Watch the Low-Code/No-Code Summit on-demand sessions to learn how to successfully innovate and achieve efficiencies by upskilling and scaling citizen developers. Watch now.
It’s been six years since Geoffrey Hinton said “We need to stop educating radiologists now,” stressed that “it’s perfectly clear that deep learning will outperform radiologists in five years.” Instead, the future of medical imaging appears to rest firmly in the hands of radiologists – who have used artificial intelligence (AI) as a collaborative tool to drive medical imaging, one of the most essential areas of healthcare that is used by the whole patient. used travel.
What is evolving, however, is significant open-source efforts to bring medical imaging AI models into clinical settings at scale, and to ensure that the medical imaging data that trains those AI models is robust, diverse, and available to everyone. to be.
Integration of AI models into clinical workflows
To address the first goal, Nvidia announced today at the annual meeting of the Radiology Society of North America (RSNA) that MONAI, an open-source AI framework for medical imaging accelerated by Nvidia, makes it easier to create AI models integrate into clinical workflows with MONAI Application Packages (MAPs), delivered through MONAI Deploy.
Nvidia and King’s College London introduced MONAI in April 2020 to simplify AI workflows for medical imaging. This helps convert raw image data into interactive digital twins to enhance analysis or diagnostics or guide surgical instruments. The development and adoption of the platform now has more than 600,000 downloads, half of them in the past six months.
Medical imaging leaders, including UCSF, Cincinnati Children’s Hospital and startup Qure AI, are using MONAI Deploy to turn research breakthroughs into clinical impact, Nvidia said in a press release. In addition, all major cloud providers, including Amazon, Google, Microsoft, and Oracle, support MAPs, enabling researchers and companies using MONAI Deploy to run AI applications on their platform, either by using containers or with native app integration.
“MONAI has really established itself in the research and development community as the PyTorch of healthcare,” David Niewolny, director of healthcare business development at Nvidia, said in a press conference ahead of the announcements. “It was built specifically for radiology, but now expanding into pathology and digital surgery, really addressing the entire AI lifecycle, bridging the gap between this research community and implementation.”
For example, Cincinnati Children’s Hospital is creating a MAP for an AI model that automates segmentation of total heart volume based on CT images, helping pediatric heart transplant patients project funded by the National Institutes of Health. “It speeds up decision-making time for pediatric transplant patients,” he said. “It really has the potential to save the lives of a number of children.”
Scaling AI and medical imaging to a wider audience
The integration of MONAI by all hyperscalers in the cloud will allow this research to go beyond one hospital to a much wider audience, Niewolny added. For example, the MAP connector is integrated with Amazon HealthLake Imaging, which allows clinicians to view, process and segment medical images in real time. And Google Cloud medical imaging suite has integrated MONAI into its platform to enable clinicians to deploy AI-assisted annotation tools that help automate the highly manual and repetitive task of labeling medical images.
Additionally, “Oracle Cloud infrastructure has some pretty big things planned,” he added, especially in light of Oracle’s recent acquisition of Cernerone of the largest medical record companies in the world.
“It’s great to see this gap between the model developers and the people actually doing the clinical implementation closing,” he said. “That’s really a boost for AI innovation across the entire medical imaging ecosystem.”
Development of various medical image datasets
Of course, even with better hardware and infrastructure, advances in medical imaging, AI, and data science require the right medical imaging datasets to ensure algorithms are not biased. To that end, a Harvard Medical School AI research lab has just announced a new initiative called MAIDAto develop and share diverse medical image datasets from around the world.
According to lab leader Pranav Rajpurkar, assistant professor at Harvard Medical School, the problem they decided to solve is that medical imaging data is rarely shared between institutions due to concerns about data security, vendor lock-in and data infrastructure costs.
In addition, the existing data lacks a diverse view. Algorithms for clinical applications are disproportionately trained in a few hospitals, with little to no representation at the national or global level. Populations that are not sufficiently represented in the training cohort are likely to get biased results. For example, dark skin is underrepresented in commonly used dermatology datasets.
“There is an urgent need to democratize medical imaging datasets and ensure diversity in the data used for data science and AI development,” Rajpurkar told VentureBeat. “The current data that is in the public domain is not only a small sliver, but a highly selective sliver that is not diverse and has no international representation.”
About 40 hospitals are already involved in MAIDA’s dataset curation, said Rajpurkar, which is starting with datasets of chest X-rays, which are the most common imaging study worldwide. The lab is also developing AI models for other common radiologist tasks, including placing an endotracheal tube and detecting pneumonia in the emergency room.
“We expect MAIDA to be a key ingredient for medical AI and data science, enabling tools to work on more diverse populations than they currently are,” he said.
VentureBeat’s mission is to become a digital city plaza where tech decision makers can learn about transformative business technology and execute transactions. Discover our Briefings.
