CADS (Newly Released) - CT-TRI (Triphasic Contrast-Enhanced Abdominal CTs)

License

CC BY-NC-SA 4.0
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License

A few images containing parts of the face have been defaced. It is the responsibility of every user to ensure that the data is still considered anonymized.

Citation

Paper BibTeX:

@article{ruhling2022automated,
  title={Automated detection of the contrast phase in MDCT by an artificial neural network improves the accuracy of opportunistic bone mineral density measurements},
  author={R{\"u}hling, Sebastian and Navarro, Fernando and Sekuboyina, Anjany and El Husseini, Malek and Baum, Thomas and Menze, Bjoern and Braren, Rickmer and Zimmer, Claus and Kirschke, Jan S},
  journal={European Radiology},
  volume={32},
  number={3},
  pages={1465--1474},
  year={2022},
  publisher={Springer}
}

Dataset description

CT images were retrospectively selected from our digital picture archiving communication system (PACS) (Sectra AB). We included 206 consecutive patients with a routine abdominal triphasic MDCT scan (dedicated to investigating liver or kidney pathologies) acquired between September 2016 and November 2019.

Exclusion criteria were:

Previous contrast application < 2 h prior to the triphasic CT (n = 6)
Contrast administration via the inferior vena cava (n = 2)
Insufficient coverage of the abdomen (n = 5)

The final dataset consisted of 193 adults (48 women and 145 men), with a mean age of 62.4 ± 14.6 years. Most patients included were suspected or confirmed to have liver or kidney cancer, resulting in a higher male-to-female ratio.

All CT scans were performed on the same MDCT scanner (IQon Spectral CT; Philips Medical Care) using a standardized protocol. The routine abdominal contrast-enhanced images were acquired in helical mode with:

Peak tube voltage of 120 kVp
Axial slice thickness of 0.9–1 mm
Adaptive tube load

After the acquisition of pre-contrast images, all patients received standardized intravenous administration of contrast agent (Iomeron 400; Bracco) using a high-pressure injector (Fresenius Pilot C; Fresenius Kabi). Thirty-seven patients additionally received oral contrast (Barilux Scan; Sanochemia Diagnostics).

Post-contrast scans were performed in both arterial (AR) and portal venous (PV) phases:

AR phase: Triggered after a threshold of 120 HU was reached in a region of interest (ROI) in the aorta
PV phase: Performed after a standard delay of 80 s

For further analysis, spine reformations were reconstructed using a filtered back projection favoring sharpness over noise (bone kernel). The contrast phase was visually assessed by two radiologists (2 and 19 years of clinical experience) and served as the ground truth.

The CT data were converted into the Neuroimaging Informatics Technology Initiative (NIfTI) format and resampled to a maximum of 1 mm isotropic spatial resolution.

(Additional information on the dataset can be found in the paper citation above)

Number of CT volumes: 586

Contrast: Triphasic contrast-enhanced (non-enhanced, arterial, and portal venous phases)

CT body coverage: Abdomen (very few images, 6 cases, included partial face coverage)

Does the dataset include any ground truth annotations?: No

Original GT annotation targets: -

Number of annotated CT volumes: -

Annotator: -

Acquisition centers: TUM Klinikum Rechts der Isar, Munich, Germany

Pathology/Disease: Higher prevalence of liver and kidney pathologies

Original dataset download link: CADS-dataset HuggingFace repository

Original dataset format: nifti