Wenjie Zhao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Wenjie Zhao has a total of 31 co-authors. Here are the top co-authors by number of shared publications:
- Dong Liang
Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences
6 shared publications - Hairong Zheng
Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China
6 shared publications - Yongfeng Yang
Research Center for Medical AI, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China
5 shared publications - Yunlong Gao
Chinese Academy of Sciences Shenzhen University of Technology: Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences, CHINA
4 shared publications - Zhanli Hu
Research Center for Medical AI, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China
4 shared publications - Zhenxing Huang
Research Center for Medical AI, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China
4 shared publications
Latest Publications
- Open Access
- 2 September 2025
- Published by IOP Publishing in Physics in Medicine & Biology
- Vol. 70 (17) , 175019
Reconstruction of total-body multi parametric images with shortened-duration dynamic [68Ga]Ga-PSMA-11 and [68Ga]Ga-FAPI-04 PET scans
Objective. The lengthy 1 h dynamic positron emission tomography (PET) scans discomfort patients, add motion artifacts, and inflate costs, highlighting the need for tech advancements to reduce scan times. Therefore, we attempted to reconstruct multi-parametric imaging with only 20 min dynamic PET frames through deep learning techniques. Approach. Total-body PET data from 49 subjects (18 [68Ga]Ga-FAPI-04 and 31 [68Ga]Ga-prostate-specific membrane antigen (PSMA)-11) at Renji Hospital were used. We employed the dynamic PET frames prediction, parametric linear fitting and generative adversarial network to form the whole deep learning framework accomplishing reconstruction of multi-parametric images ([68Ga]Ga-FAPI-04: K 1 , DV and v B ; [68Ga]Ga-PSMA-11: K 1 , K i and v B ) from 20 min dynamic PET frames. The peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) were used to evaluate the quality of the synthesized results. Statistical analyses, including Pearson correlation coefficient (PCC) and violin plot, were also used for the reconstruction assessments. Main results. The reconstructed multi-parametric images using the proposed technique closely resembled their full-time parametric counterparts, demonstrating high image quality across various tissue types. The reconstructed images resulting from the [68Ga]Ga-FAPI-04 and [68Ga]Ga-PSMA-11 tracers outperform 5 comparison methods, yielding average PSNR values of 35.5535 dB and 36.4040 dB, along with SSIM scores of 0.9798 and 0.9630, respectively. Additionally, the PCCs within the region of interests exhibited strong correlations (above 0.92) with the ground truth. Moreover, the imaging results in the pancreatic and prostate areas matched well with full-time parametric images. Significance. Our approach shortens the scan duration of dynamic PET from above one hour to 20 min, which greatly improves the experience for patients and reduces motion artifacts and costs.
- 2 September 2025
- Published by IOP Publishing in Physics in Medicine & Biology
- Vol. 70 (17) , 175019
Open AccessReconstruction of total-body multi parametric images with shortened-duration dynamic [68Ga]Ga-PSMA-11 and [68Ga]Ga-FAPI-04 PET scans
Objective. The lengthy 1 h dynamic positron emission tomography (PET) scans discomfort patients, add motion artifacts, and inflate costs, highlighting the need for tech advancements to reduce scan times. Therefore, we attempted to reconstruct multi-parametric imaging with only 20 min dynamic PET frames through deep learning techniques. Approach. Total-body PET data from 49 subjects (18 [68Ga]Ga-FAPI-04 and 31 [68Ga]Ga-prostate-specific membrane antigen (PSMA)-11) at Renji Hospital were used. We employed the dynamic PET frames prediction, parametric linear fitting and generative adversarial network to form the whole deep learning framework accomplishing reconstruction of multi-parametric images ([68Ga]Ga-FAPI-04: K 1 , DV and v B ; [68Ga]Ga-PSMA-11: K 1 , K i and v B ) from 20 min dynamic PET frames. The peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) were used to evaluate the quality of the synthesized results. Statistical analyses, including Pearson correlation coefficient (PCC) and violin plot, were also used for the reconstruction assessments. Main results. The reconstructed multi-parametric images using the proposed technique closely resembled their full-time parametric counterparts, demonstrating high image quality across various tissue types. The reconstructed images resulting from the [68Ga]Ga-FAPI-04 and [68Ga]Ga-PSMA-11 tracers outperform 5 comparison methods, yielding average PSNR values of 35.5535 dB and 36.4040 dB, along with SSIM scores of 0.9798 and 0.9630, respectively. Additionally, the PCCs within the region of interests exhibited strong correlations (above 0.92) with the ground truth. Moreover, the imaging results in the pancreatic and prostate areas matched well with full-time parametric images. Significance. Our approach shortens the scan duration of dynamic PET from above one hour to 20 min, which greatly improves the experience for patients and reduces motion artifacts and costs.
- 7 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 29 (10) , 7373-7386
Multistage Diffusion Model With Phase Error Correction for Fast PET Imaging
Fast PET imaging is clinically important for reducing motion artifacts and improving patient comfort. While recent diffusion-based deep learning methods have shown promise, they often fail to capture the true PET degradation process, suffer from accumulated inference errors, introduce artifacts, and require extensive reconstruction iterations. To address these challenges, we propose a novel multistage diffusion framework tailored for fast PET imaging. At the coarse level, we design a multistage structure to approximate the temporal non-linear PET degradation process in a data-driven manner, using paired PET images collected under different acquisition duration. A Phase Error Correction Network (PECNet) ensures consistency across stages by correcting accumulated deviations. At the fine level, we introduce a deterministic cold diffusion mechanism, which simulates intra-stage degradation through interpolation between known acquisition durations—significantly reducing reconstruction iterations to as few as 10. Evaluations on [68Ga]FAPI and [18F]FDG PET datasets demonstrate the superiority of our approach, achieving peak PSNRs of 36.2 dB and 39.0 dB, respectively, with average SSIMs over 0.97. Our framework offers high-fidelity PET imaging with fewer iterations, making it practical for accelerated clinical imaging.
- 7 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 29 (10) , 7373-7386
Multistage Diffusion Model With Phase Error Correction for Fast PET Imaging
Fast PET imaging is clinically important for reducing motion artifacts and improving patient comfort. While recent diffusion-based deep learning methods have shown promise, they often fail to capture the true PET degradation process, suffer from accumulated inference errors, introduce artifacts, and require extensive reconstruction iterations. To address these challenges, we propose a novel multistage diffusion framework tailored for fast PET imaging. At the coarse level, we design a multistage structure to approximate the temporal non-linear PET degradation process in a data-driven manner, using paired PET images collected under different acquisition duration. A Phase Error Correction Network (PECNet) ensures consistency across stages by correcting accumulated deviations. At the fine level, we introduce a deterministic cold diffusion mechanism, which simulates intra-stage degradation through interpolation between known acquisition durations—significantly reducing reconstruction iterations to as few as 10. Evaluations on [68Ga]FAPI and [18F]FDG PET datasets demonstrate the superiority of our approach, achieving peak PSNRs of 36.2 dB and 39.0 dB, respectively, with average SSIMs over 0.97. Our framework offers high-fidelity PET imaging with fewer iterations, making it practical for accelerated clinical imaging.
- 18 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Radiation and Plasma Medical Sciences
- Vol. 9 (8) , 1062-1073
Dual-Prompt-Enhanced Multiorgan Segmentation Model for Total-Body PET Images
Multi-organ segmentation in total-body PET images is crucial for accurately locating abnormalities and assisting in the observation of corresponding metabolic regions in the human body. Despite the emergence of numerous advanced methods in the field of multi-organ segmentation in recent years, available PET image segmentation techniques remain relatively limited. The complexity and variability of textures in PET images, the varying visibility and contrast of organs due to different metabolic activities, and the challenges posed by blurred organ boundaries in PET images all contribute to the increased difficulty of multi-organ segmentation. In this paper, we propose the Dual-Prompt Enhanced Multi-Organ Segmentation Model (DPESeg) for total-body PET image segmentation. Our approach focuses on enhancing the model’s ability to perceive organ thresholds and shapes by introducing textual and disentangled organ features, thereby improving segmentation accuracy. We validate our model on a dataset of total-body PET images obtained from 110 patients. Both visual and quantitative results demonstrate that DPESeg performs well in the multi-organ segmentation task, with a 2.02% improvement in the Dice coefficient and a 1.90% improvement in the Jaccard index compared to the best-performing comparison algorithm.
- 18 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Radiation and Plasma Medical Sciences
- Vol. 9 (8) , 1062-1073
Dual-Prompt-Enhanced Multiorgan Segmentation Model for Total-Body PET Images
Multi-organ segmentation in total-body PET images is crucial for accurately locating abnormalities and assisting in the observation of corresponding metabolic regions in the human body. Despite the emergence of numerous advanced methods in the field of multi-organ segmentation in recent years, available PET image segmentation techniques remain relatively limited. The complexity and variability of textures in PET images, the varying visibility and contrast of organs due to different metabolic activities, and the challenges posed by blurred organ boundaries in PET images all contribute to the increased difficulty of multi-organ segmentation. In this paper, we propose the Dual-Prompt Enhanced Multi-Organ Segmentation Model (DPESeg) for total-body PET image segmentation. Our approach focuses on enhancing the model’s ability to perceive organ thresholds and shapes by introducing textual and disentangled organ features, thereby improving segmentation accuracy. We validate our model on a dataset of total-body PET images obtained from 110 patients. Both visual and quantitative results demonstrate that DPESeg performs well in the multi-organ segmentation task, with a 2.02% improvement in the Dice coefficient and a 1.90% improvement in the Jaccard index compared to the best-performing comparison algorithm.
- 8 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 44 (10) , 4073-4086
Prompt-Agent-Driven Integration of Foundation Model Priors for Low-Count PET Reconstruction
Low-count Positron Emission Tomography reconstruction is critical for maintaining high imaging quality while minimizing tracer doses and radiation exposure. Although integrating structural information from CT and MR data has been shown to enhance PET reconstruction, this typically requires simultaneous PET and CT/MRI scans, complicating workflows and increasing radiation exposure. Recent advancements in foundation models offer a promising alternative to in-person CT/MRI imaging, potentially overcoming these limitations. However, the use of foundation models’ segmentation masks as semantic guides has been observed to introduce erroneous structures in low-count PET reconstructions. To address this challenge, this work introduces an innovative prompting agent-based framework that dynamically interacts with the foundation model to retrieve and refine priors, minimizing undue influence on the reconstruction process. Specifically, a box agent is designed for single-instance local area information retrieval, while a point agent is introduced to progressively prompt broader semantic structures globally, utilizing history point prompts. Additionally, an MDP paradigm has been developed to address the challenges of utilizing historical point prompts while maintaining the independence required by MDPs. Evaluated on both simulated and real datasets, the proposed method demonstrates superior qualitative and quantitative performance compared to state-of-the-art methods, even those leveraging in-person CT/MRI priors.
- 8 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 44 (10) , 4073-4086
Prompt-Agent-Driven Integration of Foundation Model Priors for Low-Count PET Reconstruction
Low-count Positron Emission Tomography reconstruction is critical for maintaining high imaging quality while minimizing tracer doses and radiation exposure. Although integrating structural information from CT and MR data has been shown to enhance PET reconstruction, this typically requires simultaneous PET and CT/MRI scans, complicating workflows and increasing radiation exposure. Recent advancements in foundation models offer a promising alternative to in-person CT/MRI imaging, potentially overcoming these limitations. However, the use of foundation models’ segmentation masks as semantic guides has been observed to introduce erroneous structures in low-count PET reconstructions. To address this challenge, this work introduces an innovative prompting agent-based framework that dynamically interacts with the foundation model to retrieve and refine priors, minimizing undue influence on the reconstruction process. Specifically, a box agent is designed for single-instance local area information retrieval, while a point agent is introduced to progressively prompt broader semantic structures globally, utilizing history point prompts. Additionally, an MDP paradigm has been developed to address the challenges of utilizing historical point prompts while maintaining the independence required by MDPs. Evaluated on both simulated and real datasets, the proposed method demonstrates superior qualitative and quantitative performance compared to state-of-the-art methods, even those leveraging in-person CT/MRI priors.
- 2 December 2024
- Published by Springer Nature
- p. 284-294
Leveraging Computer Vision for Automatic Modulation Classification: Insights from Spectrum and Constellation Diagram Analysis
- 2 December 2024
- Published by Springer Nature
- p. 284-294
Leveraging Computer Vision for Automatic Modulation Classification: Insights from Spectrum and Constellation Diagram Analysis
- 13 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Radiation and Plasma Medical Sciences
- Vol. 9 (1) , 47-56
Accurate Whole-Brain Segmentation for Bimodal PET/MR Images via a Cross-Attention Mechanism
The PET/MRI system plays a significant role in the functional and anatomical quantification of the brain, providing accurate diagnostic data for a variety of brain disorders. However, most of the current methods for segmenting the brain are based on unimodal MRI and rarely combine structural and functional dual-modality information. Therefore, we aimed to employ deep learning techniques to achieve automatic and accurate segmentation of the whole brain while incorporating functional and anatomical information. To leverage dual-modality information, a novel 3D network with a cross-attention module was proposed to capture the correlation between dual-modality features and improve segmentation accuracy. Moreover, several deep learning methods were employed as comparison measures to evaluate the model performance, with the Dice similarity coefficient (DSC), Jaccard index (JAC), recall, and precision serving as quantitative metrics. Experimental results demonstrated our advantages in whole-brain segmentation, achieving an 85.35% DSC, 77.22% JAC, 88.86% recall, and 84.81% precision, which were better than those comparative methods. In addition, consistent and correlated analyses based on segmentation results also demonstrated that our approach achieved superior performance. In future work, we will try to apply our method to other multimodal tasks, such as PET/CT data analysis.
- 13 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Radiation and Plasma Medical Sciences
- Vol. 9 (1) , 47-56
Accurate Whole-Brain Segmentation for Bimodal PET/MR Images via a Cross-Attention Mechanism
The PET/MRI system plays a significant role in the functional and anatomical quantification of the brain, providing accurate diagnostic data for a variety of brain disorders. However, most of the current methods for segmenting the brain are based on unimodal MRI and rarely combine structural and functional dual-modality information. Therefore, we aimed to employ deep learning techniques to achieve automatic and accurate segmentation of the whole brain while incorporating functional and anatomical information. To leverage dual-modality information, a novel 3D network with a cross-attention module was proposed to capture the correlation between dual-modality features and improve segmentation accuracy. Moreover, several deep learning methods were employed as comparison measures to evaluate the model performance, with the Dice similarity coefficient (DSC), Jaccard index (JAC), recall, and precision serving as quantitative metrics. Experimental results demonstrated our advantages in whole-brain segmentation, achieving an 85.35% DSC, 77.22% JAC, 88.86% recall, and 84.81% precision, which were better than those comparative methods. In addition, consistent and correlated analyses based on segmentation results also demonstrated that our approach achieved superior performance. In future work, we will try to apply our method to other multimodal tasks, such as PET/CT data analysis.
- 28 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 28 (9) , 5447-5458
MMCA-NET: A Multimodal Cross Attention Transformer Network for Nasopharyngeal Carcinoma Tumor Segmentation Based on a Total-Body PET/CT System
Nasopharyngeal carcinoma (NPC) is a malignant tumor primarily treated by radiotherapy. Accurate delineation of the target tumor is essential for improving the effectiveness of radiotherapy. However, the segmentation performance of current models is unsatisfactory due to poor boundaries, large-scale tumor volume variation, and the labor-intensive nature of manual delineation for radiotherapy. In this paper, MMCA-Net, a novel segmentation network for NPC using PET/CT images that incorporates an innovative multimodal cross attention transformer (MCA-Transformer) and a modified U-Net architecture, is introduced to enhance modal fusion by leveraging cross-attention mechanisms between CT and PET data. Our method, tested against ten algorithms via fivefold cross-validation on samples from Sun Yat-sen University Cancer Center and the public HECKTOR dataset, consistently topped all four evaluation metrics with average Dice similarity coefficients of 0.815 and 0.7944, respectively. Furthermore, ablation experiments were conducted to demonstrate the superiority of our method over multiple baseline and variant techniques. The proposed method has promising potential for application in other tasks.
- 28 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Biomedical and Health Informatics
- Vol. 28 (9) , 5447-5458
MMCA-NET: A Multimodal Cross Attention Transformer Network for Nasopharyngeal Carcinoma Tumor Segmentation Based on a Total-Body PET/CT System
Nasopharyngeal carcinoma (NPC) is a malignant tumor primarily treated by radiotherapy. Accurate delineation of the target tumor is essential for improving the effectiveness of radiotherapy. However, the segmentation performance of current models is unsatisfactory due to poor boundaries, large-scale tumor volume variation, and the labor-intensive nature of manual delineation for radiotherapy. In this paper, MMCA-Net, a novel segmentation network for NPC using PET/CT images that incorporates an innovative multimodal cross attention transformer (MCA-Transformer) and a modified U-Net architecture, is introduced to enhance modal fusion by leveraging cross-attention mechanisms between CT and PET data. Our method, tested against ten algorithms via fivefold cross-validation on samples from Sun Yat-sen University Cancer Center and the public HECKTOR dataset, consistently topped all four evaluation metrics with average Dice similarity coefficients of 0.815 and 0.7944, respectively. Furthermore, ablation experiments were conducted to demonstrate the superiority of our method over multiple baseline and variant techniques. The proposed method has promising potential for application in other tasks.