How can I obtain medical imaging data from TCGA?

[ayiiiii1105]

I would like to download medical imaging data from the TCGA database for research purposes; however, as I am based in China, my institution is not among those authorized for access. Is there any way for me to obtain the data I need?

[Avalon2222]

The imaging side of TCGA is open access, so the institutional authorization you are hitting should not apply there.

• Radiology (CT/MR/PET DICOM): the TCGA imaging collections are hosted on The Cancer Imaging Archive (TCIA), not on the GDC. They are public collections: no account, no dbGaP, no institutional approval. Right now TCIA lists 18 of them (TCGA-BLCA, TCGA-BRCA, TCGA-CESC, TCGA-COAD, TCGA-ESCA, TCGA-KICH, TCGA-KIRC, TCGA-KIRP, TCGA-LIHC, TCGA-LUAD, TCGA-LUSC, TCGA-OV, TCGA-PRAD, TCGA-READ, TCGA-SARC, TCGA-STAD, TCGA-THCA, TCGA-UCEC). You can check for yourself with the public REST API, which also works without a login:

  https://services.cancerimagingarchive.net/nbia-api/services/v1/getCollectionValues
  https://services.cancerimagingarchive.net/nbia-api/services/v1/getSeries?Collection=TCGA-LUAD
  

  Download options: the NBIA Data Retriever (GUI), the REST API (getImage?SeriesInstanceUID=... returns a zip per series), or the tcia_utils Python package. MONAI also has this built in: monai.apps.TciaDataset downloads a public TCIA collection and gives you a cached dataset, e.g. TciaDataset(root_dir="data", collection="TCGA-LUAD", section="training", download=True).

• Histopathology (whole-slide images): the diagnostic and tissue slides (.svs) are on the GDC Data Portal and are also open access; only the controlled-access genomic/clinical files (raw sequencing, germline data) need dbGaP authorization, which is the part that requires an approved institution. Filter by Data Type = Slide Image and Access = open, then download with the gdc-client and a manifest.

So: for radiology go through TCIA, for slides use the GDC open-access tier. If a download from China is slow or blocked, the REST API lets you fetch series one by one and resume, which is easier than the big manifest downloads.

[ayiiiii1105]

Thank you very much for your reply. However, what I am looking for is glioma imaging data, such as that for TCGA-GBM and TCGA-LGG. In my latest attempt to access them, I found that these datasets are not publicly available. Furthermore, obtaining the necessary authorization is effectively impossible for a student based in China. I will continue to explore other avenues to see if I can locate the data elsewhere—perhaps from someone who has already downloaded it. Thanks again for your response.

Ayiiii

[Avalon2222]

You're right, I should have checked those two specifically: TCGA-GBM and TCGA-LGG are listed under TCIA's "NIH Controlled Data Access Policy", because the raw head MRIs could be used to reconstruct a face. That's why they are not in the public API.

The good news is that the same TCGA glioma patients are available in skull-stripped, processed form under CC BY, which is what most people train on anyway:

• BraTS-TCGA-GBM and BraTS-TCGA-LGG (TCIA "analysis results", Bakas et al. 2017): co-registered, skull-stripped T1 / T1-Gd / T2 / FLAIR NIfTI with tumour segmentations and radiomic features, licensed CC BY 3.0. The downloadable training set is 102 GBM and 65 LGG patients (767 MB and 536 MB). Files are named by TCGA patient ID, so you can still join them to the open clinical/molecular data on GDC (IDH, 1p/19q, survival).
  • https://www.cancerimagingarchive.net/analysis-result/brats-tcga-gbm/
  • https://www.cancerimagingarchive.net/analysis-result/brats-tcga-lgg/
• MONAI's DecathlonDataset(task="Task01_BrainTumour"): 750 glioma cases from BraTS 2016/2017, which include these TCGA cases, downloaded straight from a public S3 bucket with no account (7.6 GB tar). Easiest route if you just need glioma MRI + labels inside MONAI.

If you need more patients or raw-ish clinical MRI rather than BraTS preprocessing, two large glioma collections on TCIA are fully public (CC BY 4.0):

• UPENN-GBM: 630 GBM patients, multiparametric MRI (DICOM and NIfTI) with segmentations, plus clinical and molecular data.
• UCSF-PDGM: 495 diffuse glioma patients (grades 2–4) with IDH/MGMT/1p19q status, NIfTI including DWI/DTI.

The last three don't overlap with TCGA, but they are bigger and cover the same tumour types, and none of them needs institutional approval.