---
license: cc-by-nc-4.0
tags:
- medical-imaging
- missing-modality
- mri
- ct
---

# SphereFlow: Missing Modality Imputation via Geometric Transport on Hypersphere

Pretrained weights and preprocessed data for our NeurIPS 2026 paper.

[[Paper](https://arxiv.org/abs/2611.04414)] [[Code](https://github.com/WTHH031230/SphereFlow)]

### Contents

```text
checkpoints/
├── sphereflow/
│   ├── brats/
│   │   ├── rae.pt      # shared DINOv2-B encoder + 4 ViT-G decoders
│   │   └── flow.pt     # DiT^DH-XXL flow
│   └── mri_ct/
│       ├── rae.pt      # shared DINOv2-B encoder + 2 ViT-G decoders
│       └── flow.pt     # DiT^DH-XL flow
├── baselines/
│   ├── brats/{MMSyn,MMGAN,MMT,ModImp,M2DN,Zhang}.pt
│   └── mri_ct/{MMSyn,MMGAN,MMT,ModImp,M2DN,Zhang}.pt
└── unet3d/brats.pt     # 3D U-Net segmenter
data/
├── brats_{train_00,train_01,train_02,val,test}.tar
└── mri_ct_{train,val,test}.tar
```

### Usage

Run from the root of the code repository.

```
hf download WTHH031230/SphereFlow --include "checkpoints/*" --local-dir .
hf download WTHH031230/SphereFlow --include "data/*" --local-dir .
for f in data/*.tar; do tar -xf "$f" -C /path/to/processed; done
DATA_ROOT=/path/to/processed bash scripts/preprocess/make_splits.sh --check
```

### Data

The `.npy` slices were produced from the following datasets with the preprocessing scripts in `scripts/preprocess/` of the code repository. They are released under CC BY-NC 4.0, for non-commercial use only.

- **BraTS 2023 adult glioma (BraTS-GLI)**, CC BY-NC 4.0. Data used in this publication were obtained as part of the Brain Tumor Segmentation (BraTS) Challenge project through Synapse ID: [syn51156910](https://www.synapse.org/brats2023). Please cite:
  - U. Baid, et al., The RSNA-ASNR-MICCAI BraTS 2021 Benchmark on Brain Tumor Segmentation and Radiogenomic Classification, arXiv:2107.02314, 2021.
  - B. H. Menze, et al., The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS), IEEE TMI 34(10), 2015.
  - S. Bakas, et al., Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features, Scientific Data 4:170117, 2017.
  - A. Karargyris, et al., Federated benchmarking of medical artificial intelligence with MedPerf, Nature Machine Intelligence 5:799–810, 2023.
- **SynthRAD2023, Task 1 (brain MRI ↔ CT)**, CC BY-NC 4.0, [Zenodo 10.5281/zenodo.7260705](https://doi.org/10.5281/zenodo.7260705). Please cite:
  - A. Thummerer, et al., SynthRAD2023 Grand Challenge dataset: Generating synthetic CT for radiotherapy, Medical Physics 50(7), 2023.

### Citation

    @inproceedings{wang2026sphereflow,
        title     = {SphereFlow: Missing Modality Imputation via Geometric Transport on Hypersphere},
        author    = {Wang, Huihan and Wang, Juncheng and Feng, Yuxiang and Hou, Wenlong and Yu, Guoqi and Xu, Chao and Liu, Yang and Sun, Baigui and Wang, Shujun},
        booktitle = {Advances in Neural Information Processing Systems},
        year      = {2026}
    }