NVIDIA has announced the launch of MONAI, a new cloud platform aimed at accelerating the creation and deployment of AI solutions for medical imaging. MONAI provides a set of robust application programming interfaces (APIs) to help developers more easily integrate automated workflows into existing medical imaging software.
At the core of MONAI is the VISTA-3D (Vision Imaging Segmentation and Annotation) foundation model for interactive AI annotation. Purpose-built by NVIDIA researchers for medical images, VISTA-3D can generate high-quality segmentation masks to identify areas of interest in 3D CT scans. The model employs a technique called continuous learning, which means its performance improves over time based on user feedback and new scan data.
Beyond interactive labeling, MONAI features additional APIs to simplify other key steps of the AI development process. This includes model training APIs leveraging NVIDIA's open-source MONAI library of pretrained models fine-tuned for common medical imaging tasks. MONAI also provides Auto3DSeg, an automated tool for hyperparameter tuning and model selection when creating custom 3D image segmentation models.
According to NVIDIA, these APIs will help medical software providers, researchers, and other healthcare stakeholders build the specialized "AI factories" needed to produce reliable AI solutions at scale. The company announced several partners, including medical data platform Flywheel and annotation company RedBrick AI, that plan to integrate MONAI into their existing tools and services.
By providing turnkey access to domain-optimized models and streamlined model building workflows, NVIDIA hopes the new MONAI platform will lower barriers for creating performant medical imaging AI. This has the potential to accelerate innovation of AI-assisted diagnostic, research, and clinical trial analysis tools across healthcare.
The MONAI cloud APIs are now available in early access on NVIDIA's DGX Cloud AI infrastructure.