IDC Releases 'Specialized Medical AI Platform' White Paper; NVIDIA Clara Platform Accelerates Healthcare AI Transformation

Dec 19, 2018 16:05 CST Updated 16:05

At the VB100 launch event for the 2018 Future Healthcare 100, International Data Corporation (IDC), a leading global provider of market intelligence, advisory services, and events for the information technology, telecommunications, and consumer technology markets, released the white paper titled “Specialized Medical AI Platforms Drive the Transition of Healthcare AI from Experimentation to Practice.” The application of artificial intelligence in healthcare faces triple challenges related to data, algorithms, and computational power. In the field of medical AI, GPU computing platforms provide unparalleled core computational support for healthcare AI applications.

From left to right: Yu Chang, Chief Architect for Healthcare and Life Sciences at NVIDIA China,Liu Nianning, Senior Marketing Director for NVIDIA China、

Xiao Hongliang, Senior Research Manager at International Data Corporation (IDC)

NVIDIA, a leading provider of AI computing platforms, has launched the CLARA open platform, designed to help the healthcare industry build and deploy breakthrough algorithms, thereby enabling the intelligence of medical devices and the automation of clinical workflows. CLARA provides an open, scalable, and remotely accessible general-purpose platform that transforms data stored in medical devices into dynamic medical imaging data, with data volume and dimensionality exceeding those of the raw imaging data originally acquired by the devices. CLARA can help improve image quality in medical equipment and reduce misdiagnosis rates associated with medical devices.

The NVIDIA CLARA AGX system is the core of the CLARA platform, providing developers with a rich set of development tools. By leveraging the software development kit (SDK) provided by the CLARA AGX system, developers can redesign and redevelop medical workflows. Furthermore, developers can deploy their applications across diverse computing environments, including embedded, on-premises, and cloud-based systems.

NVIDIA’s future roadmap for its CLARA platform aims to provide unified support services for all medical imaging. The goal is to deliver integrated diagnostic capabilities for individual medical diagnoses, facilitating interoperability among future medical imaging systems and across broader healthcare ecosystems.

The increasing maturity of artificial intelligence (AI) technologies has driven the development of the healthcare industry. As a core component in the advancement of medical AI, chips provide the computational power necessary for upgrading healthcare service systems. Graphics Processing Units (GPUs) are currently the primary platform for AI computing. For instance, in the field of medical imaging, which has seen the most prominent progress, the key to assisting image diagnosis lies in using deep learning to identify and extract feature points from images, based on model training with large volumes of imaging data. Employing GPUs as an acceleration solution can significantly improve the efficiency of image classification. Within the Convolutional Neural Network framework (Caffe), a single GPU can perform tens of millions of image computations within a day.

Taking Convolutional Neural Networks (CNNs) as an example, the massive number of neurons within neural networks are inherently highly parallel, a characteristic that aligns perfectly with the computational architecture of GPUs. Consequently, training models on GPUs is significantly faster than on CPUs. Deep learning involves extensive dense matrix operations, and utilizing GPUs for these computations can improve efficiency by a factor of 6 to 17. In addition to GPUs, Application-Specific Integrated Circuits (ASICs) and Field-Programmable Gate Arrays (FPGAs) are also representative acceleration solutions for deep learning. However, due to factors such as performance and cost, GPUs are more widely accepted in the field of medical artificial intelligence.

Building a high-performance computing platform is one of the fundamental elements for the success of artificial intelligence development. Since deep learning requires feeding massive amounts of data into training models, and these models necessitate large-scale computations to acquire intelligent capabilities, the computational power of an AI platform is a key factor in its success.

Currently, AI computing platforms are primarily based on GPUs for training and inference. Some AI systems also utilize chips such as CPUs, FPGAs, and Tensor Processing Units (TPUs). Major server manufacturers, including Inspur, Lenovo, Dell, and H3C, have developed servers designed for machine learning and running AI systems. NVIDIA has also developed the DGX supercomputer system specifically for AI applications.

NVIDIA’s latest medical imaging supercomputer—CLARA. It leverages deep learning technology to convert medical imaging data stored in three million medical imaging devices worldwide into “state-of-the-art” full-color animated images. This opens up possibilities for the reuse of large-scale datasets within the artificial intelligence and deep learning industries.

NVIDIA CLARA, as an open platform, assists the healthcare industry in building and deploying breakthrough algorithms to enable the intelligentization of medical devices and the automation of clinical workflows. CLARA provides an open, scalable, and remotely accessible general-purpose platform that transforms data stored in medical devices into dynamic medical imaging data, with data volume and dimensionality exceeding those of the raw imaging data acquired by the devices themselves. CLARA helps enhance imaging quality in medical equipment and reduce misdiagnosis rates associated with medical devices.

One feature in NVIDIA CLARA’s future roadmap is to provide unified support services for all medical imaging. The CLARA platform offers an opportunity to integrate diagnostic services that are typically performed separately. Although this capability is not yet fully available, it will be realized in future interactions among medical imaging systems and across various healthcare systems.