Scaling Medical Image Segmentation on the Cloud

March 4, 2025

Michael VanDenburgh, Forrest Li, Zach Mullen and Andinet Enquobahrie

A Cloud-Based Segmentation-as-a-Service Application Built with Powerful Open Source Libraries

In the rapidly evolving field of medical imaging, segmentation plays a critical role in identifying and delineating regions of interest (ROIs) from complex medical scans. High-quality segmentation can unlock significant insights in oncology, cardiology, or neurology. However, integrating advanced segmentation capabilities into existing pipelines can be challenging. This is where VAMAS (Visualize, Analyze, Manage, medical images As a Service), our cloud-based Segmentation-as-a-Service application, steps in. The application uses VolView, Resonant, and MONAI—powerful open source platforms used in medical imaging product development.

What is VAMAS?

VAMAS is an end-to-end application deployed on the cloud that allows users to perform segmentation seamlessly using pre-trained models from the MONAI Model Zoo or their own custom models. Unlike traditional methods where users must manage their own infrastructure, VAMAS simplifies this process by providing pre-configured cloud environments for model inference, user management, and session tracking.

Why VAMAS?

While it is certainly possible for organizations to configure, deploy, and manage their own segmentation pipelines, this process requires significant time, expertise, and infrastructure investment. Setting up a scalable and secure environment involves provisioning cloud resources, ensuring compatibility with deep learning frameworks, managing dependencies, and implementing user access controls—all of which can be complex and resource-intensive. VAMAS eliminates these challenges by providing a fully managed, cloud-based solution that streamlines the entire process. With VAMAS, users can focus on deriving insights from their segmentation results rather than dealing with the operational complexities of deployment, making advanced medical image analysis more accessible, efficient, and scalable.

Key Features of VAMAS

1) Pre-Configured Cloud Environments Using MONAI
The MONAI framework offers a robust suite of pre-trained models for medical image analysis. However, setting up the necessary environment for deploying and using these models requires substantial effort, from infrastructure setup to ensuring compatibility with model dependencies. VAMAS handles all of this using a suite of integrated cloud infrastructure. Whether you’re using a model from the MONAI Model Zoo or bringing your own pre-trained model, VAMAS provides the infrastructure and environment ready to go—no configuration required.

2) User Management and Inference Sessions with Provenance
Managing access to an inference service is a crucial aspect of any enterprise-grade application. VAMAS makes this easy with a built-in user access and management system. Users can create accounts, upload their data, and run inference jobs in a secure, isolated environment. Provenance is another key feature: inference sessions are logged and traceable, enabling users to track the entire history of their segmentation tasks, ensuring data integrity and reproducibility.

3) Extensibility: Adding New Models with Ease
VAMAS is designed with extensibility in mind. Leveraging the MONAI bundle format, adding a new model to the application is straightforward. Users can train their own models, package them as MONAI bundles, and add them to VAMAS with minimal effort. Once added, the new model appears as a selectable option in the application’s drop-down menu, making it easy to switch between models. This extensibility ensures that VAMAS can grow alongside evolving research and development needs.

4) Advanced Visualization with VolView
VAMAS integrates with VolView, a cutting-edge DICOM visualization tool, to offer real-time visualization of both original medical images and segmentation results. This integration allows users to interact with their data, compare inference results with the raw imaging data, and derive insights efficiently. The ability to visualize segmentation overlays directly on DICOM images brings a level of transparency and ease of use that is essential for any segmentation-as-a-service application.

5) Scalability and Asynchronous Processing with RabbitMQ and Celery
At its core, VAMAS is built to scale. We understand that in a production environment, multiple users might need to run segmentation tasks concurrently, and a single user might need to perform multiple segmentations simultaneously. To meet this demand, VAMAS employs RabbitMQ for message queuing and Celery for task distribution. This infrastructure allows segmentation tasks to be processed asynchronously, ensuring that the application remains responsive and performant, no matter the workload. Users are free to queue up their tasks without fear of slowing down the system or blocking other operations.

Segmentation-as-a-Service: A Proof of Concept

VAMAS is more than just a tool—it’s a proof of concept for companies interested in integrating segmentation into their web-based applications. It leverages the flexibility of the MONAI bundle format and the scalability of cloud to offer a powerful, cloud-hosted segmentation service. By providing an easily extensible and scalable platform, VAMAS empowers organizations to focus on their core business while incorporating cutting-edge segmentation technology into their workflows.

Request a Demo

We’re excited to share VAMAS with the world. See firsthand how VAMAS can help accelerate your medical image analysis projects and streamline your segmentation workflows.

We invite you to watch our short demo video!