interpretability. We evaluated the diagnostic performance of model using diffusion tensor imaging as input on a rigorously matched early-stage PD dataset with strictly controlled age and gender distributions, and employed activation heatmaps to visualize the interpretability of model. Results: SSKEA-Net achieved superior diagnostic performance in five-fold cross-validation, with an ACC of 0.8798±0.0158, positive predictive value of 0.9185±0.0138, true positive rate of 0.8067±0.0267, specificity of 0.9406±0.0091, and an area under the curve of 0.9301±0.0146, outperforming both classic three-dimensional DL models and the current top-performing neuroimaging model, Simple Fully Convolutional Network. Compared to the baseline model, the cumulative activation heatmaps demonstrated that SSKEA-Net achieved precise anatomical localization, with activation specifically concentrated on the substantia nigra, putamen, midbrain, and corpus callosum, which correspond to clinically relevant brain structures associated with early PD, confirming the enhanced interpretability and clinical relevance. Conclusions: The proposed SSKEA-Net effectively combines domain-specific structural and statistical knowledge with DL, achieving both high diagnostic ACC and clinical interpretability for early PD detection. By incorporating neuroimaging priors into the network architecture, this approach provides a valuable framework for developing more reliable and interpretable artificial intelligence systems in clinical neuroimaging applications.
- 1 February 2026
- Published by AME Publishing Company in Quantitative Imaging in Medicine and Surgery
- Vol. 16 (2) , 137
Open AccessStructural and Statistical Knowledge-Enhanced Attention Network for early Parkinson’s disease diagnosis
Background: Early diagnosis of Parkinson’s disease (PD) is crucial for prompt treatment and improved clinical outcomes, but accurate early diagnosis remains challenging. Deep learning (DL) methods have demonstrated significant potential for diagnosing neurological diseases using magnetic resonance imaging (MRI). However, most existing models use generic pattern recognition approaches without incorporating the distinctive characteristics of neuroimaging data and domain-specific knowledge. These limitations restrict both diagnostic accuracy (ACC) and clinical interpretability. This diagnostic ACC study aims to develop a specialized DL framework that effectively integrates neuroimaging domain knowledge to enhance both diagnostic performance and clinical interpretability for early PD detection. Methods: In this study, we propose a Structural and Statistical Knowledge-Enhanced Attention Network (SSKEA-Net) for early PD diagnosis, consisting of two innovative cascaded modules: the Gray-White Interactive Modulation (GWIM) module, which utilizes a structurally specific gray-white matter separation mechanism to modulate channel-wise attention and further enhances tissue-specific features; and the Statistical Prior-Guided Attention (SPGA) module, which incorporates voxel-level statistically significant maps as spatial attention weights to guide feature extraction toward disease-related brain regions. By incorporating the domain knowledge-guided explicit feature enhancement strategy, SSKEA-Net effectively reduces feature interference between gray and white matter, significantly enhancing both model performance and interpretability. We evaluated the diagnostic performance of model using diffusion tensor imaging as input on a rigorously matched early-stage PD dataset with strictly controlled age and gender distributions, and employed activation heatmaps to visualize the interpretability of model. Results: SSKEA-Net achieved superior diagnostic performance in five-fold cross-validation, with an ACC of 0.8798±0.0158, positive predictive value of 0.9185±0.0138, true positive rate of 0.8067±0.0267, specificity of 0.9406±0.0091, and an area under the curve of 0.9301±0.0146, outperforming both classic three-dimensional DL models and the current top-performing neuroimaging model, Simple Fully Convolutional Network. Compared to the baseline model, the cumulative activation heatmaps demonstrated that SSKEA-Net achieved precise anatomical localization, with activation specifically concentrated on the substantia nigra, putamen, midbrain, and corpus callosum, which correspond to clinically relevant brain structures associated with early PD, confirming the enhanced interpretability and clinical relevance. Conclusions: The proposed SSKEA-Net effectively combines domain-specific structural and statistical knowledge with DL, achieving both high diagnostic ACC and clinical interpretability for early PD detection. By incorporating neuroimaging priors into the network architecture, this approach provides a valuable framework for developing more reliable and interpretable artificial intelligence systems in clinical neuroimaging applications.
- 27 October 2025
- Published by SAGE Publications in Journal of Alzheimer’s Disease
- Vol. 108 (3) , 1388-1397
Relationship between striatal dopaminergic depletion and cognitive milestone in Parkinson's disease: A comprehensive analysis of striatal subregions, laterality, and nonlinear relationships
Background: In Parkinson's disease (PD), striatal dopaminergic degeneration follows a posterior-anterior gradient, with asymmetric involvement and an exponential decline in early disease stages. Previous investigations of the relationship between striatal dopamine transporter (DAT) availability and cognitive impairment in PD have been insufficient, especially in longitudinal studies. Objective: To examine the longitudinal associations between baseline DAT availability and cognitive milestone (CogM) from the perspectives of striatal subregions, laterality, and nonlinear threshold effects. Methods: A total of 467 patients with early, sporadic PD without CogM and with available striatal DAT data at baseline were included from the Parkinson's Progression Markers Initiative database. Annual follow-up data over eight years were analyzed. Standard multivariable Cox and restricted cubic spline (RCS) Cox models were used to assess the associations between DAT uptake and CogM. Two-piecewise Cox regression was used to explore potential threshold effects. Results: The 8-year cumulative probability of being CogM-free was 66.0%. In multivariable Cox models, DAT uptake in the ipsilateral, but not contralateral, caudate and putamen was inversely associated with CogM. RCS Cox models revealed a linear negative association for the ipsilateral caudate, but an L-shaped association for the ipsilateral putamen. Two-piecewise Cox regression identified a threshold effect at a DAT uptake of 1.04 in the ipsilateral putamen, below which the risk of CogM increased significantly. Conclusions: Baseline ipsilateral caudate DAT uptake shows a linear inverse relationship with CogM, highlighting its potential as an early predictive biomarker. The L-shaped association observed for ipsilateral putamen DAT warrants further mechanistic investigation.
- 27 October 2025
- Published by SAGE Publications in Journal of Alzheimer’s Disease
- Vol. 108 (3) , 1388-1397
Relationship between striatal dopaminergic depletion and cognitive milestone in Parkinson's disease: A comprehensive analysis of striatal subregions, laterality, and nonlinear relationships
Background: In Parkinson's disease (PD), striatal dopaminergic degeneration follows a posterior-anterior gradient, with asymmetric involvement and an exponential decline in early disease stages. Previous investigations of the relationship between striatal dopamine transporter (DAT) availability and cognitive impairment in PD have been insufficient, especially in longitudinal studies. Objective: To examine the longitudinal associations between baseline DAT availability and cognitive milestone (CogM) from the perspectives of striatal subregions, laterality, and nonlinear threshold effects. Methods: A total of 467 patients with early, sporadic PD without CogM and with available striatal DAT data at baseline were included from the Parkinson's Progression Markers Initiative database. Annual follow-up data over eight years were analyzed. Standard multivariable Cox and restricted cubic spline (RCS) Cox models were used to assess the associations between DAT uptake and CogM. Two-piecewise Cox regression was used to explore potential threshold effects. Results: The 8-year cumulative probability of being CogM-free was 66.0%. In multivariable Cox models, DAT uptake in the ipsilateral, but not contralateral, caudate and putamen was inversely associated with CogM. RCS Cox models revealed a linear negative association for the ipsilateral caudate, but an L-shaped association for the ipsilateral putamen. Two-piecewise Cox regression identified a threshold effect at a DAT uptake of 1.04 in the ipsilateral putamen, below which the risk of CogM increased significantly. Conclusions: Baseline ipsilateral caudate DAT uptake shows a linear inverse relationship with CogM, highlighting its potential as an early predictive biomarker. The L-shaped association observed for ipsilateral putamen DAT warrants further mechanistic investigation.
- 17 November 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 263
A model-based infrared and visible image fusion network with cooperative optimization
- 17 November 2024
- Published by Elsevier in Expert Systems with Applications
- Vol. 263
A model-based infrared and visible image fusion network with cooperative optimization