{"success":true,"data":{"author":"NCI Imaging Data Commons; FDG-PET-CT-Lesions contributors; tables and sample by Scigantic","bucketAccessType":"public","bucketListable":true,"bucketName":"scigantic-idc-open","bucketPrefix":"fdg_pet_ct_lesions/","category":"Health & Biomedical","cloudProvider":"aws","companionArchiveIds":["270d1e86-d234-4234-9cac-dc846bcad87f"],"createdAt":"2026-11-17T16:33:07.411Z","description":"FDG-PET-CT-Lesions on the NCI Imaging Data Commons (IDC v25): 900 patients, 1,014 studies, 3,042 DICOM series, 409.13 GB; segmentations of Tissue. Cancer type: Lymphoma, Melanoma, Non-small Cell Lung Cancer; location: lung, lymph node, skin of body. Modalities: CT 1,014 series, PT 1,014 series, SEG 1,014 series. Segmentations (1,014 SEG series; manual 1,014) cover, by number of series: Tissue (1,014). This Scigantic archive is the collection's complete series-, study- and patient-level metadata as parquet (from idc-index, IDC v25), 1 IDC clinical table(s) with a column dictionary, and a real sample of the images: one PT series, complete (79 MB, CC BY 4.0); one SEG series, complete (4 MB, CC BY 4.0). Every series row carries its public S3 location (s3://idc-open-data/<crdc_series_uuid>/), so any series can be pulled on demand with anonymous boto3; the collection itself is not copied. License: CC BY 4.0 (3,042 series). IDC's description: 1,014 whole-body FDG-PET/CT studies from 900 patients, acquired at University Hospital Tübingen (2014-2018) on a single Siemens Biograph mCT: 501 studies of patients with malignant lymphoma, melanoma or non-small cell lung cancer (NSCLC), and 513 negative controls without PET-positive malignant lesions. All FDG-avid malignant lesions were manually segmented on the PET images and are provided as DICOM SEG. The dataset was used in the autoPET MICCAI 2022 (https://autopet.grand-challenge.org/) challenge. IDC hosts the defaced version of the images, which TCIA released under CC BY 4.0; the original images remain TCIA Restricted.","displayTitle":"IDC FDG-PET-CT-Lesions: Lymphoma, Melanoma, Non-small Cell Lung Cancer CT, PET (900 patients)","dockerImageUrl":"531775725781.dkr.ecr.us-east-1.amazonaws.com/scigantic-notebook-pathology:latest","doi":"10.7937/gkr0-xv29","fuseMountBrowsable":true,"id":"c90ac470-bebc-5dd4-b0ce-523636c2b89a","license":"Per series (series.parquet license_short_name): CC BY 4.0 3,042. Cite the source DOI(s) and the IDC.","referenceUrl":"https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=fdg_pet_ct_lesions","region":"us-east-1","schemaCard":{"accessHints":[{"language":"python","snippet":"# Task: read this collection's tables from anywhere, no mount and no credentials (pip install scigantic-nci)\nfrom scigantic_nci import idc\nseries = idc.series('fdg_pet_ct_lesions')      # one row per DICOM series, with its public S3 location\npatients = idc.patients('fdg_pet_ct_lesions')\nprint(len(patients), 'patients;', series.Modality.value_counts().to_dict())\n# or without the library (needs s3fs): pd.read_parquet('s3://scigantic-idc-open/fdg_pet_ct_lesions/series.parquet', storage_options={'anon': True})\n# The other snippets read /mnt/archive, the mount path inside a Scigantic notebook; elsewhere use the frames above.\n"},{"language":"python","snippet":"# Task: pull one series from the IDC bucket on demand (anonymous S3)\nimport boto3, pandas as pd, os\nfrom botocore import UNSIGNED\nfrom botocore.config import Config\nseries = pd.read_parquet('/mnt/archive/series.parquet')\nrow = series[series.Modality == 'CT'].sort_values('series_size_MB').iloc[0]\nprint(row.PatientID, row.Modality, row.instanceCount, 'instances', round(row.series_size_MB, 1), 'MB')\ns3 = boto3.client('s3', region_name='us-east-1', config=Config(signature_version=UNSIGNED))\nos.makedirs('/tmp/series', exist_ok=True)\nfor o in s3.list_objects_v2(Bucket=row.aws_bucket, Prefix=row.crdc_series_uuid + '/')['Contents']:\n    s3.download_file(row.aws_bucket, o['Key'], '/tmp/series/' + o['Key'].split('/')[-1])\n"},{"language":"python","snippet":"# Task: stack a radiology series into a volume in acquisition order\nimport glob, numpy as np, pydicom\nds = [pydicom.dcmread(f) for f in glob.glob('/tmp/series/*.dcm')]\nds.sort(key=lambda d: (float(d.ImagePositionPatient[2]) if 'ImagePositionPatient' in d else 0.0, int(d.get('InstanceNumber', 0))))\nvol = np.stack([d.pixel_array for d in ds]).astype(np.float32)\nif ds[0].Modality == 'CT':\n    vol = vol * float(ds[0].get('RescaleSlope', 1)) + float(ds[0].get('RescaleIntercept', 0))  # Hounsfield units\nprint(vol.shape, ds[0].get('PixelSpacing'), ds[0].get('SliceThickness'))\n"},{"language":"python","snippet":"# Task: assemble one level of a DICOM whole-slide image from its tiles (pydicom only)\nimport numpy as np, pydicom\nd = pydicom.dcmread(path_to_level_dcm)\nframes = d.pixel_array\nframes = frames[None] if frames.ndim == 3 else frames\nR, C, r, c = int(d.TotalPixelMatrixRows), int(d.TotalPixelMatrixColumns), int(d.Rows), int(d.Columns)\nncol = -(-C // c)\ncanvas = np.zeros((-(-R // r) * r, ncol * c, 3), dtype=frames.dtype)\nfor i, f in enumerate(frames):\n    y, x = (i // ncol) * r, (i % ncol) * c\n    canvas[y:y + r, x:x + c] = f\nimage = canvas[:R, :C]  # or: from wsidicom import WsiDicom; WsiDicom.open(folder).read_region(...)\n"},{"language":"python","snippet":"# Task: join the clinical table to the imaging series\nimport pandas as pd\nclin = pd.read_parquet('/mnt/archive/clinical/fdg_pet_ct_lesions_clinical.parquet')\ndictionary = pd.read_parquet('/mnt/archive/clinical/dictionary.parquet')\nseries = pd.read_parquet('/mnt/archive/series.parquet')\nprint(dictionary[['column', 'column_label']].head(20).to_string())\njoined = series.merge(clin, left_on='PatientID', right_on='dicom_patient_id', how='left')\nprint(joined.shape, joined.dicom_patient_id.notna().mean().round(2), 'of series have a clinical row')\n"}],"caveats":["Layout: series.parquet (3,042 rows, one per DICOM series), studies.parquet, patients.parquet, clinical/<table>.parquet + clinical/dictionary.parquet, samples.parquet (one row per mirrored sample with its license), sample/<Modality>_<SeriesInstanceUID>/*.dcm, README.md, BUILD_REPORT.json. The full collection (409.13 GB) is NOT on this mount; only the sample folders hold pixels.","Fetching a series: s3://idc-open-data/<crdc_series_uuid>/ holds one <instance_uuid>.dcm per SOP instance; the buckets are public and anonymous (botocore.UNSIGNED) reads work. Take crdc_series_uuid, aws_bucket and series_size_MB from series.parquet and check the size before pulling into a pod with a few GB of disk. s5cmd and `pip install idc-index` are the bulk tools.","PatientID in IDC is the TCIA/collection patient id (e.g. PETCT_0011f3deaf). StudyInstanceUID and SeriesInstanceUID are the DICOM UIDs used by every viewer and by the IDC portal (see get_viewer_URL in idc-index).","Radiology series are one .dcm per slice; sort by ImagePositionPatient[2] (fall back to InstanceNumber) before stacking. CT pixel values need RescaleSlope/RescaleIntercept to become Hounsfield units; window for the organ you are looking at (soft tissue about -160 to 240 HU). PET needs SUV scaling from the RadiopharmaceuticalInformationSequence and patient weight.","Derived series (SEG) are analysis results (segmentations, structure sets, structured reports, annotations) stored as their own series; they reference the source image series through ReferencedSeriesSequence, and analysis_results.parquet lists which analysis produced them and under what license. They are not in the sample folders.","Clinical: fdg_pet_ct_lesions_clinical come from IDC's clinical tables (TCIA-supplied); join on dicom_patient_id = PatientID. clinical/dictionary.parquet has the human label and the coded values for every column; coded columns hold the code, not the label.","Licenses are per series (CC BY 4.0: 3,042); series.parquet license_short_name is authoritative. Cite the DOI of the series you use (series.parquet source_DOI). The collection's own DOI is 10.7937/gkr0-xv29. Also cite the IDC (Fedorov et al., Cancer Research 2021, doi:10.1158/0008-5472.CAN-21-0950; Fedorov et al., RadioGraphics 2023, doi:10.1148/rg.230180).","PatientAge is a DICOM age string (e.g. 062Y) on 3,042 series; PatientSex is set on 894 of 900 patients (M 500, F 394). Manufacturers: SIEMENS (2028), QIICR (1014)."],"columns":[{"name":"collection_id","type":"str"},{"name":"PatientID","type":"str"},{"name":"SeriesInstanceUID","type":"str"},{"name":"StudyInstanceUID","type":"str"},{"name":"source_DOI","type":"str"},{"name":"PatientAge","type":"str"},{"name":"PatientSex","type":"str"},{"name":"StudyDate","type":"str"},{"name":"BodyPartExamined","type":"str"},{"name":"Modality","type":"str"},{"name":"Manufacturer","type":"str"},{"name":"SeriesDescription","type":"str"},{"name":"instanceCount","type":"Int64"},{"name":"license_short_name","type":"str"},{"name":"aws_bucket","type":"str"},{"name":"crdc_series_uuid","type":"str"},{"name":"series_aws_url","type":"str"},{"name":"series_size_MB","type":"float64"}],"format":"v2-hybrid","generatedAt":"2026-11-19T22:51:36.130Z","layout":{"example":"series.parquet row: PatientID PETCT_0011f3deaf, Modality CT, 391 instances, 206.5 MB, s3://idc-open-data/fddb1857-7a7a-438c-b3b4-7257fd4ecf3d/*","pattern":"clinical/dictionary.parquet  22 x 4\nclinical/fdg_pet_ct_lesions_clinical.parquet  3,042 x 22\npatients.parquet  900 x 7\nsamples.parquet  2 x 13\nseries.parquet  3,042 x 31\nstudies.parquet  1,014 x 9\nsample/PT_1.3.6.1.4.1.14519.5.2.1.269817471495327477714083258815666184166/  243 .dcm (78.9 MB, CC BY 4.0)\nsample/SEG_1.3.6.1.4.1.14519.5.2.1.307310350378268960217419474630983173839/  1 .dcm (4.1 MB, CC BY 4.0)"},"mountPath":"/mnt/archive","rowCount":"900 patients, 1,014 studies, 3,042 series; 409.13 GB in IDC","sampleObjects":["README.md: IDC description, modality table, licenses, DOIs, how to fetch","series.parquet: the index; one row per series with its S3 location","sample/PT_1.3.6.1.4.1.14519.5.2.1.269817471495327477714083258815666184166/: PT sample (CC BY 4.0)","sample/SEG_1.3.6.1.4.1.14519.5.2.1.307310350378268960217419474630983173839/: SEG sample (CC BY 4.0)"],"starterCell":"# FDG-PET-CT-Lesions: what is in the collection, and one real PT series from the sample folder\nimport os, glob, pandas as pd, numpy as np\nM = os.environ.get('SCIGANTIC_MOUNT_PATH', '/mnt/archive')\nseries = pd.read_parquet(f'{M}/series.parquet')\nprint(series.groupby('Modality').agg(series=('SeriesInstanceUID', 'count'), patients=('PatientID', 'nunique'), GB=('series_size_MB', lambda s: round(s.sum() / 1024, 1))))\nimport pydicom\nfiles = sorted(glob.glob(f'{M}/sample/PT_1.3.6.1.4.1.14519.5.2.1.269817471495327477714083258815666184166/*.dcm'))\nds = pydicom.dcmread(files[0])\nprint(len(files), 'files;', ds.Modality, ds.get('Rows'), 'x', ds.get('Columns'), ds.get('ImageType'))\n","totalObjects":"6 parquet tables plus 244 sample DICOM files (83 MB)"},"summary":"FDG-PET-CT-Lesions on the NCI Imaging Data Commons (IDC v25): 900 patients, 1,014 studies, 3,042 DICOM series, 409.13 GB; segmentations of Tissue.","title":"IDC FDG-PET-CT-Lesions: Lymphoma, Melanoma, Non-small Cell Lung Cancer CT, PET (900 patients)","updatedAt":"2026-11-19T22:52:46.131Z","visibility":"public","websiteUrl":"https://imaging.datacommons.cancer.gov/"}}