Accelerate Healthcare Applications with NVIDIA Clara

 What is NVIDIA Clara?

NVIDIA Clara is a graphics and simulation solution that supports healthcare applications and frameworks for AI-powered imaging and genomics. Clara can be broken up into multiple parts: healthcare applications, application frameworks and the Edge stack. Each of these parts can also be broken up in more sub-parts. From full-stack GPU-accelerated libraries to secure and scalable solutions, Clara allows data scientists and researchers to perform optimized research in medical imaging and genomics.


Healthcare Applications

 Clara provides support for a variety of healthcare applications such as early detection, clinical workflows, real time analysis and others. This allows healthcare professionals like neurologists to speed up AI development for quicker model enhancement. With advancements in AI and software solutions, applications such as annotations of medical imaging can be significantly sped up via NVIDIA’s AI-Assisted Annotation (AIAA).

This graph compares the speed of annotation manually vs using their AIAA server. To achieve this speedup, data scientists and doctors can use tools such as The Medical Imaging Interaction Toolkit (MITK) or 3D Slicer, both open-source visualization and medical imaging computing platforms.


Clara Application Framework – Medical Imaging

 The Clara Application Framework can be split up into two categories: Clara for medical imaging and Clara for genomics. The medical imaging framework allows training frameworks such as federated learning and transferred learning. Federated learning allows a quality centralized model from training data distributed over a large number of clients. The data is kept private and the weights are modified instead of the data. Transferred learning is the process of taking a model trained on a large data-set and tuning it with new data to support your task.

To integrate the Clara Application Framework, there are multiple SDKs that are used. The first SDK is Clara Train SDK, including the AIAA server and training. Pre-trained models are loaded into into the AIAA where the annotation takes placed. Unlabeled data can then become labeled data, where it is then loaded into the training section of Clara Train. Here, pre-trained models go through either federated learning or transferred learning where the models are then optimized for your workload.

The next part of the Clara Application Framework is the Clara Deploy SDK. This SDK allows for connection to the Picture Archiving and Communication System or PACS. Various scanners and machines send data to the PACS servers, where the Clara Application can run the models that were trained.

The third part is the Clara AGX SDK, where embedded AI takes place. AGXs are edge devices that allow for real-time support, various sensor processing, and other edge capabilities.


Clara Application Framework – Genomics

Genomics is the process of studying genomes or organisms and processing them into genetic mapping and DNA sequencing. Clara for Genomics allows for GPU-accelerated genomics analysis. Clara provides three AI enabled platforms for Genomics: Parabricks for GPU Accelerated Genome Analysis Toolkit (GATK), Long Read De Novo Assembly and Single Cell Sequencing. Since analyzing genomics in very data involved and complex process, Parabricks allows for a scalable and increased performace GATK workflows. NVIDIA Parabricks comes fully loaded with readily customizable pipelines to begin analysis pipelines.

Here is an example of a workflow that may take place with DNA. The first step is to isolate the DNA, then analyze it by sequence analysis and further tertiary analysis. The steps below in green utilize NVIDIA CUDA.

Connecting to SDKs through CUDA allows for GPU-accelerated applications to bring an acceleration factor up to just under 11x more than when using CPU powered servers. This acceleration can bring a pipeline from taking hours to completion to just minutes.

Thanks to NVIDIA Clara’s healthcare application frameworks and SDKs, data scientists, medical professionals and engineers are able to quickly and effectively perform medical imaging and genomics research at a faster rate. These tools that are laid out allow for GPU acceleration to create faster and stronger models for research and development.


Ben Siegel is a Data & AI Technical Support Engineer at Groupware Technology.