HajihajihaJimmy/RBG-Diff-SVCT-data

RBG-Diff Sparse-View CT Data (AAPM + Siemens real)

Preprocessed CT data for reproducing RBG-Diff (Residual-Bootstrapping Generalized Diffusion) sparse-view CT reconstruction, in the exact on-disk layout the RBG-Diff training/evaluation code expects. Two subsets:

• aapm16/ — simulation experiments (AAPM-Mayo abdomen low-dose CT), HU images.
• real_siemens/ — real clinical experiments (Mayo Siemens abdomen + chest), HU images/volumes reconstructed from flat-fan-rebinned projection data.

⚠️ Source & license. This is a processed/derivative version of the TCIA collection Low Dose CT Image and Projection Data (LDCT-and-Projection-data) (DOI 10.7937/9npb-2637), which curates the AAPM Low Dose CT Grand Challenge data. Only the chest and abdomen/liver components are included; these are licensed CC BY 4.0. The head component (NIH controlled access) is not included. Redistribution here is under CC BY 4.0: the data has been modified (see Processing below) and must be used with the attribution in Citation.

Contents & layout

aapm16/                     # simulation (AAPM-Mayo abdomen)
├── train_img/    *.npy      # 5410 slices — one 256×256 HU image per slice (float)
└── test_img/     *.npy      # 526 slices
real_siemens/               # real (Mayo Siemens abdomen + chest)
├── train_img/    *.npy      # 5345 slices — one 256×256 HU image per slice (int16)
├── test_vol/     *.npy      # 12 per-case full-view GT volumes (nz, 256, 256) int16 HU
└── preprocess_manifest.csv  # per-file provenance of the real preprocessing

Sparse-view sinograms are not stored: RBG-Diff simulates them on the fly from each reference image (the reference/GT is the 720-view FBP reconstruction for the simulation path; the full-view real recon for the real path). Metrics are reported under HU window (3000, 500) at 18 / 36 / 72 views.

Usage with RBG-Diff

Download and point the code at the two roots:

hf download HajihajihaJimmy/RBG-Diff-SVCT-data --repo-type dataset --local-dir ./rbgdiff_data

# Simulation (AAPM): reproduce the main table
DATA_ROOT=./rbgdiff_data/aapm16 bash train.sh <GPU>
DATA_ROOT=./rbgdiff_data/aapm16 bash test.sh  <GPU>

# Real (Siemens)
bash train_real.sh <GPU> 20 ./rbgdiff_data/real_siemens/train_img
python eval_real.py --gpu <GPU> --test_vol_dir ./rbgdiff_data/real_siemens/test_vol

The directory names (train_img / test_img / test_vol) match the code's --dataset_path / --test_vol_dir arguments exactly, so no reformatting is needed.

Processing

Derived from the TCIA source as follows:

• aapm16/ — AAPM-Mayo abdomen CT images exported to one HU image per slice, loaded and resized to 256×256.
• real_siemens/ — Siemens helical projection data rebinned to flat-fan with Helix2Fan, reconstructed on a 256² / FOV=330 grid, and written as AAPM-format HU int16 .npy (per-slice training images; per-case full-view test volumes). preprocess_manifest.csv records the provenance of each output.

To rebuild from the original data instead of using this mirror, obtain the raw data from TCIA under its terms and run the preprocessing scripts in the RBG-Diff repository (preprocess_real.py for the real path).

Citation

If you use this data, cite the original dataset and acknowledge the funding (required by CC BY 4.0 / TCIA):

@misc{ldct_projection_2020,
  author       = {McCollough, C. and Chen, B. and Holmes III, D. and Duan, X. and
                  Yu, Z. and Yu, L. and Leng, S. and Fletcher, J.},
  title        = {Low Dose CT Image and Projection Data (LDCT-and-Projection-data)},
  year         = {2020},
  version      = {7},
  publisher    = {The Cancer Imaging Archive},
  doi          = {10.7937/9npb-2637}
}

Data collection was supported by NIBIB grants EB017095 and EB017185.

Please also cite the RBG-Diff paper if you use this data with the RBG-Diff method.

Disclaimer

Research use only; not for clinical or diagnostic use. Redistribution of these files or any further-processed derivatives must keep the CC BY 4.0 license, carry the citation above, and indicate that the data was modified.