NVIDIA CLARA HEALTH PLATFORM
NVIDIA Clara is a collection of developer tools specifically for healthcare, built on the NVIDIA Compute Platform, that accelerate data collection, analysis and integration.
Across the healthcare industry, the number of software-defined solutions has grown exponentially, driving efficiencies through automation and generating large volumes of digital data. Artificial intelligence has made it possible to integrate and analyse this data in new ways to gain deeper insights. NVIDIA Clara aims to give medical imaging and genomics developers access to technological advances in hardware and software to accelerate the future of medicine.
From automating workflows to improving processing speed and image quality, medical imaging developers around the world are discovering numerous ways to use AI to help doctors detect and diagnose disease. With the Clara platform, they are harnessing AI to transform healthcare workflows.
CLARA MEDICAL IMAGING
Today, GPUs are found in almost all imaging modalities, including CT, MRI/MRI, X-ray and ultrasound, bringing more computing power to the edge devices. Deep learning research in medical imaging is also booming, with more efficient and improved approaches being developed to enable AI-powered workflows. Today, most of this AI research is conducted in isolation and with limited datasets, which can lead to oversimplified models. Even when a fully validated model is available, it is challenging to deploy the algorithm in a local environment. With the latest release of Clara AI for Medical Imaging, data scientists, researchers and software developers now have the necessary tools, APIs and development frameworks to train and deploy AI workflows.
DEEP LEARNING LIBRARIES
IMAGE & SIGNAL PROCESSING
- More cone-beam CT research is done with CUDA than with any other.
- Accelerator technology CUDA outperforms other accelerator technologies by an order of magnitude or more.
- The latest algorithmic developments being performed are all CUDA-accelerated.
CLARA GENOMICS is designed to address the growing scale and complexity of genomics sequencing and analysis with accelerated and intelligent computing.
The Clara Genomics Analysis SDK is now available under open source terms to provide free and open access to developers; please click below to access the release via GitHub:
GPU-accelerated implementation of the Racon consensus module for de novo genome assembly.
This open source version adds the cudaAligner module for accelerated alignment, including Ukkonen's algorithm and Myers' bit algorithm
A high-level workflow from sample preparation to final analysis that begins with the isolation of an organism's DNA. This isolated sample is then loaded onto a sequencing device where embedded GPUs are used to accelerate primary analysis and enable next-generation base calling with Deep Neural Networks (DNNs).
Secondary analysis or sequence analysis uses NVIDIA GPU computing for the Genome Analysis Toolkit (GATK), DNN-based variant calling and de novo genome assembly.
Our first release of the Clara Genomics SDK will focus on de novo assembly of long read sequencing from Oxford Nanopore and Pacific Biosciences, reducing analysis time from days to hours. The first version includes GPU-accelerated libraries and GPU-accelerated applications.
CLARA GENOMICS TECHNOLOGY STACK
- CUDA Mapper - CUDA-based library sharing algorithms for overlapping sequencing reads.
- CUDA Aligner - CUDA-accelerated library of sequencing read alignment algorithms used for genome assembly applications such as Racon and for variant calling.
- CUDA POA - CUDA accelerated partial sequence alignment library used for genome assembly polishing with applications such as Racon.
- Racon Polisher - An extension of the open source Racon Consensus genome assembly module that uses cudaPoa for accelerated partial order alignment.
- Racon Aligner and Mapper - Will be available in upcoming versions.
CLARA TRAINING SDK
- APIs to add AI-powered annotation to any medical viewer with new features such as auto-annotation and interactive annotation modes, Annotation Server that provides pre-trained models to the client application, and client APIs hosted on Github that allow seamless integration with your medical viewer application.These capabilities are already built into the latest MITK Workbench plugin.
- The SDK provides the ability to use techniques such as transfer learning to customise or train deep learning models from scratch, allowing Data Scientists to bring in their own model architectures and run Workflows This is made possible by a unified foundation of the Medical Model Archive (MMAR).
- MMAR (Medical Model Archive) provides a model development environment; it defines a standard structure for storing and organising all artefacts created during the model development lifecycle.
- MMAR contains NVIDIA pre-trained models based on AH-Net, DenseNet, ResNet, Dextr3D, packaged as complete 2D/3D model applications for organ-based segmentation, classification and annotation.
MAIN FEATURES OF CLARA TRAINING SDK 1.1
- Bring your own models, transformations, readers, losses and metrics.
- Configurable framework to facilitate deep learning tasks from medical images.
- Medical Model Archive (MMAR) with deep learning models and artefacts.
- Model adaptation and retraining that can be easily used in heterogeneous environments with multiple GPUs.
- Model export API for easier deployment of applications for TensorRT-based inference.
with NVIDIA to solve these critical medical imaging problems."
SDK FOR CLARA INSERT
- Data ingestion interface for communication with the hospital's PACs system.
- Core services to orchestrate and manage resources for workflow deployment and development.
- Reference AI applications that can be used as-is with custom data or modified with custom AI algorithms.
- Finally, the Clara Deploy framework also includes visualisation capabilities to monitor progress and display final results.