Terrill Dickey 2024-07-28 07:58
NVIDIA has introduced the VISTA-3D NIM microservice, which enhances CT scan analysis with advanced organ and disease segmentation capabilities.
More than 300 million computed tomography (CT) scans are performed worldwide each year, with 85 million in the United States alone. Radiologists are constantly looking for ways to speed up their workflows and produce accurate reports. To address this need, NVIDIA Research developed a new foundational model, VISTA-3D, which is integrated with the NVIDIA NIM, an optimized microservice designed for scalable deployment, according to an NVIDIA technical blog.
VISTA – 3D model
The VISTA-3D (Versatile Imaging Segmentation and Annotation) model is trained on over 12,000 volumes, covering 127 different human anatomical structures and a variety of pathologies, including lung nodules, liver tumors, bone lesions, etc. It provides accurate out-of-the-box segmentation and state-of-the-art zero-shot interactive segmentation, making it a versatile tool for medical image processing.
This model includes three core workflows:
Segment Everything: Enables comprehensive body exploration and helps understand complex diseases affecting multiple organs. Segment Using Classes: Provides a detailed view based on specific classes, essential for targeted disease analysis. Segment Point Prompts: User-driven selections improve segmentation accuracy and accelerate the creation of accurate ground truth data.
VISTA-3D’s architecture includes an encoder layer followed by two parallel decoder layers, one for automatic segmentation and one for point prompting. This structure ensures high accuracy and adaptability across different anatomical regions.
VISTA-3D NIM Microservices
The VISTA-3D NIM microservice, hosted in the NVIDIA API catalog, allows you to test the functionality with sample data, and can segment over 100 organs or specific classes of interest, providing views in axial, coronal, or sagittal planes.
Using NIM Microservices
Users can run VISTA-3D on their own data by signing up for a personal key from NVIDIA, which provides 1,000 free credits to try out the NIM microservices. Detailed instructions on generating an API key and running models are provided, along with example code in various programming languages.
If you run VISTA-3D with your own data, you will need to set up an FTP server to serve your medical images. This approach accommodates medical images that are typically too large to send directly in the API payload.
Running the NIM microservice locally
To run the NIM microservices locally, users must apply for NVIDIA NIM access. Once approved, they will be provided with Docker containers to run the VISTA-3D NIM microservices on their preferred hardware. As a prerequisite, Docker, Docker Compose, and NVIDIA drivers must be installed.
To help you get started quickly, we provide a sample Docker Compose file and instructions for setting up an NGINX server to serve the images.
Conclusion
NVIDIA’s VISTA-3D foundational model represents a major advancement in medical imaging, capable of accurately segmenting over 100 organs and a variety of diseases in CT scans. NVIDIA NIM microservices simplify the deployment and use of this powerful model, improving workflow and accuracy for radiologists.
Interested parties can apply for access to the VISTA-3D NIM microservices to leverage its capabilities on their own hardware and streamline their medical imaging processes.
Image credit: Shutterstock
Source link
