anyBrainer 🧠

anyBrainer is a PyTorch Lightning + MONAI–based framework for pretraining and finetuning foundation models for brain MRI analysis.

It extends MONAI and Lightning with custom transforms, optimizers/schedulers, and inferers to provide a flexible end-to-end pipeline for a wide range of downstream tasks.

The framework supports contrastive learning pretraining as well as downstream tasks:

• Classification (e.g., brain infarct detection)
• Segmentation (e.g., meningioma segmentation)
• Regression (e.g., brain age prediction)
• Multimodal fusion setups

anyBrainer was used for the MICCAI FOMO25 challenge across all tasks:

• Pretraining
• Finetuning for classification, segmentation, and regression
• Inference and containerized submission

Features

• Config-driven workflows: Run pretraining/finetuning/inference via .yaml configs
  (pretrain_cl.yaml, finetune_cls.yaml, finetune_seg.yaml, finetune_reg.yaml)
• Custom MONAI transforms for MRI-specific augmentation and preprocessing
• Multi-param optimizer & scheduler factories for flexible training setups
• Custom inferers for efficient sliding-window and multimodal inference
• Contrastive learning setup for foundation model pretraining
• Built on PyTorch Lightning for reproducibility and scalability
• Compatible with Weights & Biases logging for detailed experiment tracking
• Containerization-ready: inference scripts and Apptainer specs in app/

Installation

Clone the repository and install via:

# For basic dependencies:
pip install -e .

# For all dependencies:
pip install -e ".[dev,test]"

Usage

Training and inference are fully config-driven. Example configs are provided in the repo.

# Pretraining (contrastive)
anyBrainer TrainWorkflow "pretrain_cl.yaml"

# Finetuning for classification
anyBrainer CVWorkflow "finetune_cls.yaml"

# Finetuning for segmentation
anyBrainer CVWorkflow "finetune_seg.yaml"

# Finetuning for regression
anyBrainer CVWorkflow "finetune_reg.yaml"

Project Structure

anyBrainer/
├── app/                  # Inference scripts and Apptainer files (containerization)
├── src/                  # Source code
│   └── anyBrainer/       # Main package
│       ├── config/       # Config management
│       ├── core/         # Core modules: data, engines, networks, losses, transforms, etc.
│       ├── factories/    # Factories for models, optimizers, schedulers
│       ├── interfaces/   # Base classes and interfaces
│       ├── log/          # Logging utilities
│       ├── registry/     # Registries for models, transforms, losses, etc.
│       ├── __main__.py   # CLI entry point
│       └── __init__.py
├── tests/                # Unit and integration tests
├── finetune_cls.yaml     # Example config: classification
├── finetune_reg.yaml     # Example config: regression
├── finetune_seg.yaml     # Example config: segmentation
├── pretrain_cl.yaml      # Example config: contrastive pretraining
├── pyproject.toml        # Project dependencies
├── LICENSE
└── README.md

Citation

If you use anyBrainer in your research, please cite this repository:

@misc{anybrainer2025,
  title        = {anyBrainer: A PyTorch Lightning + MONAI Framework for Brain MRI Pretraining and Finetuning},
  author       = {Petros Koutsouvelis},
  year         = {2025},
  url          = {https://github.com/MaastrichtU-CDS/anyBrainer},
  note         = {GitHub repository}
}

License

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