Synthetic Data Generation for Healthcare Innovation

Generative AI is revolutionizing the medical imaging field, enabling researchers and developers to leverage digital twins of real as well as fully synthetic patients and medical imaging devices to overcome training data limitations, validate models, and explore clinical hypotheses.

Combining advanced gen AI algorithms with the computational power of GPUs enables large-scale synthetic image creation as well as real-time digital twin explorations. Synthetic image models can be made to cover a wide range of patient demographics, to insert specific disease biomarkers into existing patient data to overcome shortage of data indicative of rare diseases, and to simulate alternative imaging modalities for training or to simplify intraoperative image registration tasks. Data acquisition and annotation challenges and costs are reduced, and ethical data diversity concerns can be addressed.

Project MONAI contains multiple generative AI capabilities for synthetic image generation for 2D and 3D imaging modalities. The data needs and privacy concerns of nearly any and every medical imaging application are met when those gen AI capabilities are combined with MONAI Label for AI-assisted data annotation and with MONAI’s federated learning methods for model training without requiring data centralization.

MAISI is NVIDIA’s foundational model for medical image generation. Built using Project MONAI, it can generate synthetic 3D x-ray computed tomography (CT) images and corresponding segmentation masks with up to 127 anatomical classes (including bones, organs, and tumors), while achieving the landmark voxel dimensions of up to 512 × 512 × 768 and spacing ranging from 0.5mm³ to 5.0mm³. The overarching goal of MAISI is to revolutionize the field of medical imaging by providing a reliable and efficient way to generate high-quality synthetic images that can be used for various research and clinical applications. By overcoming the challenges of data scarcity and privacy concerns, MAISI aims to enhance the accessibility and usability of medical imaging data.

With NVIDIA’s accelerated computing and MONAI platform for medical imaging AI, researchers and application developers can use MAISI to:

• Reliably and efficiently generate high-quality synthetic images that can be used for various research and clinical applications—from training medical students to training and validating surgical robots
• Overcome challenges associated with data scarcity and privacy concerns
• Explore new methods and fine-tune models for medical image gen AI that address new imaging modalities, new patient demographics, and rare diseases
• Interactively evaluate the baseline capabilities of MAISI using NVAIE resources and the portable, scalable MAISI NIM™: https://build.nvidia.com/nvidia/maisi