title: How to download a TCIA collection in Python description: Download DICOM series from The Cancer Imaging Archive in Python without the NBIA Data Retriever: tcia://Collection through the public REST API, series as folders, one volume per series, licence attached.

How to download a TCIA collection in Python¶

The Cancer Imaging Archive serves public collections through the NBIA REST API: getSeries lists the series of a collection, getImage returns one series as a zip of DICOM files. pardesline wraps both as a dataset source: every series is a sample whose role is a DICOM folder that opens as one Volume.

import pardesline as pl

ds = pl.load("tcia_pancreas_ct")  # bundled: Pancreas-CT, 82 CTs, CC-BY-3.0
print(ds.describe())
s = ds[0]
print(s.id, s.fields)  # PatientID / Modality / SeriesInstanceUID, ImageCount, ...
ct = s.ct  # the series zip is fetched and unpacked once (about 100 MB)
print(ct.info())  # one volume, LPS mm from the DICOM headers
ct.save("pancreas_0001.nii.gz")  # NIfTI + sidecar

Four collections ship with a manifest: tcia_pancreas_ct, tcia_lidc_idri, tcia_nsclc_radiomics_dicom, tcia_prostate_3t.

Any public collection¶

ds = pl.load("tcia://Pancreas-CT")  # same as the bundled one
ds = pl.load("tcia://CPTAC-PDA/CT,MR")  # a collection, restricted to the CT and MR modalities

The collection name is the one on cancerimagingarchive.net. Restricted collections answer 401: they need the TCIA Data Retriever and your credentials, which pardesline does not handle. Licences are per collection (most CC BY 3.0 or 4.0, some non-commercial): the drafted manifest asks you to confirm it from the collection page, the bundled ones record it.

Segmentations and RT structures¶

LIDC-IDRI and NSCLC-Radiomics also carry DICOM SEG and RTSTRUCT series; they appear as their own samples (Modality field SEG / RTSTRUCT). Converting them to label maps on the CT grid is not built in yet: pydicom-seg and rt-utils do it from the folder s.path("seg") returns.

Size and time¶

describe_dataset and s.fields["ImageCount"] tell you the slice count before downloading. LIDC-IDRI is 122 GB in all; ds.filter(Modality="CT").take(5).fetch_all() gets five CTs.