SimNICT: Simulated Non-Ideal measurement CT Dataset

SimNICT is the first comprehensive dataset for training universal non-ideal measurement CT (NICT) enhancement models, containing simulated low-dose, limited-angle, and sparse-view CT from different body regions. We release the SimNICT Dataset (823 GB, 8 datasets) for comprehensive NICT research, and provide SimNICT-AMOS-Sample (78 MB) for quick exploration and prototyping.

💡 Recommendation: Start with SimNICT-AMOS-Sample Dataset for initial exploration and prototyping, then download specific datasets from the SimNICT Dataset based on your research needs.

---

Part 1: SimNICT Dataset

Dataset	Volumes	Body Regions	License	Download Link
AMOS	500	Abdomen	CC BY 4.0	simnict-amos
COVID-19-NY-SBU	459	Chest	CC BY 4.0	simnict-covid-19-ny-sbu
CT Images in COVID-19	771	Chest	CC BY 4.0	simnict-ct-images-in-covid-19
CT_COLONOGRAPHY	1,730	Abdomen	CC BY 4.0	simnict-ct-colonography
LNDb	294	Chest	CC BY-NC-ND 4.0	simnict-lndb
LUNA	888	Chest	CC BY 4.0	simnict-luna
MELA	1,100	Chest	CC BY 4.0	simnict-mela
STOIC	2,000	Chest	CC BY-NC 4.0	simnict-stoic
AutoPET	1,014	Whole-body	NIH Controlled Data Access Policy	-
HECKTOR22	882	Head, neck	Custom Research License	-

Note: AutoPET and HECKTOR22 datasets are not publicly available due to licensing restrictions.

Dataset Overview

The SimNICT dataset is a large-scale medical imaging dataset containing:

• 📊 9,513 CT volumes from 10 medical imaging datasets (2 out of 10 datasets are not open-source due to licensing restrictions)
• 🔬 3 NICT types: Low-dose CT (LDCT), Sparse-view CT (SVCT), Limited-angle CT (LACT)
• ⚙️ Randomized parameters:
  • SVCT: Views randomly sampled from 15-360 range
  • LACT: Angular range randomly sampled from 75°-270°
  • LDCT: Dose levels randomly sampled from 5%-75% range
• 💾 Total size: ~823 GB
• 📁 File format: NIfTI (.nii.gz), 16-bit, gzip compressed

Data Release Strategy

The SimNICT dataset provides preprocessed ICT data with NICT simulation code:

1. 🔍 Preprocessed ICT Data

The preprocessed ICT data of SimNICT dataset is hosted on Internet Archive, ensuring stable access for the global research community. You can download data through the dataset table above or execute batch download scripts below:

# Download batch download script
wget https://huggingface.co/datasets/YutingHe-list/SimNICT/blob/main/simnict_download.py

# Download all datasets (~823 GB)
python simnict_download.py --all --output_dir ./data

# Download specific datasets
python simnict_download.py --datasets AMOS LUNA --output_dir ./data

2. ⚙️ NICT Simulation Code

After downloading preprocessed ICT data, generate NICT data with the following simulation code to construct complete SimNICT dataset:

# Download simulation code
wget https://huggingface.co/datasets/YutingHe-list/SimNICT/blob/main/simnict_generator.py

# Configure paths and run
python simnict_generator.py

The simulation code uses advanced physics-based modeling with ODL (Operator Discretization Library) and ASTRA Toolbox for accurate CT reconstruction simulation.

---

Part 2: SimNICT-AMOS-Sample

SimNICT-AMOS-Sample is a preview subset of SimNICT dataset for quick exploration and prototyping.

Sample Dataset Specifications

• 📂 Source: Selected from AMOS dataset (part of SimNICT)
• 📊 Content: 55 CT volumes (44 train + 11 test)
• 🔬 Coverage: 3 NICT types × 3 fixed severity levels (different from SimNICT dataset's randomized parameters)
• 💾 Size: ~78 MB (1000× smaller than SimNICT dataset)
• 🚀 Format: Preprocessed and optimized for Hugging Face platform

Quick Start with Sample Dataset

from datasets import load_dataset

# Load the preview sample dataset
dataset = load_dataset("YutingHe-list/SimNICT")
sample = dataset["train_previews"][0]

# Access different NICT simulations
ict_image = sample["ICT"]        # Ground truth
ldct_low = sample["LDCT_Low"]    # Low-dose simulation
svct_mid = sample["SVCT_Mid"]    # Sparse-view simulation  
lact_high = sample["LACT_High"]  # Limited-angle simulation

NICT Simulation Parameters (Sample Dataset Only)

Type	Low	Mid	High
LDCT	I₀=1×10⁵	I₀=1×10⁴	I₀=1×10³
SVCT	120 views	60 views	30 views
LACT	120° range	90° range	60° range

---

Citation

@article{liu2024imaging,
        title={Imaging foundation model for universal enhancement of non-ideal measurement ct},
        author={Liu, Yuxin and Ge, Rongjun and He, Yuting and Wu, Zhan and Yang, Shangwen and Gao, Yuan and You, Chenyu and Wang, Ge and Chen, Yang and Li, Shuo},
        journal={arXiv preprint arXiv:2410.01591},
        year={2024}
}

Links

• 📊 Browse sample data
• 📥 Download script
• 📑 Research paper

---

For questions or collaborations, contact via the arXiv paper. Contributions welcome!