rarf-brats

RARF for BraTS inpainting

This model is presented in RARF: Region-Aware Rectified Flows for 3D Brain MRI Inpainting.

RARF formulates 3D inpainting as a region-restricted noise-to-data transport: during training, the model learns the velocity inside the complete inpainting mask while supervision is restricted to its healthy-tissue subset and the observed anatomy remains fixed; at inference, Gaussian noise inside the mask is transported into a completed volume using four integration points, after which the visible voxels are restored. This checkpoint was developed for the BraTS Local Synthesis of Healthy Brain Tissue via Inpainting Challenge. Complete training, preprocessing, inference, and configuration details are available in the RARF repository.

Citation

Coming soon.

References

1. Ujjwal Baid, Satyam Ghodasara, Suyash Mohan, et al. “The RSNA-ASNR-MICCAI BraTS 2021 Benchmark on Brain Tumor Segmentation and Radiogenomic Classification.” arXiv preprint arXiv:2107.02314, 2021. arXiv:2107.02314.

2. Florian Kofler, Felix Meissen, Felix Steinbauer, et al. “The Brain Tumor Segmentation (BraTS) Challenge: Local Synthesis of Healthy Brain Tissue via Inpainting.” arXiv preprint arXiv:2305.08992, 2024. arXiv:2305.08992.