Deep learning in healthcare is booming — but when it comes to production-ready pipelines for medical imaging, even the best researchers hit roadblocks. Enter MONAI: NVIDIA’s open-source framework built for healthcare AI.

And when you bring Valohai into the mix, you unlock the operational muscle to take your imaging models from prototype to production.

In this technical spotlight, we showcase how you can take a MONAI training pipeline and make it fully reproducible, trackable, and scalable using Valohai — from data preprocessing to evaluation.

👉 See it live on GitHub

Why MONAI?

NVIDIA MONAI (Medical Open Network for AI) is the go-to framework for medical imaging AI. With MONAI, researchers and teams can:

• Work with medical image formats like NIfTI and DICOM
• Leverage domain-specific pre-processing transforms and augmentations
• Train high-performance models tailored for segmentation, classification, and more
• Build with PyTorch, but for healthcare

It’s research-grade AI with production intentions — but like most frameworks, MONAI needs help when it comes to MLOps.

Enter Valohai

Valohai is the MLOps platform purpose-built for deep learning workflows in regulated and high-compliance environments.

• Track every dataset version, training run, and model artifact
• Reproduce any experiment, anywhere — from cloud to air-gapped hospital environments
• Deploy MONAI pipelines as reproducible workflows — not ad hoc scripts
• Automate everything, without compromising control

The Project: MONAI + Valohai in Action

This example project demonstrates how to:

1. Preprocess Medical Image Data

Convert and standardize raw imaging datasets with Valohai-managed pipelines. Ensure every image transform is tracked and repeatable.

2. Train a UNet Model for Segmentation

Launch reproducible GPU-backed training using MONAI’s components and PyTorch backend. Configuration is managed in Git, while execution is managed by Valohai.

3. Evaluate Model Accuracy

Run evaluation workflows post-training, with metrics logged and associated with each pipeline version.

4. Version, Track & Share

Every step is logged, every result reproducible. Need to re-run with different parameters? It’s one click away.

👉 Explore the full MONAI + Valohai example on GitHub

Why This Combo Matters

Let’s face it: most medical imaging AI workflows are stitched together with notebooks and tribal knowledge. That’s fine for prototypes — but for real-world impact, you need:

• Pipelines you can reproduce under audit
• Automation that respects compliance boundaries
• Version control for every data and model state
• Easy onboarding for new researchers and collaborators

Valohai gives MONAI users everything they need to go from research to deployment without changing tools — or compromising control.

Get Started Now

Want to accelerate your healthcare AI projects with confidence?

➡️ Check out the GitHub example and clone it for your own use

➡️ Sign up for Valohai and run your first MONAI job today

➡️ Explore the Valohai + NVIDIA Partnership

➡️ Or just contact us — we’ll walk you through it

Medical imaging AI at scale. MLOps that’s audit-ready. That’s the Valohai way.