Xu Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Xu Wang has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Danail Stoyanov
University College London
3 shared publications - Evangelos B. Mazomenos
University College London
3 shared publications - Jialang Xu
UCL Hawkes Institute and the Department of Medical Physics and Biomedical Engineering, University College London, London, U.K.
3 shared publications Qin Li
Beijing Inst Technol, Sch Life Sci, Dept Biomed Engn, Beijing, Peoples R China
3 shared publications- Xiaoming Hu
School of Life Science, Beijing Institute of Technology, Beijing 100081, China
3 shared publications Jiyuan Xiong
Beijing Inst Technol, Sch Life Sci, Dept Biomed Engn, Beijing, Peoples R China
2 shared publications
Latest Publications
- Open Access
- 29 August 2026
- Published by Elsevier in Medical Image Analysis
- Vol. 114
SEDCLIP: Adapting vision-language model for multi-label surgical error detection
- 29 August 2026
- Published by Elsevier in Medical Image Analysis
- Vol. 114
Open AccessSEDCLIP: Adapting vision-language model for multi-label surgical error detection
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 45 (9) , 1
EndoLRMGS: Combining Large Reconstruction Modelling and Gaussian Splatting for Complete Endoscopic Scene Reconstruction
Reconstructing dynamic surgical scenes from endoscopic videos remains a fundamental challenge in robot-assisted surgery. Existing methods primarily focus on deformable tissues, overlooking the presence of articulated instruments. To bridge this gap, we present EndoLRMGS, the first unified framework capable of reconstructing both deformable tissue and articulated instruments in a modular approach from monocular video and depth priors. We introduce Frequency-Modulated Gaussian Splatting (FMGS), for deformable tissue reconstruction, which modulates the spatial frequency of Gaussian primitives according to the Nyquist-Shannon sampling theorem, improving the visual fidelity while maintaining robust geometric accuracy. For instrument reconstruction, we leverage the Large Reconstruction Model (LRM) to generate high-quality, watertight 3D models from single images, and introduce a novel Orthographic and Perspective joint Projection Optimization (OPjPO) module to recover metric scale and spatial alignment. Extensive experiments on public datasets demonstrate the effectiveness of EndoLRMGS, achieving PSNR values from 28.4981 to 38.4179, with Chamfer distance ranging from 1.43 to 4.71 mm in tissue reconstruction. For instrument reconstruction, EndoLRMGS achieves PSNR values ranging from 19.7341 to 22.4393 on left views and 17.7442 to 20.8436 on right views. In terms of spatial alignment accuracy, it attains IoU values between 71.56% and 85.82%, with Chamfer distance ranging from 12.59 to 17.38 mm. These results highlight EndoLRMGS as a powerful and versatile solution for accurate, complete, and photorealistic 3D reconstruction of surgical scenes. Code is available at: EndoLRMGS.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Medical Imaging
- Vol. 45 (9) , 1
EndoLRMGS: Combining Large Reconstruction Modelling and Gaussian Splatting for Complete Endoscopic Scene Reconstruction
Reconstructing dynamic surgical scenes from endoscopic videos remains a fundamental challenge in robot-assisted surgery. Existing methods primarily focus on deformable tissues, overlooking the presence of articulated instruments. To bridge this gap, we present EndoLRMGS, the first unified framework capable of reconstructing both deformable tissue and articulated instruments in a modular approach from monocular video and depth priors. We introduce Frequency-Modulated Gaussian Splatting (FMGS), for deformable tissue reconstruction, which modulates the spatial frequency of Gaussian primitives according to the Nyquist-Shannon sampling theorem, improving the visual fidelity while maintaining robust geometric accuracy. For instrument reconstruction, we leverage the Large Reconstruction Model (LRM) to generate high-quality, watertight 3D models from single images, and introduce a novel Orthographic and Perspective joint Projection Optimization (OPjPO) module to recover metric scale and spatial alignment. Extensive experiments on public datasets demonstrate the effectiveness of EndoLRMGS, achieving PSNR values from 28.4981 to 38.4179, with Chamfer distance ranging from 1.43 to 4.71 mm in tissue reconstruction. For instrument reconstruction, EndoLRMGS achieves PSNR values ranging from 19.7341 to 22.4393 on left views and 17.7442 to 20.8436 on right views. In terms of spatial alignment accuracy, it attains IoU values between 71.56% and 85.82%, with Chamfer distance ranging from 12.59 to 17.38 mm. These results highlight EndoLRMGS as a powerful and versatile solution for accurate, complete, and photorealistic 3D reconstruction of surgical scenes. Code is available at: EndoLRMGS.
- 1 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (10) , 10682-10689
StereoMamba: Real-Time and Robust Intraoperative Stereo Disparity Estimation via Long-Range Spatial Dependencies
Stereo disparity estimation is crucial for obtaining depth information in robot-assisted minimally invasive surgery (RAMIS). While current deep learning methods have made significant advancements, challenges remain in achieving an optimal balance between accuracy, robustness, and inference speed. To address these challenges, we propose the StereoMamba architecture, which is specifically designed for stereo disparity estimation in RAMIS. Our approach is based on a novel Feature Extraction Mamba (FE-Mamba) module, which enhances long-range spatial dependencies both within and across stereo images. To effectively integrate multi-scale features from FE-Mamba, we then introduce a novel Multidimensional Feature Fusion (MFF) module. Experiments against the state-of-the-art on the ex-vivo SCARED benchmark demonstrate that StereoMamba achieves superior performance on EPE of 2.64 px and depth MAE of 2.55 mm, the second-best performance on Bad2 of 41.49% and Bad3 of 26.99%, while maintaining an inference speed of 21.28 FPS for a pair of high-resolution images (1280 × 1024), striking the optimum balance between accuracy, robustness, and efficiency. Furthermore, by comparing synthesized right images, generated from warping left images using the generated disparity maps, with the actual right image, StereoMamba achieves the best average SSIM (0.8970) and PSNR (16.0761), exhibiting strong zero-shot generalization on the in-vivo RIS2017 and StereoMIS datasets.
- 1 September 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (10) , 10682-10689
StereoMamba: Real-Time and Robust Intraoperative Stereo Disparity Estimation via Long-Range Spatial Dependencies
Stereo disparity estimation is crucial for obtaining depth information in robot-assisted minimally invasive surgery (RAMIS). While current deep learning methods have made significant advancements, challenges remain in achieving an optimal balance between accuracy, robustness, and inference speed. To address these challenges, we propose the StereoMamba architecture, which is specifically designed for stereo disparity estimation in RAMIS. Our approach is based on a novel Feature Extraction Mamba (FE-Mamba) module, which enhances long-range spatial dependencies both within and across stereo images. To effectively integrate multi-scale features from FE-Mamba, we then introduce a novel Multidimensional Feature Fusion (MFF) module. Experiments against the state-of-the-art on the ex-vivo SCARED benchmark demonstrate that StereoMamba achieves superior performance on EPE of 2.64 px and depth MAE of 2.55 mm, the second-best performance on Bad2 of 41.49% and Bad3 of 26.99%, while maintaining an inference speed of 21.28 FPS for a pair of high-resolution images (1280 × 1024), striking the optimum balance between accuracy, robustness, and efficiency. Furthermore, by comparing synthesized right images, generated from warping left images using the generated disparity maps, with the actual right image, StereoMamba achieves the best average SSIM (0.8970) and PSNR (16.0761), exhibiting strong zero-shot generalization on the in-vivo RIS2017 and StereoMIS datasets.
- 25 August 2023
- Vol. 39 (4) , 329-336
Legal Theories, Disability Models and Principles of Disability Assessment.
In the personal injury compensation system, the protection and relief of the injured people's rights to life, rights to health, and body rights are generally based on the results of disability assessment. Over the years, with the increased number of personal injury compensation cases, the practice of disability assessment have been greatly developed, and the development of disability assessment standards tends to be mature. However, the lack of basic theories for disability assessment has seriously affected the construction and unification of standards. Starting from the tort legal system of personal injury compensation, this article systematically analyzes the legal theories of disability assessment, and holds that the loss of labor ability is the legal basis for disability assessment in China, and the essence of disability assessment should be understood as the quantitative assessment of an individual's permanent loss of labor ability. This article combines the international disability assessment models and the primary concepts of American Medical Association's Guides to the Evaluation of Permanent Impairment to refine the basic concepts of disability assessment in China, such as impairment, disability, handicap, disabled people and self-care ability, etc. At the same time, it sorts out the critical issues of identification time, promotion principles and compound calculation of multiple injuries in disability assessment. It is expected to be beneficial to the theory and practice of disability assessment in personal injury compensation.
- 25 August 2023
- Vol. 39 (4) , 329-336
Legal Theories, Disability Models and Principles of Disability Assessment.
In the personal injury compensation system, the protection and relief of the injured people's rights to life, rights to health, and body rights are generally based on the results of disability assessment. Over the years, with the increased number of personal injury compensation cases, the practice of disability assessment have been greatly developed, and the development of disability assessment standards tends to be mature. However, the lack of basic theories for disability assessment has seriously affected the construction and unification of standards. Starting from the tort legal system of personal injury compensation, this article systematically analyzes the legal theories of disability assessment, and holds that the loss of labor ability is the legal basis for disability assessment in China, and the essence of disability assessment should be understood as the quantitative assessment of an individual's permanent loss of labor ability. This article combines the international disability assessment models and the primary concepts of American Medical Association's Guides to the Evaluation of Permanent Impairment to refine the basic concepts of disability assessment in China, such as impairment, disability, handicap, disabled people and self-care ability, etc. At the same time, it sorts out the critical issues of identification time, promotion principles and compound calculation of multiple injuries in disability assessment. It is expected to be beneficial to the theory and practice of disability assessment in personal injury compensation.
- 20 December 2022
- Published by Optica Publishing Group in Biomedical Optics Express
- Vol. 14 (1) , 453-466
Three-dimensional image-guided topical photodynamic therapy system with light dosimetry dynamic planning and monitoring
Photodynamic therapy (PDT) has shown significant potential for skin disease treatment. As a key element, light is critical to influencing its treatment outcome, and light dosimetry is an issue of much concern for researchers. However, because of three-dimensional irregularity in shape and patient’s movement during the therapy, irradiance hardly keeps uniform on the lesion and flux measurement remains a challenge. In this work, we report the development of a three-dimensional image-guided PDT system, and the method of dynamic irradiance planning and flux monitoring for lesions in different poses. This system comprises a three-dimensional camera for monitoring patients’ movement during therapy, a computer for data analysis and processing, and a homemade LED array for forming uniform irradiance on lesions. Simulations on lesions of the face and arm show that the proposed system significantly increases effective therapy area, enhances irradiance uniformity, is able to visualize flux on the lesion, and reduces risks of burns during PDT. The developed PDT system is promising for optimizing procedures of PDT and providing better treatment outcomes by delivering controllable irradiance and flux on lesions even when a patient is moving.
- 20 December 2022
- Published by Optica Publishing Group in Biomedical Optics Express
- Vol. 14 (1) , 453-466
Three-dimensional image-guided topical photodynamic therapy system with light dosimetry dynamic planning and monitoring
Photodynamic therapy (PDT) has shown significant potential for skin disease treatment. As a key element, light is critical to influencing its treatment outcome, and light dosimetry is an issue of much concern for researchers. However, because of three-dimensional irregularity in shape and patient’s movement during the therapy, irradiance hardly keeps uniform on the lesion and flux measurement remains a challenge. In this work, we report the development of a three-dimensional image-guided PDT system, and the method of dynamic irradiance planning and flux monitoring for lesions in different poses. This system comprises a three-dimensional camera for monitoring patients’ movement during therapy, a computer for data analysis and processing, and a homemade LED array for forming uniform irradiance on lesions. Simulations on lesions of the face and arm show that the proposed system significantly increases effective therapy area, enhances irradiance uniformity, is able to visualize flux on the lesion, and reduces risks of burns during PDT. The developed PDT system is promising for optimizing procedures of PDT and providing better treatment outcomes by delivering controllable irradiance and flux on lesions even when a patient is moving.
- 24 August 2022
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 22 (23) , 4521-4530
Dual-function microneedle array for efficient photodynamic therapy with transdermal co-delivered light and photosensitizers
Photodynamic therapy (PDT), as a globally accepted method for treating different forms of skin or mucosal disorders, requires efficient co-delivery of photosensitizers and corresponding therapeutic light. The adverse effects of intravenous injection of photosensitizers have been reduced by the development of microneedle arrays for transdermal local photosensitizer delivery. However, the drawbacks of the only available therapeutic light delivery method at the moment, which is directly applying light to the skin surface, are yet to be improved. This study presents a new strategy in which therapeutic light and photosensitizer were transdermally co-delivered into local tissues. A flexible dual-function microneedle array (DfMNA) which contains 400 microneedles was developed. Each microneedle consists of a dissolvable needle tip (140 μm in height) for delivering the photosensitizer and a transparent needle body (660 μm in height) for guiding therapeutic light. Using port-wine stains, which is a frequently occurring skin disorder caused by vascular malformation, as a model disease, the effectiveness of DfMNA mediated PDT has been verified on mice. Compared with the standard operation procedure of clinical PDT, the DfMNA decreases the amount of photosensitizer from 300 μg to 0.5 μg and reduces therapeutic light irradiance from 100 mW cm−2 to 60 mW cm−2 while realizing better treatment effects. As a result, the skin damage and the burden on the metabolic system have been alleviated. The DfMNA has a remarkably reduced photosensitizer amount and, for the first time, realized transdermal delivery of therapeutic light for PDT, thus avoiding the disadvantages of existing PDT methodologies.
- 24 August 2022
- Published by Royal Society of Chemistry (RSC) in Lab on a Chip
- Vol. 22 (23) , 4521-4530
Dual-function microneedle array for efficient photodynamic therapy with transdermal co-delivered light and photosensitizers
Photodynamic therapy (PDT), as a globally accepted method for treating different forms of skin or mucosal disorders, requires efficient co-delivery of photosensitizers and corresponding therapeutic light. The adverse effects of intravenous injection of photosensitizers have been reduced by the development of microneedle arrays for transdermal local photosensitizer delivery. However, the drawbacks of the only available therapeutic light delivery method at the moment, which is directly applying light to the skin surface, are yet to be improved. This study presents a new strategy in which therapeutic light and photosensitizer were transdermally co-delivered into local tissues. A flexible dual-function microneedle array (DfMNA) which contains 400 microneedles was developed. Each microneedle consists of a dissolvable needle tip (140 μm in height) for delivering the photosensitizer and a transparent needle body (660 μm in height) for guiding therapeutic light. Using port-wine stains, which is a frequently occurring skin disorder caused by vascular malformation, as a model disease, the effectiveness of DfMNA mediated PDT has been verified on mice. Compared with the standard operation procedure of clinical PDT, the DfMNA decreases the amount of photosensitizer from 300 μg to 0.5 μg and reduces therapeutic light irradiance from 100 mW cm−2 to 60 mW cm−2 while realizing better treatment effects. As a result, the skin damage and the burden on the metabolic system have been alleviated. The DfMNA has a remarkably reduced photosensitizer amount and, for the first time, realized transdermal delivery of therapeutic light for PDT, thus avoiding the disadvantages of existing PDT methodologies.
- Open Access
- 14 July 2022
- Published by World Scientific Pub Co Pte Ltd in Journal of Innovative Optical Health Sciences
- Vol. 15 (04)
Implementation and uniformity calibration of LED array for photodynamic therapy
Irradiance uniformity is an important parameter for a Photodynamic Therapy (PDT) device. The calibration and verification of a LED array light source based on computer vision technology were implemented and carried out. First, the correlation coefficients between the pulse width modulate value and the irradiance were calibrated. Then, the correction of the actual light center and divergence angle were solved by image processing to reduce errors from each LED lens. Finally, uniformity was optimized according to the irradiance formula of the Lambertian source. The lowest coefficients of variation of irradiance were 4.87% in a [ ] area and 3.55% in a [ ] area within the depth range of 8–12[ ]cm when the expected irradiance was 100[ ]mW/cm2. This finding indicated that the light source can achieve a more uniform illumination and provide a better therapeutic effect for the PDT of port-wine stains.
- 14 July 2022
- Published by World Scientific Pub Co Pte Ltd in Journal of Innovative Optical Health Sciences
- Vol. 15 (04)
Open AccessImplementation and uniformity calibration of LED array for photodynamic therapy
Irradiance uniformity is an important parameter for a Photodynamic Therapy (PDT) device. The calibration and verification of a LED array light source based on computer vision technology were implemented and carried out. First, the correlation coefficients between the pulse width modulate value and the irradiance were calibrated. Then, the correction of the actual light center and divergence angle were solved by image processing to reduce errors from each LED lens. Finally, uniformity was optimized according to the irradiance formula of the Lambertian source. The lowest coefficients of variation of irradiance were 4.87% in a [ ] area and 3.55% in a [ ] area within the depth range of 8–12[ ]cm when the expected irradiance was 100[ ]mW/cm2. This finding indicated that the light source can achieve a more uniform illumination and provide a better therapeutic effect for the PDT of port-wine stains.