Awesome Computed Tomography

CT Reconstruction Research Groups

25 research groups working on CT reconstruction, with their institution, lead, focus and their work in this directory.

Research groups and labs whose papers, software or data appear in this directory. Each card links the group's homepage and the works of theirs already in the list.

North America

• WANG-AXIS Lab / Biomedical Imaging Center (BIC) WANG-AXIS 实验室 / 生物医学成像中心

  Rensselaer Polytechnic Institute · Troy, NY, USA·Led by Ge Wang·Diffusion & Generative Priors · Self-Supervised & Untrained Methods · Low-Dose CT Denoising · Sparse-View CT · Spectral & Dual-Energy CT

  The WANG-AXIS Lab and the Biomedical Imaging Center (BIC) at Rensselaer Polytechnic Institute, directed by Ge Wang, work on x-ray CT imaging theory, systems, and reconstruction. The group's earlier work includes spiral cone-beam CT reconstruction and interior tomography, and its recent work applies deep learning to tomographic reconstruction, including the FORCE framework (2025), which combines a Poisson flow generative prior with data fidelity for reconstruction under low-dose, sparse-view, and metal-artifact conditions. The center also operates CT hardware that includes a photon-counting (MARS) spectral micro-CT scanner and a grating-based x-ray phase-contrast imaging platform.

  In this library: Tomographic Foundation Model—FORCE: Flow-Oriented Reconstruction Conditioning Engine

  Code in the lab's GitHub organization (github.com/WANG-AXIS): FORCE, RDDM (low-dose CT denoising), LEARN (sparse-data CT reconstruction network), M3FM.

• X-ray Tomographic Imaging Group X 射线断层成像课题组

  University of Chicago · Chicago, IL, USA·Led by Xiaochuan Pan, Emil Y. Sidky·Classical & Iterative Reconstruction · Sparse-View CT · Limited-Angle Tomography · Spectral & Dual-Energy CT · Low-Dose CT Denoising

  The X-ray tomographic imaging group in the Department of Radiology at the University of Chicago, led by Xiaochuan Pan and Emil Y. Sidky, develops reconstruction algorithms and imaging systems for clinical, preclinical, security, and synchrotron CT. The group is known for constrained total-variation (TV) minimization reconstruction — the ASD-POCS algorithm of Sidky & Pan (2008) — which minimizes image TV subject to data constraints and reconstructs cone-beam CT images from sparse or limited-angle data. Recent work from the group includes directional TV algorithms, one-step spectral CT reconstruction, and reconstruction from truncated projection data.

  In this library: Image reconstruction in circular cone-beam computed tomography by constrained, total-variation minimization · leehoy/CTReconstruction

  CTReconstruction (GitHub: leehoy/CTReconstruction) — reconstruction code associated with the group.

• I-STAR Lab (Imaging for Surgery, Therapy, and Radiology) I-STAR 实验室

  Johns Hopkins University · Baltimore, MD, USA·Led by Jeffrey H. Siewerdsen, J. Webster Stayman, Wojciech Zbijewski·Cone-Beam CT · Scatter Correction · Geometry Calibration & Motion Compensation · Spectral & Dual-Energy CT · Metal Artifact Reduction

  The I-STAR Lab (Imaging for Surgery, Therapy, and Radiology) at Johns Hopkins University, led by Jeffrey Siewerdsen, J. Webster Stayman, and Wojciech Zbijewski, works on cone-beam CT (CBCT) systems and 3D reconstruction for image-guided interventions. Its research includes penalized-likelihood (model-based) reconstruction, known-component reconstruction for imaging in the presence of implants and other known objects, and physics modeling of effects such as x-ray scatter, detector lag, and system geometry. Early work from lab members quantified the magnitude and effects of x-ray scatter in flat-panel CBCT (Siewerdsen & Jaffray, 2001) and Monte Carlo-based scatter correction in cone-beam micro-CT (Zbijewski & Beekman, 2006).

  In this library: Cone-Beam Computed Tomography with a Flat-Panel Imager: Magnitude and Effects of X-Ray Scatter · Efficient Monte Carlo Based Scatter Artifact Reduction in Cone-Beam Micro-CT

  The lab site (istar.jhu.edu) maintains a Downloads section for lab software and data; no single flagship public toolkit was identified.

• Fessler Group Fessler 课题组

  University of Michigan · Ann Arbor, MI, USA·Led by Jeffrey A. Fessler·Classical & Iterative Reconstruction · Deep Unrolling & Model-Based Learning · Low-Dose CT Denoising

  Jeffrey Fessler's group at the University of Michigan works on statistical (model-based) image reconstruction and the optimization algorithms behind it, for CT as well as PET, SPECT, and MRI. Fessler & Rogers (1996) analyzed the spatial resolution properties of penalized-likelihood reconstruction for space-invariant tomographs, work that underlies later methods for designing regularizers with controlled resolution properties. The group maintains the Michigan Image Reconstruction Toolbox (MIRT) in MATLAB and Julia (MIRT.jl) versions, and the PyTorch-based MIRTorch toolbox for differentiable image reconstruction.

  In this library: Spatial resolution properties of penalized-likelihood image reconstruction: space-invariant tomographs · MIRT · MIRTorch

  MIRT — Michigan Image Reconstruction Toolbox (MATLAB; Julia port MIRT.jl under active development). MIRTorch — PyTorch-based differentiable reconstruction toolbox (PyPI: MIRTorch).

• CT Clinical Innovation Center CT 临床创新中心

  Mayo Clinic · Rochester, MN, USA·Led by Cynthia H. McCollough, Lifeng Yu·Low-Dose CT Denoising · Spectral & Dual-Energy CT · Dynamic & 4D CT

  The CT Clinical Innovation Center at Mayo Clinic, founded in 2004, develops and evaluates new CT technology, clinical applications, and dose-reduction techniques, with core faculty including Cynthia McCollough and Lifeng Yu. Its work covers dual- and multi-energy CT and photon-counting-detector CT, reviewed in McCollough et al. (2015). The center also created public data resources for the CT research community: the 2016 AAPM Low-Dose CT Grand Challenge, and the Low Dose CT Image and Projection Data library (LDCT-and-Projection-data) hosted at The Cancer Imaging Archive, which provides routine-dose and simulated low-dose projection data together with full-dose images for 299 patient exams of the head, chest, and abdomen.

  In this library: Dual- and Multi-Energy CT: Principles, Technical Approaches, and Clinical Applications · LDCT-and-Projection-data · AAPM Low-Dose CT Grand Challenge

  LDCT-and-Projection-data library (TCIA) — 299 patient exams with routine-dose and simulated low-dose projection data. DICOM-CT-PD — open, vendor-neutral data format for CT projection data developed by the center.

• CT Research Group (UW CT Lab) UW CT 实验室

  University of Wisconsin–Madison · Madison, WI, USA·Led by Guang-Hong Chen·Low-Dose CT Denoising · Spectral & Dual-Energy CT · Dynamic & 4D CT · Sparse-View CT

  Guang-Hong Chen's CT research group at the University of Wisconsin–Madison works on reconstruction methods for low-dose and undersampled CT. Chen introduced Prior Image Constrained Compressed Sensing (PICCS), which uses a prior image as a constraint to reconstruct dynamic CT images from highly undersampled projection data, and PICCS has been applied to cardiac imaging, perfusion imaging, dual-energy CT, and C-arm CT. The group also works on spectral and photon-counting detector CT; UW–Madison was the first U.S. clinical evaluation site for GE Healthcare's silicon-based photon-counting CT system, with first human scanning in November 2022.

  In this library: Prior Image Constrained Compressed Sensing (PICCS): A Method to Accurately Reconstruct Dynamic CT Images from Highly Undersampled Projection Data Sets · Accurate and robust sparse-view angle CT image reconstruction using deep learning and prior image constrained compressed sensing (DL-PICCS)

• Bouman Group Bouman 课题组

  Purdue University · West Lafayette, IN, USA·Led by Charles A. Bouman·Sparse-View CT · Low-Dose CT Denoising · Classical & Iterative Reconstruction · Geometry Calibration & Motion Compensation

  Charles Bouman's group at Purdue University works on model-based iterative reconstruction (MBIR) for X-ray CT, formulating reconstruction as statistical estimation with explicit forward models and Markov-random-field priors. The MBIR statistical reconstruction work carried out with Ken Sauer (University of Notre Dame) in the 1990s was later commercialized by GE Healthcare as the Veo MBIR product for low-dose clinical CT. The group maintains SVMBIR, an open-source implementation of MBIR for parallel- and fan-beam CT, and has extended model-based methods to distributed and multi-node reconstruction.

  In this library: A generalized Gaussian image model for edge-preserving MAP estimation · A local update strategy for iterative reconstruction from projections · svmbir

  SVMBIR (GitHub: cabouman/svmbir) — supervised MBIR for parallel/fan-beam CT.

• LEAP Project, Lawrence Livermore National Laboratory 劳伦斯利弗莫尔国家实验室 LEAP 项目

  Lawrence Livermore National Laboratory · Livermore, CA, USA·Led by Kyle Champley, Hyojin Kim·Geometry Calibration & Motion Compensation · Spectral & Dual-Energy CT · Cone-Beam CT · Other Topics

  The LEAP project at Lawrence Livermore National Laboratory provides a library of 3D transmission CT algorithms — pre-processing, forward/back projectors, and analytic (FBP/FDK) and iterative reconstruction — with Python and C++ APIs and PyTorch bindings that make the projectors differentiable for integration into deep-learning pipelines. The projectors support multiple geometries on multi-GPU and multi-core CPU hardware and are aimed at industrial and laboratory CT. Physics-based modeling and correction (scatter, beam hardening, dual-energy decomposition) is provided through the companion XrayPhysics package, a C/C++ library with Python bindings for X-ray cross sections and source-spectra modeling.

  In this library: LEAP · XrayPhysics

  LEAP — LivermorE AI Projector for Computed Tomography (GitHub: LLNL/LEAP). XrayPhysics (GitHub: kylechampley/XrayPhysics) — cross sections, spectra modeling, multi-material beam-hardening correction, dual-energy decomposition.

• APS Tomography Group 先进光子源断层成像团队

  Argonne National Laboratory · Lemont, IL, USA·Led by Francesco De Carlo, Viktor Nikitin, Doga Gürsoy·Classical & Iterative Reconstruction · Dynamic & 4D CT · Cone-Beam CT · Other Topics

  The tomography group at the Advanced Photon Source of Argonne National Laboratory develops the reconstruction and data-handling software used at synchrotron micro- and nano-tomography beamlines. Its software output includes TomoPy, a Python framework for pre-processing and reconstructing synchrotron tomographic data (Gürsoy, De Carlo et al., 2014), and TomocuPy, Viktor Nikitin's GPU pipeline that reconstructs data in concurrent chunks. The group also curates TomoBank, a repository of experimental and simulated tomography datasets with descriptors, provided so that reconstruction methods can be developed and compared on shared beamline data.

  In this library: TomoPy · TomoBank

  TomoPy (GitHub: tomopy/tomopy) — tomographic reconstruction in Python. TomocuPy — GPU-accelerated reconstruction pipeline (Nikitin, Journal of Synchrotron Radiation, 2023). TomoBank — open repository of synchrotron tomography datasets.

• Lei Xing's group, Medical Physics Division 斯坦福大学 Lei Xing 课题组（医学物理部）

  Stanford University · Stanford, CA, USA·Led by Lei Xing·Sparse-View CT · Implicit Neural Representations · Metal Artifact Reduction · Cone-Beam CT · Low-Dose CT Denoising

  Lei Xing directs the Medical Physics Division of Stanford's Department of Radiation Oncology, and his group works on artificial intelligence and imaging for radiation therapy. Its CT reconstruction papers include patient-specific volumetric reconstruction from a single projection view (Nature Biomedical Engineering, 2019), implicit neural representations with prior embedding for sparse sampling (NeRP, PINER), compressed-sensing and 4D cone-beam CT, the TransCT low-dose network and deep-learning metal artifact reduction.

  In this library: Patient-specific reconstruction of volumetric computed tomography images from a single projection view via deep learning · NeRP: Implicit Neural Representation Learning With Prior Embedding for Sparsely Sampled Image Reconstruction · PINER: Prior-Informed Implicit Neural Representation Learning for Test-Time Adaptation in Sparse-View CT Reconstruction · TransCT: Dual-Path Transformer for Low Dose Computed Tomography · Deep Sinogram Completion With Image Prior for Metal Artifact Reduction in CT Images · Metal artifact reduction in 2D CT images with self-supervised cross-domain learning · Compressed sensing based cone-beam computed tomography reconstruction with a first-order method · 4D Cone Beam CT via Spatiotemporal Tensor Framelet

• Imaging and Informatics Lab (CT Lab) 麻省大学洛厄尔分校成像与信息学实验室（Hengyong Yu）

  University of Massachusetts Lowell · Lowell, MA, USA·Led by Hengyong Yu·Metal Artifact Reduction · Low-Dose CT Denoising · Spectral & Dual-Energy CT · Dynamic & 4D CT · Limited-Angle Tomography

  The Imaging and Informatics Lab in the Department of Electrical and Computer Engineering at UMass Lowell is directed by Hengyong Yu and works on computed tomography and medical image processing, from compressed sensing based interior tomography to machine-learning based tomographic imaging. Its papers in this directory include CNN-based metal artifact reduction (CNN-MAR, IEEE TMI 2018), the CTformer transformer for low-dose CT, adversarial reconstruction (CLEAR), tensor methods for dynamic and spectral CT, and limited-angle cardiac CT.

  In this library: Convolutional Neural Network Based Metal Artifact Reduction in X-Ray Computed Tomography · CTformer: Convolution-Free Token2Token Dilated Vision Transformer for Low-Dose CT Denoising · CLEAR: Comprehensive Learning Enabled Adversarial Reconstruction for Subtle Structure Enhanced Low-Dose CT Imaging · Tensor-based dictionary learning for dynamic tomographic reconstruction · Non-Local Low-Rank Cube-Based Tensor Factorization for Spectral CT Reconstruction · Hybrid U-Net and Swin-transformer network for limited-angle cardiac computed tomography

Europe

• Pattern Recognition Lab (LME) 模式识别实验室

  Friedrich-Alexander-Universität Erlangen-Nürnberg · Erlangen, Germany·Led by Andreas Maier·Geometry Calibration & Motion Compensation · Cone-Beam CT · Deep Unrolling & Model-Based Learning · Diffusion & Generative Priors

  The Pattern Recognition Lab at FAU Erlangen-Nürnberg, headed by Andreas Maier since 2015, applies machine learning to medical imaging, with a CT line built around known operators — embedding the physics of projection and reconstruction as differentiable layers inside neural networks. Its software output includes PYRO-NN, a library of differentiable CT operators for parallel, fan, and cone-beam geometries, and diffct, a CUDA-accelerated differentiable CT library whose later versions support non-circular and arbitrary acquisition trajectories. A DFG-funded project, "Deep-Learning-basierte CBCT-Rekonstruktion beliebiger Trajektorien" (36 months, 324,459 EUR, announced September 2025), develops deep-learning-based CBCT reconstruction for arbitrary trajectories with partners at TH Deggendorf and Stanford University, using FAU's RRZE computing infrastructure.

  In this library: diffct · PYRO-NN

  diffct (GitHub: sypsyp97/diffct) — CUDA-accelerated differentiable CT library. PYRO-NN (GitHub: csyben/PYRO-NN) — differentiable CT reconstruction operators with PyTorch support and native CUDA kernels.

• Computational Image Analysis (CIA) Group 剑桥大学计算图像分析组

  University of Cambridge · Cambridge, United Kingdom·Led by Carola-Bibiane Schönlieb, Ander Biguri·Deep Unrolling & Model-Based Learning · Sparse-View CT · Low-Dose CT Denoising · Self-Supervised & Untrained Methods · Cone-Beam CT

  The Computational Image Analysis group at Cambridge works on mathematical imaging and inverse problems, with a dedicated CT remit on closing the gap between mathematical advances and clinical practice. Its tomographic work covers data-driven reconstruction methods — including iterative unrolled methods, learned regularizers, and Plug-and-Play approaches — alongside variational regularization, Krylov subspace solvers, and 4D imaging. The group develops LION (Learned Iterative Optimization Networks), an open software package that collects post-processing, unrolled, learned-regularizer, Plug-and-Play and self-supervised CT reconstruction methods with reproducible training pipelines and CT datasets.

  In this library: LION

  LION (Learned Iterative Optimization Networks) — data-driven CT reconstruction methods with reproducible training pipelines; the group also co-maintains the TIGRE toolbox for cone-beam reconstruction.

• Computational Imaging Group 阿姆斯特丹 CWI 计算成像组

  Centrum Wiskunde & Informatica (CWI) · Amsterdam, Netherlands·Led by Tristan van Leeuwen·Classical & Iterative Reconstruction · Sparse-View CT · Spectral & Dual-Energy CT · Geometry Calibration & Motion Compensation · Dynamic & 4D CT

  The Computational Imaging group at CWI develops mathematics- and algorithm-driven methods for 3D imaging across science, medicine and industry, combining advanced acquisition, parameter estimation and discrete tomography — including real-time reconstruction via RECAST3D and the in-house FleX-ray lab that couples a custom CT scanner to advanced processing and AI algorithms. It jointly developed the ASTRA Toolbox with the University of Antwerp — a MATLAB/Python platform of high-performance GPU primitives for 2D and 3D tomography used by researchers, industrial partners, and large synchrotrons (ESRF, APS) — and tomosipo, its Pythonic wrapper for 3D tomography with complex geometries, aimed at pairing fast GPU compute with modern deep learning tools.

  In this library: ASTRA Toolbox · tomosipo · FleX-ray Walnut CBCT Collection (CWI)

  ASTRA Toolbox (astra-toolbox.com) and tomosipo; RECAST3D for real-time visualization of tomographic imaging.

• imec-Vision Lab (ASTRA Group) 安特卫普大学 imec 视觉实验室（ASTRA 组）

  University of Antwerp · Antwerp, Belgium·Led by Jan Sijbers, Jan De Beenhouwer·Classical & Iterative Reconstruction · Geometry Calibration & Motion Compensation · Dynamic & 4D CT · Spectral & Dual-Energy CT · Other Topics

  The ASTRA research group of imec-Vision Lab develops tomographic reconstruction algorithms for electron, X-ray, neutron and magnetic resonance imaging, focusing on the general reconstruction problems shared across application fields rather than any single application. Key expertises include phase-contrast X-ray CT, X-ray tomography for non-standard acquisition geometries (e.g., laminography, tomosynthesis), dynamic (4D) tomography and discrete tomography. The group developed the ASTRA Toolbox — an open-source, GPU-accelerated software platform for developing novel tomographic reconstruction methods — jointly with CWI's Computational Imaging group.

  In this library: ASTRA Toolbox

  ASTRA Toolbox (developed jointly with CWI); DART-related discrete tomography methods.

• Mathematical Imaging Group (Ozan Öktem) 瑞典皇家理工学院 Öktem 课题组

  KTH Royal Institute of Technology · Stockholm, Sweden·Led by Ozan Öktem·Deep Unrolling & Model-Based Learning · Classical & Iterative Reconstruction · Other Topics · Low-Dose CT Denoising · Sparse-View CT

  Ozan Öktem's research at KTH combines inverse problems, convex optimization and machine learning for medical imaging, with a line of work on learned iterative (unrolled) reconstruction schemes. Together with Jonas Adler he introduced the Learned Primal-Dual reconstruction method (IEEE T-MI 2018) and the ODL (Operator Discretization Library) Python library for inverse problems, which provides forward operators, geometries and analytical methods used by many subsequent CT reconstruction studies. The learned gradient tomography codebase implements the earlier iterative deep network work (Adler & Öktem, "Solving ill-posed inverse problems using iterative deep neural networks", Inverse Problems 2017).

  In this library: ODL · Learned Primal-Dual · Learned Gradient Tomography

  ODL (odlgroup/odl) — operator discretization library for inverse problems; learned primal-dual and learned gradient tomography implementations.

• CREATIS Tomographic Reconstruction Group 里昂 CREATIS 层析重建组

  CREATIS (CNRS / Inserm / INSA Lyon) · Lyon, France·Led by Simon Rit·Cone-Beam CT · Dynamic & 4D CT · Classical & Iterative Reconstruction · Sparse-View CT · Other Topics

  Simon Rit's group at CREATIS works on tomographic reconstruction with a focus on cone-beam CT for image-guided radiotherapy, including regularized and sparse (compressive-sensing) CT methods and motion-compensated X-ray CT for cardiac and respiratory motion — methods that were transferred to a clinical solution in the Netherlands. The group develops the Reconstruction Toolkit (RTK, openrtk.org), an open-source cone-beam CT reconstruction toolkit built on ITK with GPU code from Plastimatch, released under Apache 2.0 and distributed both as C++ classes/command-line tools and as Python packages (itkrtk) via PyPI. RTK supports Elekta, Varian and IBA scanner inputs, CPU/GPU Feldkamp-David-Kress reconstruction, 4D algorithms such as 4D ROOSTER, and cylindrical-detector geometries used by diagnostic CT scanners.

  In this library: RTK

  RTK (Reconstruction Toolkit; GitHub: RTKConsortium/RTK, PyPI: itkrtk) — open-source cone-beam CT reconstruction toolkit built on ITK.

• Tomographic Imaging / Core Imaging Library (CIL) Group

  Science and Technology Facilities Council (STFC) · Harwell / Manchester, United Kingdom·Led by Edoardo Pasca, Daniil Kazantsev, Jakob Sauer Jørgensen·Classical & Iterative Reconstruction · Dynamic & 4D CT · Spectral & Dual-Energy CT · Other Topics

  The Tomographic Imaging group at the UK Science and Technology Facilities Council (STFC), working with partners including the University of Manchester through the Collaborative Computational Project in Tomographic Imaging (CCPi), develops open-source software for tomographic image reconstruction. Its Core Imaging Library (CIL) is a Python framework for iterative, regularised reconstruction of datasets where filtered back-projection is insufficient, such as noisy, incomplete, non-standard-geometry or multichannel (dynamic and spectral) data. The group also works on synergistic imaging, combining tomography with other modalities and data sources in joint reconstruction problems.

  In this library: Core Imaging Library (CIL) · CCPi Regularisation Toolkit · TomoPhantom

  Core Imaging Library (CIL) — open-source Python reconstruction framework (GitHub: TomographicImaging/CIL), Apache-2.0. CCPi Regularisation Toolkit. TomoPhantom — 2D–4D analytical phantoms for reconstruction benchmarking.

Asia

• BISPL (Bio Imaging & Signal Processing Lab)

  Korea Advanced Institute of Science and Technology (KAIST) · Daejeon, South Korea·Led by Jong Chul Ye·Diffusion & Generative Priors · Sparse-View CT · Limited-Angle Tomography · Self-Supervised & Untrained Methods

  The Bio Imaging & Signal Processing Lab (BISPL) at KAIST, led by Jong Chul Ye, works on signal processing and machine learning methods for biomedical imaging. A central research line uses diffusion and score-based generative models as priors for imaging inverse problems, including sparse-view and limited-angle CT; its DiffusionMBIR method combines a pre-trained 2D diffusion model with model-based iterative reconstruction and a total-variation prior along the slice direction to solve 3D reconstruction problems such as sparse-view and limited-angle tomography.

  In this library: Solving 3D Inverse Problems Using Pre-Trained 2D Diffusion Models

• CT Imaging Group, Department of Engineering Physics 清华大学工程物理系 CT 成像组

  Tsinghua University · Beijing, China·Led by Yuxiang Xing (邢宇翔), Li Zhang (张丽)·Spectral & Dual-Energy CT · Classical & Iterative Reconstruction · Geometry Calibration & Motion Compensation · Metal Artifact Reduction

  The CT imaging group in Tsinghua University's Department of Engineering Physics, led by Yuxiang Xing and Li Zhang, studies X-ray CT imaging systems and image reconstruction methods. Its research covers tomographic reconstruction algorithms, photon-counting and spectral (multi-energy) CT, and fixed multi-source CT system designs in which source arrays replace gantry rotation. The group is based in the department's Key Laboratory of Particle & Radiation Imaging (Ministry of Education).

• CT Reconstruction Groups (Chen / Luo) 东南大学 CT 重建组（陈阳 / 罗守华）

  Southeast University · Nanjing, China·Led by Yang Chen (陈阳), Shouhua Luo (罗守华)·Low-Dose CT Denoising · Sparse-View CT · Limited-Angle Tomography · Other Topics

  This entry covers two Southeast University groups working on CT reconstruction in Nanjing: Yang Chen's group at the School of Computer Science and Engineering (Laboratory of Image Science and Technology) and Shouhua Luo's group at the School of Biological Science and Medical Engineering. Their research addresses low-dose CT reconstruction and image enhancement under non-ideal measurement conditions such as sparse views and limited angles. In 2026 the CSE-led team published TAMP, an imaging foundation model pre-trained on physics-simulated data for universal enhancement of non-ideal measurement CT, in Nature Communications.

  In this library: Imaging foundation model for universal enhancement of non-ideal measurement CT

  TAMP model code (announced in the paper for release at github.com/YutingHelist/TAMP); SimNICT — the physics-simulated pre-training dataset built for TAMP (announced via Hugging Face, YutingHelist/SimNICT).

• XMed Lab XMed 实验室

  Hong Kong University of Science and Technology · Hong Kong, China·Led by Xiaomeng Li (李小萌)·Sparse-View CT · Cone-Beam CT · Other Topics

  The XMed Lab at the Hong Kong University of Science and Technology, led by Xiaomeng Li in the Department of Electronic and Computer Engineering, works on medical AI and computer vision. Its DeepSparse model (IEEE Transactions on Medical Imaging, 2026) is a foundation model for sparse-view cone-beam CT reconstruction: it is pre-trained on a large-scale abdominal CT dataset (AtlasAbdomenMini, ~7k volumes) and can be fine-tuned to diverse anatomical targets with few labelled examples.

  In this library: DeepSparse: A Foundation Model for Sparse-View CBCT Reconstruction

  DeepSparse code (GitHub: xmed-lab/DeepSparse); pretrained and fine-tuned model weights released on Hugging Face.

• MIAH Lab 复旦大学 MIAH 实验室（单洪明）

  Fudan University · Shanghai, China·Led by Hongming Shan (单洪明)·Low-Dose CT Denoising · Diffusion & Generative Priors · Sparse-View CT · Metal Artifact Reduction

  The MIAH Lab at Fudan University's Institute of Science and Technology for Brain-inspired Intelligence is led by Hongming Shan and works on machine learning for medical imaging, with an emphasis on generative AI and image reconstruction. Its CT work centres on low-dose CT: Shan's earlier MAP-NN (Nature Machine Intelligence, 2019) and 3D CPCE networks, built with Ge Wang's group, were followed at Fudan by diffusion-based denoising such as CoreDiff (IEEE TMI, 2024) and FoundDiff. The group also works on sparse- and incomplete-view reconstruction (FreeSeed, GloReDi, ProCT) and metal artifact reduction (Quad-Net).

  In this library: CoreDiff: Contextual Error-Modulated Generalized Diffusion Model for Low-Dose CT Denoising and Generalization · FoundDiff: Foundational Diffusion Model for Generalizable Low-Dose CT Denoising · FreeSeed: Frequency-Band-Aware and Self-Guided Network for Sparse-View CT Reconstruction · Learning to Distill Global Representation for Sparse-View CT · Universal Pre-training for Generalizable Incomplete-View CT Reconstruction · Quad-Net: Quad-Domain Network for CT Metal Artifact Reduction · ASCON: Anatomy-Aware Supervised Contrastive Learning Framework for Low-Dose CT Denoising · DU-GAN: Generative Adversarial Networks with Dual-Domain U-Net Based Discriminators for Low-Dose CT Denoising · Competitive performance of a modularized deep neural network compared to commercial algorithms for low-dose CT image reconstruction · 3D Convolutional Encoder-Decoder Network for Low-Dose CT via Transfer Learning from a 2D Trained Network

• Deep Imaging Group (DIG) 四川大学深度成像课题组（张意）

  Sichuan University · Chengdu, China·Led by Yi Zhang (张意)·Low-Dose CT Denoising · Metal Artifact Reduction · Sparse-View CT · Spectral & Dual-Energy CT

  The Deep Imaging Group (DIG) is led by Yi Zhang, a professor in the School of Cyber Science and Engineering at Sichuan University, and works on medical image processing with deep learning. Its CT reconstruction papers cover low-dose CT (MAGIC, IEEE TMI 2021), reconstruction across scan geometries (PDF), metal artifact reduction (DAN-Net, IDOL-Net, SemiMAR), sparse-view and spectral CT with unrolled networks (RegFormer, SOUL-Net), and a 2024 review of deep-learning CT reconstruction from incomplete projection data.

  In this library: MAGIC: Manifold and graph integrative convolutional network for low-dose CT reconstruction · CT Reconstruction With PDF: Parameter-Dependent Framework for Data From Multiple Geometries and Dose Levels · Dual-Domain Adaptive-Scaling Non-Local Network for CT Metal Artifact Reduction · IDOL-Net: An Interactive Dual-Domain Parallel Network for CT Metal Artifact Reduction · SemiMAR: Semi-Supervised Learning for CT Metal Artifact Reduction · RegFormer: A Local–Nonlocal Regularization-Based Model for Sparse-View CT Reconstruction · SOUL-Net: A Sparse and Low-Rank Unrolling Network for Spectral CT Image Reconstruction · A Review of Deep Learning CT Reconstruction From Incomplete Projection Data

• Imaging and Visual Representation Laboratory 南昌大学成像与视觉表示实验室（刘且根）

  Nanchang University · Nanchang, China·Led by Qiegen Liu (刘且根)·Diffusion & Generative Priors · Sparse-View CT · Low-Dose CT Denoising

  Qiegen Liu's Imaging and Visual Representation Laboratory in the School of Information Engineering at Nanchang University works on generative models for image reconstruction in MRI and CT. Its CT papers use learned and score-based priors: deep gradient priors for low-dose CT (EASEL, IEEE TRPMS 2022), a denoising-autoencoder prior for sparse-view CT (REDAEP), generative modelling in the sinogram domain (IEEE TRPMS 2024) and a wavelet-refined diffusion model for sparse-view CT (SWORD, IEEE TMI 2024).

  In this library: Iterative Reconstruction for Low-Dose CT using Deep Gradient Priors of Generative Model · REDAEP: Robust and Enhanced Denoising Autoencoding Prior for Sparse-View CT Reconstruction · Generative Modeling in Sinogram Domain for Sparse-view CT Reconstruction · Stage-by-Stage Wavelet Optimization Refinement Diffusion Model for Sparse-View CT Reconstruction · Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary Tomography