Project Clara

2020: AI models for diagnostics and treatment of COVID-19

On May 14, 2020 the NVIDIA company declared expansion of opportunities of the platform for problems of health care of NVIDIA Clara to help medical community better to diagnose and treat COVID-19.

According to the developer, the platform integrates AI and the accelerated calculations, helping researchers in the field of medicine, to suppliers of technological solutions and hospitals to cope with a pandemic by fast methods:

• Record speed of sequencing of a genome — software of NVIDIA Clara Parabricks for genomics, available according to the free 90-day license for researchers of COVID-19, shows the high speed of work: the analysis of the complete sequence of DNA of human genome takes less than 20 minutes.
• AI models for recognition of diseases — available for May, 2020, AI models were developed together with National Institutes of health care for studying effects of COVID-19 on KT of pictures of a breast and creation of tools for the best understanding, measurement and recognition of infections.
• NVIDIA Clara Guardian to small hospitals — the provided NVIDIA Clara Guardian system applies technologies of the smart video analysis and automatic speech recognition that "smart" hospital could execute monitoring of the key vital indicators of patients at the minimum employee involvement.

"The pandemic of COVID-19 accelerated interaction of the technology, research and medical industries for the purpose of development of the new computing solutions allowing to understand quicker how this disease extends and is how dangerous. Never before need for use of the best AI technologies and the accelerated calculations to all aspects of health care was so sharp. Results of use of the latest technologies will be notable far outside a pandemic", noted Kimberly Powell, the vice president of the Healthcare direction in NVIDIA

NVIDIA Clara Parabricks works for GPU NVIDIA A100. NVIDIA also provided the GPU accelerated pipelines of sequencing of RNA which issue results less than in two hours that gives researchers important information on susceptibility of the patient to a disease, its progressing and reactions to treatment, emphasized in NVIDIA.

"In our work every minute is expensive. NVIDIA Clara Parabricks is applied by researchers in the Center of rare children's diseases to define the most effective methods of treatment for each patient and to provide results during the day", noted Glen Otero, the vice president for scientific calculations of the Research institute of transmitting genomics

That users could get advantages of acceleration of NVIDIA Clara Parabricks in a cloud, NVIDIA contributes also to development of a corporate cloud platform of DNAnexus.

To provide specialists with additional tools in fight against COVID-19, NVIDIA provided a number of AI models which will help them to define and to study sick people using these KT-scans of a thorax. According to NVIDIA for May, 2020 these models are already available in Clara Imaging release in NGC library.

In the models created by a team of researchers of NVIDIA together with doctors and data analysis specialists in the National center of researches of health (NIH) data from regions with the high level of a disease of COVID-19, including China, Italy, Japan and the USA were applied. These AI models were created using NVIDIA Clara framework for Apparaty_meditsinskoy_vizualization_(mirovoy_rynok)|medical visualization]]. NVIDIA Clara contains instruments of training and implementation, characteristic of this area, that allowed NVIDIA and NIH to create these models less than in three weeks, claim in NVIDIA.

Creation of reliable AI models is one of global priorities, consider in NVIDIA, however data exchange often rests against the existing interstate restrictions. As of May, 2020 the international initiative of studying COVID-19 headed by NVIDIA and the Massachusetts Hospital Brigham is developed for application of COVID-19 AI models to x-rays. It will allow to adapt work methods locally, providing security of personal data of patients.

According to the statement of the developer, the framework of NVIDIA Clara Guardian opens access to a crucial ecosystem of solutions AI for security of the population and monitoring of patients by replacement of the used sensors by smart. Automatic temperature measurement of a body, detection of wearing protective masks, determination of observance of a distance and remote monitoring of patients belongs to important scenarios of use.

Partners on ecosystems apply the trained models and transfer of training to creation and implementation of the AI-applications integrating processing of visual, speech and language data. The NVIDIA EGX Edge AI platform platform gives the chance to develop and manage safely the park of devices in a hospital environment.

According to NVIDIA for May, 2020 the solutions based on NVIDIA Clara Guardian apply already dozens of partners in more than 50 hospitals and 10,000 chambers around the world.

2018

Announcement of Clara AI

On March 19, 2019 Nvidia provided the new Clara AI tool kit intended for training and deployment of algorithms of artificial intelligence in the field of medical visualization.

According to the developer, Clara AI includes 13 AI algorithms and software tools for classification and segmentation of images. Besides, its function of creation of auxiliary summaries is directed to acceleration of creation of the structured data sets and allows to create summaries to images in a few minutes. Tools also support training transfer, allowing to adapt the existing models to local indicators. Using transfer of training of AI can be adapted to the local database of patients without need to move this confidential information or to share it.

Nvidia provided the Clara AI tool kit for training and deployment of algorithms of AI in the field of medical visualization

These functions are included in a set of software development of Clara Train SDK. Besides, Clara Deploy SDK providing a development environment and deployments of AI for medical visualization was developed. According to representatives of Nvidia, for simplification of integration of the AI models into the existing workflow in Deploy SDK such standards as DICOM are used.

By March, 2019 Clara AI is actively used by the German center of a research of cancer. It put these tools into program open source for operation with medical images (MITK). In particular, the Center integrated Clara Train SDK into a plug-in MITK Workbench Segmentation, having provided 2D - and 3D - tools for editing and segmentation of polygons. In Clinical center of National Institute health care USA the researchers and scientists of Nvidia used Clara AI for development of a method of generalization of domains for the purpose of selection of a prostate from surrounding fabric on MPT- pictures. When testing the localized model reached productivity of the radiologist and exceeded other algorithms which studied and estimated on data from the same area.

Since April, 2019 Clara AI will become available to all hospitals, research institutes and the centers of medical visualization.^[1]

Clara SDK exit in a public access

At the end of November, 2018 Nvidia released a set of software development of Clara in a public access - a set of computing algorithms with acceleration of graphic processing, examples of applications, etc. Besides, the company provided the Transfer Learning Toolkit and AI Assisted Annotation SDK tool kit — two AI tools intended for processing of medical images.

The set for software development of Clara provides to developers tools for graphical analysis (Vulkan and Optix), calculations (CUDA, cuFFT and cuBlas) and AI libraries (CT Recon, Volume Segmentation, Lung Detection, Render Server); examples of applications for processing and rendering of images and the application for optimization of workflows at KT, MPT and ultrasonography. In the future he will also use Kubernetes, a storage system open source of Google to automate deployment and management of applications in the built-in, local and cloud environments. A number of partners of Nvidia, including ImFusion, Aidence, Arterys, Visage Imaging, Nuance, InferVision, Imagia, Trtle Medical and Kheiron, tested Clara within the pilot program.

Announcement of Project Clara

At the end of March, 2018 the producer of graphic solutions Nvidia announced the computing Project Clara platform intended for medical visualization.

The supercomputer provided at the Nvidia GTC 2018 conference allows to start a set of computing tools at the same time. In Clara the virtual graphic processors Nvidia are used to provide simultaneous access to many users.

Clara is capable to execute calculations for any tool — KT, MPT, ultrasonography, X-ray or mammography. The balancer of load of Kubernetes of GPU allows to scale effectively calculations on demand.

New technologies cardinally change health care —   doctor Greg Zaharchuk, the founder of Subtle Medical, the radiologist and the associate professor at department of radiology in Stanford says. — The idea of Nvidia about the virtualized medical supercomputer are new chapter which will cardinally change our opportunities in the field of medicine.

Medical visualization helps to define diseases at an early stage. Thanks to improvement of technologies of diagnostics to begin to work with many diseases today it is possible at the earliest stages. Now in development of this area also parallel computings which are developed by Nvidia make a contribution.

The company cooperates with a large number of the companies and startups in the field of health care and also with numerous research medical institutions. Their AI-applications, such as AutoMap and V-Net, make a powerful contribution to development of radiology.

The original ultrasonography picture of the left ventricle of heart (at the left) and the image with V-Net processing (on the right); in an algorithm V-Net it is applied 3D - segmentation to automatic measurement of the volume of blood passing through heart

For modern applications of medical visualization new levels of calculations are necessary, to scalability and availability. Project Clara is the computing Nvidia platform designed to make a revolution in the field, says Nvidia.^[2]

Notes