FLARE-MedFM/FLARE-Task1-Pancancer

FLARE Task1 Pancancer Seg Dataset

Data Description

This is the dataset for MICCAI FLARE 2025 Task 1: Pan-cancer segmentation in CT scans. We aim to further promote the development of pan-cancer segmentation and model deployment on low-resource settings.

Data Structure

train_label： This is a partially labeled dataset. Only the primary lesion is labeled in each case. The other lesions may not be labeled (e.g., metastatic lesions).

train_unlabel： Only images are provided and the lesion annotations are not available. README.md For the abdominal images, we also provide the pseudo labels, which are generated by MedSAM2.

validation-public： Both images and annotations are provided. Please do not use them for model training!

validation-hidden： The labels of hidden validation set will not be released. Please submit the segmentation results on codabench to get the metrics.

FLARE-Task1-Pancancer/
├── train_label/
│    ├── DeepLesion5K-MedSAM2/
│        ├── images/
│        ├── labels/
│    ├── imagesTr/
│    └── labelsTr/
├── train_unlabel/
│    ├── AMOS-2350/
│    ├── MSD-506/
│        ├── MSD-Colon/
│        ├── MSD-HapaticVessel/
│        ├── MSD-Liver/
│        ├── MSD-Lung/
│        ├── MSD-Pancreas/
│        └── MSD-Spleen/ ├── validation/
│    ├── HealthyImages-noLesion/
│    ├── Validation-Hidden-Images/
│    ├── Validation-Public-Images
│    └── Validation-Public-Labels └── README.md

Dataset Download Instructions

Participants can download the complete dataset using the following Python script:

from huggingface_hub import snapshot_download

local_dir = "./FLARE-Task1-Pancancer"
snapshot_download(
    repo_id="FLARE-MedFM/FLARE-Task1-Pancancer",
    repo_type="dataset",
    local_dir=local_dir,
    local_dir_use_symlinks=False,
    resume_download=True,
)