MeisenMeister

MeisenMeister is a framework for breast cancer classification on DCE-MRI. It is designed to help develop reproducible multi-stage pipelines, from dataset fingerprinting and experiment planning to ROI preprocessing, fold-safe training, benchmarking, and ROI-level inference.

🏆 Winning solution of the MICCAI 2025 ODELIA Breast MRI Challenge on Grand Challenge: https://odelia2025.grand-challenge.org/

Installation

conda create -n meisenmeister python=3.12 -y
conda activate meisenmeister
pip install -e .

Environment Configuration

MeisenMeister resolves datasets and outputs via three required storage roots:

• MM_RAW
• MM_PREPROCESSED
• MM_RESULTS

The code also accepts lowercase variants (mm_raw, mm_preprocessed, mm_results).

export MM_RAW=/path/to/mm_raw
export MM_PREPROCESSED=/path/to/mm_preprocessed
export MM_RESULTS=/path/to/mm_results

Workflow At A Glance

1. Raw data is discovered under MM_RAW (Dataset_<id>_* naming).
2. Fingerprinting and planning define preprocessing geometry and output format.
3. Preprocessing writes ROI tensors and metadata to MM_PREPROCESSED.
4. Split generation writes leakage-safe five-fold splits.
5. Training and benchmarking write experiment artifacts to MM_RESULTS.
6. Inference runs either from local dataset context or from a portable model folder.

Quickstart

# 1) Optional: create breast masks if masksTr is missing
mm_create_breast_segmentations -d 1

# 2) Fingerprint + plan + preprocess
mm_extract_dataset_fingerprint -d 1 --num-workers 8
mm_plan_and_preprocess -d 1 --num-workers 8

# 3) Build 5-fold split file
mm_create_5fold -d 1

# 4) Train fold 0
mm_train -d 1 -f 0

# 5) Predict with fold ensemble
mm_predict -d 1 -i /path/to/images -o /path/to/preds -f 0 1 2 3 4

Installed CLI Commands

Command	Purpose
mm_extract_dataset_fingerprint	Compute dataset fingerprint statistics used by planning.
mm_create_breast_segmentations	Create breast segmentations in masksTr.
mm_homogenize	Resample channels into _0000 space and overwrite raw NIfTI files.
mm_plan_experment	Build mmPlans.json from fingerprint data.
mm_preprocess	Generate preprocessed ROI data from mmPlans.json.
mm_plan_and_preprocess	Run planning and preprocessing in one step.
mm_create_5fold	Create case-aware leakage-safe splits.json.
mm_train	Run fold-based training with registered trainer classes.
mm_benchmark_train	Benchmark train/validation throughput with warmup controls.
mm_predict	Run ROI-level inference using local dataset + results roots.
mm_predict_from_modelfolder	Run inference from a portable experiment folder.

Storage Layout

• MM_RAW: source datasets (Dataset_001_*, imagesTr, masksTr, dataset.json)
• MM_PREPROCESSED: dataset_fingerprint.json, mmPlans.json, splits.json, ROI outputs
• MM_RESULTS: experiment folders, checkpoints, logs, plots, evaluation outputs

Documentation

• High-Level Overview
• Setup and Paths
• CLI Guide
• Training and Experiments
• Data and Pipeline Notes

License

The repository source code is licensed under the Apache License 2.0 (see LICENSE).

Model weights are licensed under CC BY-NC-SA 4.0 due to downstream dataset licensing constraints from the data used for training.

Citation

If you use MeisenMeister in research, please cite:

Hamm, B., Kirchhoff, Y., Rokuss, M., and Maier-Hein, K., MeisenMeister: A Simple Two Stage Pipeline for Breast Cancer Classification on MRI, arXiv:2510.27326 [cs.CV], 2025.

Paper: https://arxiv.org/pdf/2510.27326

@article{hamm2025meisenmeister,
  title={MeisenMeister: A Simple Two Stage Pipeline for Breast Cancer Classification on MRI},
  author={Hamm, Benjamin and Kirchhoff, Yannick and Rokuss, Maximilian and Maier-Hein, Klaus},
  journal={arXiv preprint arXiv:2510.27326},
  year={2025}
}