# IDC data release v25 Oct 2026: new ground truth for AI

**URL:** <https://discourse.canceridc.dev/t/idc-data-release-v25-oct-2026-new-ground-truth-for-ai/810>\
**Category:** Announcements\
**Created:** [October 14, 2026, 9:02pm UTC](https://discourse.canceridc.dev/t/idc-data-release-v25-oct-2026-new-ground-truth-for-ai/810 "2026-10-14T21:02:28Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![fedorov](https://sea1.discourse-cdn.com/flex015/user_avatar/discourse.canceridc.dev/fedorov/32/3_2.png) [@fedorov](https://discourse.canceridc.dev/u/fedorov)\
**Post date:** [October 14, 2026, 9:02pm UTC](https://discourse.canceridc.dev/t/idc-data-release-v25-oct-2026-new-ground-truth-for-ai/810/1 "2026-10-14T21:02:29Z")

</div>

NCI Imaging Data Commons (IDC) v25 is live!

This release is about **new ground truth for imaging AI**. It adds **3 new collections** and **2 new analysis results** , about **0.65 terabytes (TB)** of new data, bringing IDC to **99.9 TB** across 179 collections, 86,362 cases and 1,044,191 DICOM series. Highlights include 900 whole-body FDG-PET/CT patients with segmented lesions, 2,004 independent radiologist reads of lung nodules on National Lung Screening Trial (NLST) CT, and IDC’s first labelmap segmentations. The release also adds **80 clinical data tables**. Here is a summary of what’s new.

 ![IDC Data Release v25 — Highlights - slide 1](https://us1.discourse-cdn.com/flex015/uploads/canceridc/original/1X/350ddbecf00e5945d1dbdd518c3226edf0957ea0.jpeg)

## 🩻 Radiology

New collections curated by The Cancer Imaging Archive (TCIA), ingested “as is” and now available via the resources and tools maintained by IDC:

- **[FDG-PET-CT-Lesions](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=fdg_pet_ct_lesions)** — The largest addition in v25: 1,014 whole-body FDG-PET/CT studies from 900 patients acquired at University Hospital Tübingen, including 501 studies of patients with lymphoma, melanoma or non-small cell lung cancer (NSCLC) and 513 negative controls. All FDG-avid lesions are manually segmented and provided as DICOM SEG (CT + PT + SEG, ~409 GB). This is the dataset behind the autoPET MICCAI 2022 challenge, and a natural companion to PSMA-PET-CT-Lesions added in v24. IDC hosts the defaced images released by TCIA under CC BY 4.0 ([DOI](https://doi.org/10.7937/gkr0-xv29)).
- **[QIN-SARCOMA](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=qin_sarcoma)** — Longitudinal dynamic contrast-enhanced (DCE) MRI of 15 soft-tissue sarcoma patients from Oregon Health & Science University, imaged up to three times to assess response to preoperative chemoradiation: 38 studies, 2,168 MR series. Formerly limited access; TCIA made it public in April 2026 ([DOI](https://doi.org/10.7937/K9/TCIA.2016.fXL9sESs)).
- **[MRI-DIR](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=mri_dir)** — Two MD Anderson image sets for measuring deformable image registration accuracy in MRI. The first is a porcine phantom with ten implanted gold markers, imaged with CT and T1/T2 MR in four compression states. The second is eight synthetic head and neck subjects with pre- and post-radiotherapy T1/T2 images and RTSTRUCT contours of the parotid, submandibular and sublingual glands (9 subjects, 61 series) ([DOI](https://doi.org/10.7937/K9/TCIA.2018.3f08iejt)).

Updated collections:

- **Cancer Moonshot Biobank (CMB)** — TCIA published new versions of nine CMB collections: [CMB-LCA](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=cmb_lca), [CMB-CRC](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=cmb_crc), [CMB-BRCA](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=cmb_brca), [CMB-MML](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=cmb_mml), [CMB-OV](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=cmb_ov), [CMB-GEC](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=cmb_gec), [CMB-PCA](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=cmb_pca), [CMB-MEL](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=cmb_mel) and [CMB-AML](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=cmb_aml). Together with Vestibular-Schwannoma-MC-RC (below), they add 3,512 radiology series (~224 GB) for 246 patients, 84 of them new to IDC. New content includes breast MRI with post-contrast subtractions in CMB-BRCA, and CMB-OV’s first MR series. The update is purely additive.
- **[Vestibular-Schwannoma-MC-RC](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=vestibular_schwannoma_mc_rc)** — Updated to TCIA version 2, adding 133 MR series and 36 new patients ([DOI](https://doi.org/10.7937/hrzh-2n82)).
- **[EA1141](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=ea1141)** — Updated to TCIA version 2. It adds 444 previously missing images and 3 new mammography series, and replaces 45,121 instances to correct a malformed UID in `ReferencedStudySequence` ([DOI](https://doi.org/10.7937/2bas-hr33)).

## 🎯 Analysis results

- **[LIDC-annot-NLST501](https://portal.imaging.datacommons.cancer.gov/explore/filters/?analysis_results_id=LIDC-annot-NLST501)** — LIDC-IDRI-style lung nodule annotations on 501 NLST chest CTs, built and curated by Median Technologies. Each scan was read independently four times by a panel of 20 board-certified radiologists. That gives 2,004 DICOM SEG series with 6,505 nodule and mass segments, plus a TotalSegmentator lung mask; the reader is recorded in each SEG. The cohort is cancer-enriched: 152 of 501 patients have histopathology-confirmed lung cancer. The release also includes 24 corrected CT series that replace NLST source series with slice problems ([DOI](https://doi.org/10.7937/y2dx-e334)).

- **[RCC-AID](https://portal.imaging.datacommons.cancer.gov/explore/filters/?analysis_results_id=RCC-AID)** — Voxel-level kidney, renal mass and cyst annotations for 129 CT scans from 91 patients across TCGA-KIRC, TCGA-KIRP and TCGA-KICH. It was annotated with a model-initialized, human-corrected workflow (described in [MELBA 2026](https://doi.org/10.59275/j.melba.2026-2dg2)) and converted to DICOM SEG by IDC with dcmqi ([DOI](https://doi.org/10.5281/zenodo.22284133)). **These are IDC’s first LABELMAP segmentations.** If you run into problems reading them with your tools, please let us know in this thread!

## 🔬 Digital pathology

- **[CATCH](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=catch)** — The canine cutaneous tumor slides now include their expert annotations as DICOM: 350 Microscopy Bulk Simple Annotations (ANN) series with polygon annotations for 13 histologic classes, viewable directly in the IDC slide viewer ([DOI](https://doi.org/10.5281/zenodo.18526942)).
- **JPEG 2000 color metadata corrected across 52 collections.** 40,660 slide microscopy instances, about 10% of all SM instances in IDC, carried an incorrect Photometric Interpretation. Decoders that trust this attribute could render the slides with wrong colors, as reported in [openslide/openslide#772](https://github.com/openslide/openslide/issues/772). The JPEG 2000 pixel data is carried over byte for byte, and SOPInstanceUIDs are unchanged, so existing SEG/ANN references still resolve; only `crdc_instance_uuid` changes. The IDC slide viewer (SLIM) renders these slides correctly before and after the fix. The change matters if you load the files into other tools that rely on the header, such as OpenSlide. If you downloaded affected slides before v25 and use such tools, re-download them to get the corrected files.
- **HTAN multiplexed imaging presentation states re-encoded.** All 324 Advanced Blending Presentation State instances in HTAN-WUSTL, HTAN-HMS, HTAN-OHSU and HTAN-TNP-SARDANA now use a standard 8-bit palette lookup table, as the DICOM standard requires. Note: these 107 series have **new SeriesInstanceUIDs and SOPInstanceUIDs** , so update any scripts that reference them by UID.
- **[BoneMarrowWSI-PediatricLeukemia](https://portal.imaging.datacommons.cancer.gov/explore/filters/?collection_id=bonemarrowwsi_pediatricleukemia)** gains one additional patient, and 20 duplicate ANN/SEG series on 10 TCGA-LUAD slides were removed from **[Pan-Cancer-Nuclei-Seg-DICOM](https://portal.imaging.datacommons.cancer.gov/explore/filters/?analysis_results_id=Pan-Cancer-Nuclei-Seg-DICOM)**.

## 📋 Clinical data

80 new clinical tables:

- TCGA clinical data for 32 projects
- 10 CPTAC projects
- All five HTAN atlases
- FDG-PET-CT-Lesions
- Updated LDCT-and-Projection-data tables

Discover them with `list_clinical_tables` in the IDC MCP server or the skill, or in the `idc_v25_clinical` BigQuery dataset.

## 🚀 Get started

- Instructions on setting up agentic interfaces to IDC: [https://learn.canceridc.dev/ai-assistants/agents](https://learn.canceridc.dev/ai-assistants/agents) - the easiest interface for interacting with IDC
- Explore in the portal: [https://portal.imaging.datacommons.cancer.gov](https://portal.imaging.datacommons.cancer.gov)
- Programmatic access: `pip install --upgrade idc-index`
- Getting started notebooks: [GitHub - ImagingDataCommons/IDC-Tutorials](https://github.com/ImagingDataCommons/IDC-Tutorials)

Full data release notes: [Data release notes | IDC User Guide](https://learn.canceridc.dev/data/data-release-notes#v25-oct-2026)