As the radiology community continues to learn and capitalize on the latest advances in the use of artificial intelligence (AI), a respected panel of leading experts in the field announced on the final day of the 108th Congress of Radiology It was fitting to announce an important and comprehensive session. North America (RSNA) Scientific and Annual Meetings.
One of the hottest topics during the conference combined with a line-up of Catherine P. Andrior, M.D., Linda Moy, M.D., Dania Day, M.D., and Walter F. Wiggins, M.D., generated strong interest and participation. . An overview of basic terminology, leveling of important tasks, notes on clinical applications, and predictions for future applications were well explained and well received.
With the presence of these people and this session “Back to Basics: What Should Radiology Departments Know About AI?” up, reflecting the rapid growth of AI software solutions and clinical applications in today’s market.
ITNThe editorial team of has shared a summary of the information presented in multiple installments. This first segment contains an excerpt from Dr. Day and Wiggins. Daye, along with Wiggins and other authors, recently published a paper in the RSNA journal. Radiology, she reviewed it. In the next second segment, ITN We publish educational and insightful session segments by Dr. Andrior and Moi.
Setting the stage for deeper understanding
before the session ITN I spoke with moderator and panelist Day, an interventional radiologist at Massachusetts General Hospital and a faculty member at Harvard Medical School and the MGH/HST Martinos Biomedical Imaging Center.
Daye shared the following insight: ITN About the goal of the session.
“I think there has been an explosion of AI applications in the market over the last few years. There are still a great many radiologists who are just starting out in this area and we are very interested in learning more about the basics of implementation and learning in practice.Therefore, the purpose of this session is to It’s about getting people to understand the basics, learn a little more about data science, implementation, governance, see what’s out there, and then use it to give people the infrastructure to start working. This is certainly the beginning of a very long journey, but we want to help people get started by dipping their feet in the water, so to speak.”
Addressing AI Governance: Who Decides How?
Daye’s participation in the panel focused on the RSNA journal in introducing the current focus of the panel discussion on the use of AI in radiology. Radiology Paper entitled “Clinical Artificial Intelligence Implementation in Radiology: Who Decides How?” (published online August 2, 2022, in print December 2022) She was its lead author. Day joined the manuscript with Wiggins and other leading radiologists, including 2023 RSNA gold medalist James A. Brink, MD.
Focusing on key points from the paper, Daye emphasized that there are four key requirements for successful implementation of a clinical imaging AI program. Cross-platform and cross-domain integration. Clinical translation and delivery. And leadership that supports innovation. She detailed her needs as follows:
•Data access: We recognize that complete anonymization is difficult, and we build an environment with data security to protect patient information.
•Data science: Provide clinical researchers with new data science training and tools to harness this data.
•Cross modality: Consolidate access to EMR, medical imaging, genomics, and physiological monitoring data in one place.
•delivery: Bring new discoveries to clinical care through context-integrated clinical decision support platforms through collaborative processes across the healthcare enterprise.
Daye highlighted key points from the paper by stating: “Successful clinical implementation of artificial intelligence is facilitated by establishing robust organizational structures that properly oversee the implementation, maintenance, and monitoring of algorithms.” As it evolves, governance structures will oversee the implementation, maintenance, and oversight of clinical AI algorithms to improve quality, manage resources, and ensure patient safety.”
Emphasizing the value of this article, Daye explained that she and her co-authors established a governance roadmap and answered four key questions. Who decides which tools to implement? What factors should be considered when evaluating application implementation? How should the application be implemented in the clinical setting? To ensure quality patient care and implement improvement goals, devising a flexible governance structure that can quickly adapt to a changing environment is essential, he said. pointed out.
Daye further emphasized that AI oversight in medical imaging needs to consider the multidisciplinary stakeholders who use radiology services. Inevitably, all panelists spoke about the need for AI governance and shared their top priorities for successful AI adoption.
The Committee noted that the elements of good governance should have the following characteristics: consensus-oriented, participatory, accountable, transparent, responsive, equitable and inclusive, effective and efficient; Obey the rule of law.
Review of new developments and future impacts
Continuing AI Committee is Walter F. Wiggins, M.D., board-certified neuroradiologist, assistant professor at Duke Health University, and clinical director of the Duke Center for AI Radiology (DAIR) at Duke University School of Medicine. A Strategic Advisor to Qure.ai, Mr. Wiggins focuses on the use of advanced imaging and image analysis technologies in diagnostic imaging of the brain, head, neck and spine, with a particular focus on the clinical implementation of artificial intelligence technologies. I’m here.
medical imaging.
In a session segment, Wiggins discussed the topic “Current Applications of AI in Radiology: What’s Out There?” where are we going ” He reported on the National Institutes of Health’s (NIH) data-sharing policies and updates that apply to NIH grants received after January 25, 2023. For those advancing the application of AI in radiology: , said data management and sharing plans will be needed, Wiggins said, also identifying policy goals such as enabling verification of research findings, providing access to high-value datasets, and providing access to high-value datasets in the future. Wiggins also said there will likely be a continued increase in model/tool development and FDA clearance in the future. He suggested that oversight of deployment tools would be a necessary element of the market strategy going forward, and said analysis and decisions about who should bear the burden are ongoing.
Addressing current applications and trends, Wiggins noted that the Vendor Landscape at RSNA 2022 shows a clear and consistent increase in both the number of companies in the space and the number of FDA-approved AI tools. Did.
He noted that the American College of Radiology (ACR) recently released an AI Central Dashboard. This can be broken down by subspecialty, by modality, and allows users to track trends in approval sequences. The dashboard also provides a catalog of all FDA-approved software as medical devices (SaMDs), which now number 201, a sharp increase from 2017 and likely to continue. expected.
Wiggins identified other recent developments, reporting that chest radiographs continue to receive significant attention, and noted the proliferation of text analysis tools, namely natural language processing (NLP). He highlighted the growing awareness and concern about bias (identification, mitigation and management). Citing research published in the journal The Lancet in early 2022 by collaborators from many institutions around the world, he said more emphasis would be placed on training models to de-bias them in the future. suggested.
“If you want to see how things are going in the future, look at how the FDA and its guidance are coming out. It’s about being there,” Wiggins said. “And we also need to look at the NIH and what funding they have and what they say they have to do once they have the funding,” he added. Wiggins continued, “The FDA is moving toward a full product lifecycle concept centered around machine learning models, potentially leading to a shift to continuous learning models and the need for vendors to update their models.” It eliminates it and makes it easier to update the model.” Each time it has to go through the FDA clearance process… I don’t know exactly how this will play out as it is still pretty preliminary. I think what this really reveals, even before we bring things up to the FDA, is an increasing focus on clinical validation and post-implementation monitoring. “
He also noted the NIH’s data sharing policy, which went into effect in January 2023, saying, “What I think this will bring to AI research is the ability to validate and reuse high-value research results. It’s about providing access to datasets.” Data for future research as they are goals set by the NIH. “
Finally, Wiggins said the focus will be on the continued growth of AI, continued significant research towards clinical implementation, and reducing bias with AI tools as a necessary component of vendors’ market strategies in the coming years. predicted. Returning to common concerns going forward, he said: “The real question here is who will bear the burden? The vendor? Will it be just us radiologists, or will we collectively oversee these tools?”
Related artificial intelligence content:
AiMed Global Summit 2023 in San Diego Focuses on Transforming Healthcare, One Connection at a Time
AiMed Global Summit lineup announced
AiMed 2023: Changing healthcare, one connection at a time
Click here for full coverage of the AiMed23 conference
Advantages and disadvantages of using ChatGPT in clinical radiology: an open discussion
ChatGPT passes Radiology Board exam
RamSoft Powers Medical Imaging Patient App Blume with ChatGPT – Presented at HIMSS 2023
JNM Explores Potential Applications of ChatGPT in Nuclear Medicine and Molecular Imaging
New study suggests AI imaging with DALL-E2 is a promising future in radiology
