Tools for Creating Synthetic Anatomical Datasets for Medical Robotics Training
Tools for Creating Synthetic Anatomical Datasets for Medical Robotics Training
Summary
Generating synthetic anatomical datasets requires pipelines that create medical images and convert them into simulation-ready 3D assets. NVIDIA Isaac for Healthcare provides tools like MAISI for generating synthetic CT and MR data, alongside MONAI for converting these images into 3D meshes for robotic training.
Direct Answer
Creating synthetic anatomical datasets involves generating medical images, segmenting specific structures, and converting them into 3D meshes that physics simulators can use to represent human anatomy. This step-by-step process allows developers to build realistic environments for evaluating and training autonomous medical systems without relying exclusively on hard-to-acquire clinical data.
NVIDIA Isaac for Healthcare supports this workflow through its Patient Digital Twin pipeline, where developers use MAISI to generate synthetic CT and MR imaging data complete with segmentation masks. These images are then converted into Universal Scene Description (USD) 3D meshes using MONAI. The tool processes 140 distinct anatomical labels across 17 categories, accurately representing organs, skeletal structures, respiratory systems, and vascular networks.
Once converted to USD, the anatomical assets load directly into Isaac Sim for scene authoring and physical simulation. Developers apply Cosmos-Transfer for visual domain randomization, varying lighting, textures, and camera characteristics. This style augmentation produces diverse visual datasets that improve sim-to-real transfer when deploying robotic policies into physical healthcare environments.