MedUnify

MedUnify is a unified medical workflow SDK for Python. It wraps mature ecosystems (pydicom, nibabel, SimpleITK, WFDB, fhir.resources, optional MONAI) behind small adapters and a stable, developer-friendly surface API.

This project is not a medical device, not a diagnostic engine, and not a replacement for PACS, EHRs, or regulated clinical software. It exists to help researchers and engineers integrate imaging, biosignals, FHIR records, and research-grade ML tooling with less glue code.

Why MedUnify?

Hospital and research workflows often combine DICOM/NIfTI pipelines, waveform archives, FHIR JSON, and PyTorch/MONAI experiments. Each ecosystem is powerful on its own, but public codebases frequently reimplement the same adapters. MedUnify centralizes those integrations behind explicit adapter modules and narrow facades so applications can grow without turning into an unmaintainable monolith.

Installation

Requires Python 3.11+.

pip install -e .

Optional AI stack (PyTorch + MONAI):

pip install -e ".[ai]"

Development tools:

pip install -e ".[dev]"

Quickstart

from medunify import Imaging, Signals, Records, AI

img = Imaging.load("scan.dcm")
series = Imaging.load_series("dicom_folder/")
meta = Imaging.metadata(img)

ecg = Signals.load("100", source="wfdb", pn_dir="mitdb")

patient = Records.parse_fhir_dict({...})

result = AI.run_inference(
    image=img,
    task="segmentation",
    model_name="basic_unet",
)

Important: AI features require the [ai] extra. Built-in basic_unet is an untrained research stub for smoke tests — supply your own weights and models for real experiments.

Module overview

Architecture

• src/medunify/adapters/ — thin wrappers around third-party libraries (imported as submodules such as medunify.adapters.pydicom_adapter to avoid eager loading unrelated stacks).
• src/medunify/*/ — domain models and facades (Imaging, Signals, Records, AI).
• AI is lazy-imported from medunify so environments without PyTorch/MONAI can still use imaging/signals/FHIR workflows.
• tests/ — pytest coverage for public APIs and adapters (with mocks where needed).

Roadmap

• Additional biosignal sources (EDF, proprietary readers behind adapters).
• Stronger FHIR resource coverage and validation pipelines.
• Pluggable preprocessing for AI (spacing-aware resampling, clip/normalize presets).
• Optional plugins for deployment-specific configuration.

Disclaimer

MedUnify is provided for research, education, and integration purposes. It must not be used for autonomous clinical diagnosis or any regulated use without appropriate validation, human oversight, and compliance with local laws and institutional policies.

License

MIT — see LICENSE.