Today, GPUs are found in almost all imaging modalities, including CT, MRI, x-ray, and ultrasound - bringing compute capabilities to the edge devices. With the boom of deep learning research in medical imaging, more efficient and improved approaches are being developed to enable AI-assisted workflows.

To develop these AI capable applications, the data needs to be made AI-ready. NVIDIA Clara’s AI-Assisted Annotation does so by providing APIs and a toolkit to bring AI-assisted annotation capabilities to any medical viewer. Post annotation, diverse and high quality annotated data is required to create powerful AI models. To enable this, NVIDIA Clara Train SDK includes privacy-preserving Federated Learning, a collaborative learning paradigm that enables research hospitals and institutions to collaborate and develop more robust AI algorithms without sharing private data.

Running on a variety of hardware solutions, the Clara application framework leverages the edge stack
to deliver all the necessary tools for healthcare application developers.

Clara Federated Learning

NVIDIA’s latest release of Clara Train SDK, which features Federated Learning, makes it possible for Academic Hospitals and centers of AI excellence to collaboratively build robust AI algorithms without compromising data privacy. From peer-to-peer to cyclic and server-client, Federated Learning can be implemented in different distributed architectures.

NVIDIA Clara’s implementation is based on a server-client approach, which means that a centralized server acts as a facilitator for the overall federated training with the participation of various clients. The client only shares partial model weights back to the server for aggregation - data stays secure at the client location. The configurable MMAR (Medical Model ARchive) feature of Clara Train SDK makes it possible for developers to bring their own models and components to perform Federated Learning and have control over whether local training is run on a single GPU or multiple GPUs.

Learn More

Optimized Tensorflow solution that provides AI-Assisted annotation to any medical imaging viewer as well as pre-trained AI Models and development tools to accelerate the creation of AI algorithms for medical imaging workflows.

Clara Train SDK includes AI-Assisted Annotation APIs and Annotation server that can be seamlessly integrated into any medical viewer making them AI capable. The training framework includes decentralized learning techniques like federated learning and transfer learning. The SDK also makes available model applications packaged as MMARS (Medical Model ARchive) available to users, providing an intuitive config based environment for data scientists and researchers to get kick-started with AI development.

AI-Assisted Annotation

  • AI Annotation Server now includes NVIDIA TensorRT inference server as its inference back-end providing a more scalable inference experience to the users
  • APIs for 3D polygon editing
  • APIs for continuous training

Horovod - Automatic Mixed Precision - Smart Cache - 8 GPUs
Click on graph for more results

Training Framework

  • Federated learning is a collaborative learning technique that allows for distributed training with multiple clients. With Clara Train v2.0 we bring privacy-preserving Federated Learning that enables researcher to collaborate and build AI Models without sharing private data.
  • Automatic Mixed Precision(AMP) allows researchers to train with half precision and maintain network accuracy. AMP can reduce memory usage and provide significant speed ups to training process.
  • Deterministic training on GPUs is now available in the SDK and is crucial to guarantee reproducibility for iterative experimentation.
  • The option to use Smart Cache in new task specific ImagePipelines allows for faster and more efficient training by saving intermediate results and skipping repeated operations.
  • New loss functions and models have been added.
  • Transforms have been rewritten to be more purpose based with ShapeFormat and MedicalImage taken into account to simplify configuration and improve clarity.
  • You can use MMARs to set up training configurations with json, but you can directly use python code with the Clara Train API for greater customization including bringing your own components.

An extensible reference development framework that facilitates turning AI models into AI-powered clinical workflows with built-in support for DICOM communication and the ability to interface with existing hospital infrastructures.

Clara Deploy SDK provides a container-based development & deployment framework for building AI accelerated medical imaging workflows, it uses Kubernetes under the hood and enables developers and data scientists to define a multi-staged container-based pipeline. It can be thought of as a reference platform for building AI-based Imaging workflows.

Latest Features
  • New Clara Pipeline Orchestrator allows developers to get higher performance in a multi-container; multi-pipeline environment
  • Clara Deploy includes gRPC based platform APIs that allow the flexibility to extend the platform with newer utilities and allows developers to bring their own services to the platform
  • The latest package includes logging and performance tools for efficient monitoring of pipelines
  • Clara Deploy includes features for Faster I/O that allows for efficient memory handling between containers in the same pipeline

To enable a whole new world of devices, Clara AGX SDK enables medical devices that need the ability to perform real-time AI and advanced image, video, and signal processing.

NVIDIA Clara AGX™ Developer Kit

The NVIDIA Clara AGX™ developer kit is an embedded AI computer and software development framework for medical devices that need the ability to perform real-time AI and advanced image, video and signal processing. NVIDIA Clara AGX securely manages and orchestrates AI application deployments to fleets of medical devices or edge nodes.

Coming Soon

Other Developer Tools

Clara SDKs are built on CUDA-X, a collection of libraries, tools, and technologies that deliver dramatically higher performance than alternatives across multiple application domains—from artificial intelligence to high-performance computing. Commonly used tools and APIs are provided here for your convenience.


CUDNN TensorRT TensorRT Inference Server


Image and Signal Processing


OptiX Video Codec Deepstream SDK Optical Flow

Additional Resources

deep learning Get Started With Hands-On Training

The NVIDIA Deep Learning Institute (DLI) offers hands-on training for developers, data scientists, and researchers in AI and accelerated computing. Start your hands-on training in AI for Game Development with self-paced courses in Computer Vision, CUDA/C++, and CUDA Python. Plus, check out two-hour electives on Deep Learning for Digital Content Creation and Game Development.

Disclaimer: Clara SDKs and samples are for developmental purposes only and cannot be used directly for clinical procedures.