Improved efficiency and coordination enable several immediate applications, and clinical potential also emerges.
Artificial intelligence in healthcare still seems to have some sort of futuristic, “this will happen someday” feel to it.
That is far from the truth in radiology. Remarkably, at the recent annual meeting of the Radiological Society of North America in Chicago, the idea that “the future is now” was at the core.
AI is already bringing benefits to healthcare organizations, but more benefits are expected, especially in tandem with other technological advances taking place in the field of imaging. In summary, the promise of AI is to make radiologists more efficient, reduce their workload, and enable them to leverage technology to deliver preventive care.
AI is more than just an add-on
Traditionally, radiologists have had to navigate between different technology stacks, covering a variety of imaging devices, picture archiving and communication systems (PACS), radiology information systems, and more.
Additionally, previously, AI for radiology was applied at different stack levels and didn’t always fit neatly into workflows. This year’s RSNA Show focused on designing systems that support radiologists by leveraging AI to increase accuracy and efficiency, enhance backstops, and support radiologists throughout the imaging cycle.
Shyam Sokha, chief operating officer and technology director for digital health at RadNet, says radiology is plagued with inefficiencies and inconsistencies that impact downstream quality. “There is an urgent need to solve clinical, financial and operational challenges,” he said. “The result is disconnected patient engagement, strained workforces, and inconsistent clinical outcomes.”
Two years ago, Los Angeles-based RadNet, a provider of outpatient diagnostic imaging services, announced the launch of DeepHealth, a product line designed to use AI to “improve efficiency and transform the role of radiology in healthcare.”
DeepHealth offers TechLive, an AI-powered remote scanning platform. This enables “cooperative multimodality image processing.” This makes it vendor-independent and allows radiologists to direct technicians to perform imaging studies remotely.
Sokka said the company’s use of AI “helps us build an integrated solution on a cloud-based structure. Worklists, reports, viewers, advanced visualization and analysis… can be synchronized to enable continuous improvement and build trust in AI.”
Fusion focused on efficiency
David Niewolny, director of healthcare and medical business development at NVIDIA, said that while radiologists were once reluctant to trust AI with the clinical aspects of their work, concerns about high workloads, workforce attrition, and burnout are now making them more willing to leverage the technology to improve efficiency.
While the company has been dedicated to healthcare for five years, Niewolny says NVIDIA is working with companies and organizations to go deeper in several areas. For example, we offer MONAI, an open source framework that enables deep learning in medical image processing.
As an example of the possibilities, Aidoc noted its partnership with NVIDIA MONAI in a deal announced at RSNA25. Through this initiative, Aidoc aims to “expand the scope of clinical AI and help health systems deploy imaging AI models and bring them into routine clinical use at scale.”
Aidoc’s efforts emerge as “the demand for image processing AI continues to grow,” the company notes. “Health systems are adopting more vendor solutions, and major hospitals and academic centers are increasingly developing proprietary models tailored to local populations and priorities. However, the biggest barrier remains the last mile: effectively deploying models into clinical workflows, rather than training them, and scaling them across facilities and modalities.” Aidoc’s approach, developed through a partnership with NVIDIA and Quibim, is “imaging AI in the clinical setting. The aim is to achieve “deployment and scalability”.
Achieving clinical results
Despite the potential reluctance to use AI for clinical purposes, there is some evidence emerging that AI can be transformative in certain use cases.
For example, a recently published study found that natural health This is based on data from DeepHealth’s Breast Suite application, which RadNet calls the largest real-world analysis of AI-powered breast cancer screening in the United States. The data, derived from mammograms from more than 579,000 women across more than 100 community-based imaging sites, demonstrated a 21% increase in breast cancer detection rates.
“Furthermore, this technology has been proven to elevate the performance of general radiologists to expert levels and expand access to quality breast care in areas where experienced readers are limited,” the company claims.
In another application of AI-supported imaging, RapidAI combines technologies to better diagnose aneurysms that cause strokes, using Lumina 3D technology to detect and visualize aneurysms in three-dimensional renderings. Karim Carti, CEO of the San Mateo, Calif.-based company, said advances in technology already allow for faster diagnosis, allowing critical treatment to begin in six hours, compared to 24 hours just a few years ago.
A ruptured aneurysm “can be very fatal, so treatment with clot-busting drugs is very important. We want to be proactive, monitor the aneurysm, and use advanced technology to know when to intervene,” Carti says.
The company hopes to further develop the technology to provide insight into different types of stroke and other conditions such as pulmonary embolism, where speed of response is critical.
“Integration into the workflow is very important,” says Karti. “Radiology needs to work seamlessly with PACS and not slow down reading speeds.” The second integration point is collaboration with clinicians. One of the things we’ve continued to expand on is improving the quality of reporting, reducing cognitive burden, and helping clinicians make decisions faster.
“Radiologists don’t have to spend a lot of energy on mundane things,” he concludes. “Currently we specialize in detection, but we are ultimately moving into the realm of prediction. The earlier we intervene, the better the results and the lower the cost.”
