NVIDIA launches a cloud service for medical imaging AI

NVIDIA launched a cloud service for medical imaging AI to further streamline and accelerate the creation of ground-truth data and training of specialized AI models through fully managed, cloud-based application programming interfaces.

The new NVIDIA MONAI cloud APIs provide an expedited path for developers and platform providers to integrate AI into their medical imaging offerings using pre-trained foundation models and AI workflows for enterprises. The APIs are built on the open-source MONAI project founded by NVIDIA and King's College London.

Why does it matter?

Medical imaging is critical across healthcare, making up approximately 90% of healthcare data. It's used by radiologists and clinicians to do screening, diagnosis and intervention, by biopharma researchers to evaluate how clinical trial patients respond to new drugs and by medical device makers to provide real-time decision support.

The scale of work across each of these areas requires a medical imaging-specific AI factory — an enterprise-grade platform that delivers large-scale data management, creates ground-truth annotations, accelerates model development, and establishes seamless AI application deployment.

With NVIDIA MONAI cloud APIs, solution providers can more easily integrate AI into their medical imaging platforms — enabling them to provide supercharged tools for radiologists, researchers, and clinical trial teams to build domain-specialized AI factories. The APIs are available in early access through the NVIDIA DGX Cloud AI supercomputing service.

How does it work?

The NVIDIA MONAI cloud API is integrated into Flywheel, a leading medical imaging data and AI platform that supports end-to-end workflows for AI development.

Developers at medical image annotation companies, including RedBrick AI, and at machine learning operations (MLOps) platform providers, including Dataiku, are poised to integrate NVIDIA MONAI cloud APIs into their offerings.

These APIs feature interactive annotation powered by the VISTA-3D (Vision Imaging Segmentation and Annotation) foundation model, which is purpose-built for continuous learning.

Trained on a dataset of annotated images from 3D CT scans from more than 4,000 patients, spanning various diseases and parts of the body, VISTA-3D accelerates the creation of 3D segmentation masks for medical image analysis. With continuous learning, the AI model's annotation quality improves over time.

To further accelerate AI training, this release includes APIs that make it seamless to build custom models based on MONAI pre-trained models. NVIDIA MONAI cloud APIs also include Auto3DSeg, which automates hyperparameter tuning and AI model selection for a given 3D segmentation task, simplifying the model development process.

On the record

"NVIDIA MONAI cloud APIs lower the cost of building high-quality AI models for radiology, disease research and the evaluation of clinical trial data," said Dan Marcus, chief scientific officer at Flywheel. "With the addition of cloud APIs for interactive annotation and automated segmentation, customers of our medical imaging AI platform can accelerate AI model development to more quickly deliver innovative solutions."

"VISTA-3D allows our clients to rapidly build models across different modalities and conditions," said Shivam Sharma, CEO of RedBrick AI. "The foundation model is generalizable, making it easy to fine-tune for various clinical applications with accurate, reliable segmentation results."

"With NVIDIA MONAI cloud APIs, Dataiku users would be able to easily use Auto3DSeg, a low-code option to accelerate the development of state-of-the-art segmentation models, through Dataiku's web interface connected to an NVIDIA-hosted, GPU-accelerated service," said Kelci Miclaus, global head of AI health and life sciences solutions at Dataiku. "This democratizes AI in biomedical imaging by extending the power to create and apply AI-driven workflows to both data and domain experts."