A Large-Scale Benchmark for Source-Free Test-Time Adaptation in Medical Image Segmentation

MedSeg-TTA is a benchmark for test-time adaptation in medical image segmentation. This repository centers on the public leaderboard, the benchmark figures behind it, and the currently available local method implementations organized by paradigm.

Leaderboard

MEDSEG-TTA

👉 Click here to explore the full leaderboard in detail.

Benchmark Overview

The benchmark unifies medical TTA evaluation around a shared surface that connects source-target dataset pairs, paradigm-level comparisons, and local method code roots.

Dataset Coverage

MedSeg-TTA covers seven modalities and multiple cross-domain source-target pairs spanning MRI, CT, US, PATH, DER, OCT, and CXR.

Repository Layout

MedSeg-TTA/
├── medseg_tta/
├── site/
├── feature_level_alignment/
│   ├── GraTa/
│   ├── DANN/
│   ├── UDA-MIMA/
│   └── Testfit/
├── input_level_transformation/
│   ├── SFDA-FSM/
│   ├── DLTTA/
│   ├── AIF-SFDA/
│   ├── STDR/
│   └── RSA/
├── output_level_regularization/
│   ├── DG-TTA/
│   ├── SaTTCA/
│   ├── UPL-SFDA/
│   └── tent/
├── prior_estimation/
│   ├── ExploringTTA/
│   ├── AdaMI/
│   ├── PASS/
│   └── ProSFDA/
└── ASSETS.md

Assets and Licensing

Dataset provenance, third-party code sources, redistribution notes, and license details are documented in ASSETS.md.

Citation

If you find this project useful, please cite:

@article{anonymous2025medsegtta,
  title   = {A Large Scale Benchmark for Test Time Adaptation Methods in Medical Image Segmentation},
  author  = {Anonymous Authors},
  journal = {Anonymous preprint},
  year    = {2025}
}

License

This project is released under the MIT License. See LICENSE for details.