---
license: cc-by-nc-4.0
task_categories:
- image-segmentation
modality:
- MRI
language: []
tags:
- medical-imaging
- head-and-neck
- tumor-segmentation
- radiotherapy
- MICCAI-2024
pretty_name: HNTSMRG 2024
size_categories:
- 100<n<1K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: patient_id
    dtype: int32
  - name: timepoint
    dtype: string
  - name: num_slices
    dtype: int32
  - name: selected_slice
    dtype: int32
  - name: t2_middle_slice
    dtype: image
  - name: mask_middle_slice
    dtype: image
  - name: overlay_middle_slice
    dtype: image
  splits:
  - name: train
    num_bytes: 91595083
    num_examples: 300
  download_size: 91605316
  dataset_size: 91595083
---

# HNTSMRG 2024 — Head and Neck Tumor Segmentation for MR-Guided Applications

Public training data of the **HNTS-MRG 2024 MICCAI Challenge**: T2-weighted MRI of head-and-neck cancer patients with expert-consensus segmentations of the primary gross tumor volume (GTVp) and metastatic involved lymph nodes (GTVn), captured at two timepoints — pre-radiotherapy (pre-RT) and mid-radiotherapy (mid-RT).

## Dataset Summary

| Field | Details |
|---|---|
| Modality | T2-weighted anatomical MRI (mix of fat-sup and non-fat-sup) |
| Body Part | Head and neck (oropharyngeal cancer, cancer of unknown primary) |
| Subjects | 150 (training, public) |
| Timepoints / patient | 2 (pre-RT + mid-RT) |
| Labels | 0 = background, 1 = GTVp (primary tumor), 2 = GTVn (involved lymph nodes) |
| Annotation | 3–4 expert physicians per case, fused via STAPLE consensus (faculty arbitration in extreme-disagreement cases) |
| Total Size | ~15 GB |
| License | CC-BY-NC-4.0 |
| Source | Zenodo: https://zenodo.org/records/11199559 |

## Tasks

The HNTSMRG 2024 Challenge defines two segmentation tasks, both supported by this single training set:
- **Task 1** — pre-RT segmentation (GTVp + GTVn on baseline MRI)
- **Task 2** — mid-RT segmentation (GTVp + GTVn on intra-treatment MRI; pre-RT data registered to mid-RT space is provided as a prior)

## Data Structure

```
HNTSMRG24_train/
└── {patient_id}/
    ├── preRT/
    │   ├── {patient_id}_preRT_T2.nii.gz             # pre-RT T2w MRI
    │   └── {patient_id}_preRT_mask.nii.gz           # pre-RT GT mask (STAPLE consensus)
    └── midRT/
        ├── {patient_id}_midRT_T2.nii.gz             # mid-RT T2w MRI
        ├── {patient_id}_midRT_mask.nii.gz           # mid-RT GT mask (STAPLE consensus)
        ├── {patient_id}_preRT_T2_registered.nii.gz  # pre-RT T2 registered to mid-RT space
        └── {patient_id}_preRT_mask_registered.nii.gz # pre-RT mask registered to mid-RT space
```

Pre-RT-to-mid-RT registration was performed with SimpleITK (rigid + deformable) by the challenge organizers and is intended as an optional prior for Task 2.

## Citation

```bibtex
@inproceedings{wahid2025hntsmrg,
  title     = {Overview of the Head and Neck Tumor Segmentation for Magnetic Resonance Guided Applications (HNTS-MRG) 2024 Challenge},
  author    = {Wahid, Kareem A. and Dede, Cem and El-Habashy, Dina M. and Kamel, Serageldin and Rooney, Michael K. and Khamis, Yomna and others and Fuller, Clifton D.},
  booktitle = {Head Neck Tumor Segmentation for MR-Guided Applications},
  series    = {Lecture Notes in Computer Science},
  volume    = {15273},
  pages     = {1--35},
  year      = {2025},
  doi       = {10.1007/978-3-031-83274-1_1}
}

@dataset{wahid2024hntsmrgzenodo,
  title     = {Training Dataset for HNTSMRG 2024 Challenge},
  author    = {Wahid, Kareem and Dede, Cem and Naser, Mohamed and Fuller, Clifton},
  year      = {2024},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.11199559},
  url       = {https://doi.org/10.5281/zenodo.11199559}
}
```