Low-Dose CT Denoising

Low-dose CT denoising suppresses the noise and artifacts that appear when CT is acquired at reduced tube current or voltage to lower the radiation dose. Fewer detected photons raise the quantum noise in the projections, which after reconstruction shows up as grainy texture and streaks that can hide low-contrast lesions.

It is used in lung-cancer screening, paediatric imaging, repeated follow-up scans and perfusion studies: anywhere cumulative dose matters. The standard real-data benchmark is the Mayo Clinic data from the 2016 AAPM Low-Dose CT Grand Challenge (now LDCT-and-Projection-data on TCIA), and LoDoPaB-CT is the main simulated benchmark.

Methods work in the projection domain (sinogram filtering and noise-model-aware restoration), in the image domain (CNNs such as RED-CNN, GAN-based denoisers with perceptual losses, transformers), in both domains, or inside statistical iterative reconstruction with learned or dictionary priors. Self-supervised methods avoid the need for paired full-dose images, and diffusion models are increasingly used as priors.

Browse all 69 Low-Dose CT Denoising papers in the searchable directory

Start here

If you are new to this topic, these are good first reads, chosen by the maintainers.

• Low-Dose CT with a Residual Encoder-Decoder Convolutional Neural Network (TMI, 2017 · code)
  The residual encoder-decoder CNN that became the standard image-domain denoising baseline, with code.
• Low Dose CT Image Denoising Using a Generative Adversarial Network with Wasserstein Distance and Perceptual Loss (TMI, 2018 · code)
  Introduced Wasserstein GAN training with a perceptual loss to keep low-dose images from looking over-smoothed, with code.
• CTformer: Convolution-Free Token2Token Dilated Vision Transformer for Low-Dose CT Denoising (PMB, 2023 · code)
  A convolution-free vision transformer for low-dose CT denoising, with code; a good representative of transformer-based methods.
• LoDoPaB-CT, a benchmark dataset for low-dose computed tomography reconstruction (Scientific Data, 2021 · code)
  The descriptor of the LoDoPaB-CT benchmark dataset, with reference code; read it before reporting results on it.
• Self-Supervised Training For Low-Dose CT Reconstruction (ISBI, 2021)
  Trains a low-dose reconstruction network from the noisy measurements alone, without full-dose targets.

Papers

The 30 most recent of 69 papers, newest first.

Browse all 69 Low-Dose CT Denoising papers in the searchable directory

Datasets, benchmarks and toolkits

Entries from the directory whose tasks or tags match this topic.

• LoDoPaB-CT (official page) · Dataset — Projection + image. 42,895 simulated human thorax slice pairs (LIDC/IDRI) at a single noise level (N0 = 4096) — other dose levels have to be re-simulated.
• LDCT-and-Projection-data (official page) · Dataset — Projection + image. Mayo Clinic, released through TCIA.
• LIDC-IDRI (official page) · Dataset — 1,018 thoracic CT scans with four-radiologist nodule annotations. The source anatomy behind LoDoPaB-CT and most simulated low-dose data.
• STOIC 2021 (official page) · Dataset — 10,735 COVID-19 chest CT scans from the French first wave, collected for severe-outcome prediction; a large real-clinical distribution-shift test.
• MosMedData (official page) · Dataset — 1,110 chest CT scans with COVID-19 severity grades; frequently reused as a low-quality / low-dose robustness set. The original site is currently down — the Kaggle mirror carries the same collection.
• LoDoPaB-CT (benchmark release paper) · Dataset — The Scientific Data descriptor for the LoDoPaB-CT dataset — the citation to use when you report results on this split. Documents the simulation protocol and the FBP/U-Net reference baseline.
• COBRA2026: a large-scale multicenter pelvic cone-beam computed tomography projection dataset (official page) · Dataset — Projection + image. 867 pelvic radiotherapy CBCT exams from six European centres (Elekta and Varian systems), with raw projections, acquisition geometry, clinical CBCT and planning CT.
• Walnut-21 (two dose levels) (official page) · Dataset — Projection only. High-resolution cone-beam scans of 21 walnuts, each acquired at two radiation dose levels - a good and a noisy version of the same object.
• LoDoInd (official page) · Dataset — Five dose levels of the same industrial CT specimens, including a near-noise-free reference scan, all pre-registered so slices line up across levels.
• AAPM Low-Dose CT Grand Challenge (official page) · Benchmark — The challenge that established the quarter-dose / 10% / 5% protocol and the Mayo abdominal cohort. Nearly every low-dose CT paper still reports on this split.
• LoDoPaB-CT benchmark (official page) · Benchmark — Grand-challenge style evaluation on the LoDoPaB-CT split, with a published, tuned FBP+U-Net reference baseline so improvements are measurable.
• SPIE-AAPM-NCI LungX Challenge (official page) · Benchmark — Lung 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.
• Noise2Noise denoising protocol (official page) · Benchmark — The reference implementation that popularised training a denoiser without clean targets.
• Physics-ArX · Toolkit — Physics-based data-augmentation library that injects realistic CT/CBCT noise, scatter and artefact models into existing images to build paired training data.
• LD-CT simulator · Toolkit — Low-dose CT simulator that converts high-dose images into realistic low-dose counterparts using a validated noise model.

Guides

Guides that cover this topic.

• Low-Dose CT Denoising: Benchmarks, Datasets and Toolkits
• How to Choose a CT Reconstruction Dataset

Other topics

• Surveys & Reviews (24)
• Classical & Iterative Reconstruction (71)
• Sparse-View CT (157)
• Limited-Angle Tomography (46)
• Metal Artifact Reduction (23)
• Spectral & Dual-Energy CT (19)
• Cone-Beam CT (77)
• Dynamic & 4D CT (24)
• Scatter Correction (11)
• Geometry Calibration & Motion Compensation (28)
• Deep Unrolling & Model-Based Learning (68)
• Self-Supervised & Untrained Methods (35)
• Diffusion & Generative Priors (68)
• Implicit Neural Representations (25)
• Gaussian Splatting (26)