Awesome Computed Tomography 中文 Star on GitHub

Awesome Computed Tomography

A curated list of CT reconstruction resources. Every paper link is resolved against Crossref, arXiv or the publisher’s own record, and every URL is rechecked weekly.

Papers
678
Toolkits
60
Datasets
82
Benchmarks
22
Learning
35
Updated
2026-10-30

CT Reconstruction Benchmarks & Challenges

22 CT reconstruction benchmarks and challenges (low-dose, sparse-view, limited-angle, metal artifacts and more), with year, tasks and what each one evaluates.

Benchmark / ChallengeYearTasksNotes
Generating synthetic computed tomography for radiotherapy: SynthRAD2025 challenge report [official]2026cbct, synthetic-ct, reconstructionSynthetic CT generation from CBCT and MRI — the 2025 edition of SynthRAD, with the accompanying report documenting the tasks, data and results. Same CBCT-to-CT focus as the 2023 edition, where scatter, truncation and beam hardening dominate the input.
CT-DegradBench: A Physics-Informed Benchmark for CT Degradation Detection and Severity Estimation2026reconstructionA physics-informed benchmark for detecting CT degradations (noise, blur, artefacts) and estimating their severity — it asks what is wrong with a scan, not only how to reconstruct it. Accepted at the CVPR 2026 VISION Workshop (DEXTER track).
DeepInverse benchmark suite [official]2024reconstruction, benchmarkingA unified PyTorch benchmark harness covering CT, MRI, deblurring and inpainting, with reproducible metrics and pretrained baselines — the most practical way to compare a new CT method against strong generic solvers.
AAPM CT-MAR Grand Challenge [official]2024metal-artifact, reconstructionMetal artifact reduction on simulated CT (2023–24): 14,000 training cases with and without metal, image- or sinogram-domain methods allowed, ranked by a weighted set of clinically focused image-quality metrics. Results reported in Medical Physics (2025).
SynthRAD2023 Challenge [official]2023cbct, synthetic-ct, reconstructionSynthetic CT generation from CBCT and MRI. Relevant here because the CBCT-to-CT task is exactly where reconstruction artefacts (scatter, truncation, beam hardening) dominate — the challenge forces methods to handle them rather than assume clean input.
LDCTIQAC 2023 [official]2023low-dose, quality-assessmentLow-dose CT perceptual image quality assessment: no-reference quality models scored on low-dose CT images with mixed noise and artefacts, ranked by how well they agree with radiologists' ratings. A test of the metrics reconstruction papers report, not of the reconstructions.
Helsinki Tomography Challenge [official]2022limited-angle, reconstructionReal measured limited-angle sinograms. A purely learned network trained on synthetic data won; the only model-based entry (CIL, directional TV) placed third — evidence that classical methods stay competitive at limited angle.
AutoPET Challenge2022segmentationWhole-body PET/CT lesion segmentation, run annually since 2022 (four editions by 2025), each with a deliberately hidden multi-centre test cohort.
CTooth / CTooth+ [official]2022segmentation, cbctFully annotated 3D dental CBCT dataset and benchmark for tooth segmentation (5,803 CBCT slices), later extended to CTooth+. A small but well-annotated CBCT benchmark in a domain where artefacts are severe.
AAPM DL-Spectral CT Challenge [official]2022spectral, reconstructionReconstruct adipose, fibroglandular and calcification maps directly from simulated dual-kVp breast CT data: 1,000 training cases and a 100-case test set. The spectral counterpart of the DL-Sparse-View challenge.
AAPM DL-Sparse-View CT Challenge [official]2021sparse-view, reconstructionRecover a 2D breast-CT phantom from ideal noiseless 128-view fan-beam data: 4,000 training pairs, 100 test cases, ranked by RMSE on a public leaderboard. A common reference point for deep-learning sparse-view CT.
LoDoPaB-CT benchmark [official]2021low-dose, reconstructionGrand-challenge style evaluation on the LoDoPaB-CT split, with a published, tuned FBP+U-Net reference baseline so improvements are measurable.
MICCAI FLARE2021segmentationMulti-organ abdominal CT segmentation across four editions (FLARE21/22/23/24); the standard downstream-robustness target for CT reconstruction outputs.
HECKTOR Challenge2020segmentation, prognosisHead-and-neck PET/CT tumour delineation; five editions make it one of the longer-running medical imaging challenges.
KiTS Challenge2019segmentationKidney tumour segmentation, three editions (2019/2021/2023) with strictly controlled annotation quality and a strong emphasis on cross-institution generalisation.
VerSe Challenge2019segmentationVertebra labelling and segmentation on multi-detector CT, run in two editions with MICCAI 2019 and 2020 (374 scans, 355 patients); the hidden test set is what makes results comparable.
CHAOS Challenge2019segmentationAbdominal CT and MRI organ segmentation with explicit cross-modality evaluation; the CT scans are portal-venous-phase contrast-enhanced, not low-dose.
SPARE Challenge [official]2018dynamic, cbct, reconstructionThoracic 4D cone-beam CT reconstruction from a one-minute scan. The first widely adopted benchmark for breathing-resolved dynamic CBCT; run by the Image X Institute at the University of Sydney.
Medical Segmentation Decathlon2018segmentationTen tasks with a fixed, hidden test set and a strict rule that no task-specific tuning is allowed — a genuinely fair generalisation benchmark.
Noise2Noise denoising protocol [official]2018denoising, self-supervisedThe reference implementation that popularised training a denoiser without clean targets. On CT it became the standard 'no clean ground truth' evaluation protocol — most unsupervised CT denoising papers report on a Mayo-LDCT-style split under this setting.
AAPM Low-Dose CT Grand Challenge [official]2016low-dose, denoising, reconstructionThe 2016 AAPM–NIBIB–Mayo Clinic challenge: 30 contrast-enhanced abdominal CT cases (10 for training, 20 for testing), with quarter-dose (25 %) data simulated by inserting Poisson noise into the projections. Its cases and protocol are the basis of the Mayo low-dose data that most low-dose CT papers still report on.
SPIE-AAPM-NCI LungX Challenge [official]2015diagnosis, low-doseLung nodule malignancy classification on low-dose screening CT — the task is diagnostic and benchmarked against radiologists, not nodule detection. 70 cases (10 training, 60 test), distributed through TCIA.