NVIDIA MONAI helps promote the implementation of research and development results and empowers innovative AI applications in the medical industry

With the continuous development of technology, artificial intelligence continues to penetrate into scenarios and is widely used in all walks of life. In the medical industry, the use of artificial intelligence technology to assist doctors in reading medical images can greatly improve efficiency and reduce doctors' work intensity and patients' waiting time.

In order to allow artificial intelligence to better serve the medical industry, NVIDIA has launched two key components, MONAI and Clara Holoscan. David Niewolny, director of medical business development at NVIDIA, said that medical imaging is one of the most important tools in the healthcare industry, accounting for more than 90% of healthcare data. Therefore, the use of artificial intelligence in medical imaging systems for healthcare is a very important application scenario. According to reports, many healthcare industries are currently rapidly adopting artificial intelligence technology, and NVIDIA research data is as high as 75%.

At this media communication meeting, David Niewolny focused on sharing MONAI technology and its implementation in major hospitals.

Officially launched in 2019, MONAI is an open source healthcare-specific artificial intelligence framework for developing and deploying models at scale in artificial intelligence applications. With MONAI, developers can easily build and deploy AI applications, create models that can be used for clinical integration, and more easily interpret medical test results to gain a deeper understanding of patient conditions.

According to David Niewolny, MONAI is designed for radiology, pathology and surgical data and aims to accelerate the clinical transformation of artificial intelligence, especially in the field of medical imaging. Therefore, MONAI is called the Pytorch of healthcare. David Niewolny said that the AI ​​life cycle comes with pre-trained models, AI-assisted labeling tools, and state-of-the-art training technologies (such as federated learning and self-supervised learning).

To make it easier for MONAI to integrate models into clinical workflows, NVIDIA also provides the MONAI Application Package (MAP), whose specifications were developed by the MONAI Deploy working group, which consists of more than a dozen companies A team of experts from medical imaging institutions with the goal of supporting AI application developers as well as clinical and infrastructure platforms that run AI applications.

For developers, MAP can help researchers easily package and test models in clinical environments, thereby accelerating the evolution of AI models. This allows them to collect real-world feedback to refine and improve the AI. In addition, MAP can also simplify the deployment process. If developers use the MONAI Deploy application development kit to package an application, hospitals can easily run the application locally or in the cloud. Finally, the MAP specification also integrates medical IT standards, such as the medical imaging interoperability standard DICOM.

For cloud service providers, support for MAP (designed using cloud native technology) can help researchers and enterprises using MONAI Deploy run AI applications on their own platforms through container or native application integration. .

Because MONAI standardizes application development, packaging, and deployment in healthcare IT infrastructure, it is widely adopted in the R&D community, with over 650,000 downloads, over 450 GitHub projects, and published papers. 160 articles and won 11 Kaggle competitions.

At the communication meeting, David Niewolny also introduced in detail the implementation case of MONAI in the medical industry through the cooperation with Cincinnati Children's Hospital Medical Center. According to reports, during heart transplant surgery, since the human heart can only survive for about 4 hours, every minute is very important. An important decision point is matching the donor, with medical imaging data and body segmentation used to measure the size of the potential donor's heart. Because this process is error-prone and time-consuming, it takes more than 20 minutes to complete. To that end, a Cincinnati Children's Hospital research team developed a deep learning model that automates this critical step, estimating total myocardial volume in just seconds, greatly improving the chances of a potential match. Pediatric patients in need of heart or lung transplants often suffer unnecessarily high mortality rates and spend long periods of time waiting, even when there are large numbers of unused donors, David Niewolny said. Cincinnati Children's Hospital Medical Center uses MONAI to scale a deep learning total heart volume model that saves many children's lives.

In addition to Cincinnati Children's Hospital, many well-known medical institutions are also applying MONAI to different applications. For example, the British National Health Service Trust Fund has deployed MONAI-based AI deployment engine platform - AIDE (AI Deployment Engine) in four hospitals, committed to providing AI disease detection tools for professional medical staff.

NVIDIA Startup Acceleration Program member Qure.ai uses MAP to package solutions for deployment, developing medical imaging AI models for use cases such as lung cancer, brain trauma, and tuberculosis. Companies that are members of NVIDIA Chicago's Startup Acceleration Program build 3D virtual representations of patients' tumors and use MAP for precision medicine AI applications that help predict how patients will respond to specific treatments. UCSF is developing MAP for several AI models, including hip fracture detection, liver and brain tumor segmentation, knee joint and breast cancer classification, among other applications.

According to David Niewolny, in addition to a large number of medical industry cases, many cloud vendors are also using MONAI Deploy researchers and enterprises to run AI applications on their own platforms through container or native application integration.

For example, the MAP interface has been integrated into the HealthLake imaging service, allowing clinicians to view, process and segment medical images in real time. Google Cloud's medical imaging suite makes medical imaging data more accessible, more interoperable and more useful. The suite has integrated MONAI into its platform, enabling clinicians to deploy AI-assisted annotation tools to help automate manual and repetitive medical image labeling tasks.

The Microsoft Azure-powered Nuance Precision Imaging Network combines MONAI and the Nuance Precision Imaging Network. Oracle and NVIDIA recently announced a collaboration to bring accelerated computing solutions for the healthcare industry, including MONAIDeploy, to Oracle Cloud Infrastructure. Starting today, developers can use NVIDIA containers on Oracle CloudMarketplace to build MAP through MONAI Deploy.

As David Niewolny mentioned, currently, most AI models are in the research and development stage, mainly due to the lack of a proprietary standard. MONAI Deploy will help promote the implementation of research and development results and achieve more influential clinical AI. ​

The above is the detailed content of NVIDIA MONAI helps promote the implementation of research and development results and empowers innovative AI applications in the medical industry. For more information, please follow other related articles on the PHP Chinese website!