|
 
The Cancer Imaging Archive Updates : December 2023
|
|
The Cancer Imaging Archive at the 2023 Radiological Society of North America Meeting
 |
|
As part of the RSNA Deep Learning Lab series, Justin Kirby presented “Accessing freely available public datasets from The Cancer Imaging Archive (TCIA)”. In this course he addressed a variety of use cases for helping data scientists query and download TCIA datasets via Python and Jupyter Notebooks. |
|
|
Collection Updates
After a busy summer, the Cancer Imaging Archive Update Newsletter is back! Over the past four months, TCIA has published or updated 25 data collections and six analysis results collections. That includes over 2,500 new subjects worth of imaging data. Here are just a few of the new collections now available on TCIA:
- NCI’s Cancer Imaging Program (CIP) has funded and released comprehensive annotations of four CPTAC imaging collections (Pancreatic Ductal Adenocarcinoma, Clear Cell Renal Cell Carcinoma, Uterine Corpus Endometrial Carcinoma, and Head and Neck Squamous Cell Carcinoma). Each annotation dataset includes radiologist-reviewed 3D segmentations and seed points identifying the tumor locations. The annotations are provided in DICOM RTSTRUCT format which contains links to the CT, PET/CT, and MRI scans that were analyzed. Additional imaging was also added over the summer. Together, these data will facilitate correlation with the genomic and proteomic analysis data hosted in the Cancer Research Data Commons. Dr. Lalitha K. Shankar, MD, PhD, Chief of Clinical Trials Branch at CIP notes that, "We are pleased to contribute these valuable resources to the scientific community, enabling researchers and AI experts to make noteworthy advancements in cancer imaging analysis as well as multi-omics assessments, to ultimately, improve patient outcomes."
|
- The new Analysis Results dataset, RSNA-ASNR-MICCAI-BraTS-2021 includes imaging and segmentations of 1,480 subjects from the 2021 Brain Tumor Segmentation (BraTS) challenge. It includes ground truth labels for tumor sub-regions including enhancement, necrosis, edema, and MGMT promoter methylation status. These labels add new value to imaging that is largely derived from existing TCIA imaging datasets that included 4 structural MRI modalities (T1, T1c, T2, T2-FLAIR).
|
-
Computed Tomography Images from Large Head and Neck Cohort (RADCURE) was collected for radiation therapy treatment planning and retrospectively reconstructed for quantitative imaging research. The availability of images and radiation therapy structures, along with clinical, demographic and treatment data in RADCURE makes it useful to explore the application of machine learning methods to expedite routine clinical practices, discover new non-invasive biomarkers, or develop prognostic models.
|
- The HNSCC-mIF-mIHC-comparison collection was updated to make it more accessible for AI and deep learning algorithms. It contains re-stained and co-registered multiplex imaging of head-and-neck carcinoma.
|
|
|
The tcia_utils Python package saw the addition of many features including a new module to support mining of collection-level metadata from DataCite, new functions in the NBIA module that allow users to access full DICOM metadata, as well as a new NBIA reporting and charting function, and the ability to visualize DICOM segmentations overlaid on the related SEG/RTSTRUCT images from NBIA. |
|
Significant updates were made to TCIABrowser this summer, which is a plugin that allows users to directly browse and download TCIA datasets for analysis in 3D Slicer. |
|
|
|
|
