Understanding the Transformation Potential
Artificial Intelligence (AI) is being used all around us, from personalizing our shopping experiences to making our cars safer and our digital devices smarter. It’s easy to get carried away by the promise of AI.
Machine learning (ML), a sub-domain of AI, improves predictions, and better predictions lead to better decisions. So how will AI, including faster and more accurate predictions, transform radiology? This question was the impetus for the inaugural RP AI summit. Held in Silicon Valley Nov. 1-2, the meeting brought together 60 attendees from RP and other large radiology practices, health systems, the tech industry, and academia to engage in a dialogue on the opportunities and challenges of AI in healthcare.
We kicked things off Thursday evening with a fireside chat with NEA Venture Partner, Mo Makhzoumi and RP board member, Jeff Immelt. Jeff pointed to three forces that will continue to push practices towards AI: “One thing is clear: imaging volume is going up, reimbursement is going down, and risk is shifting to the provider.” Tools that improve prediction allow for quicker and more effective decision-making – precisely what is needed in this evolving landscape.
Greg Papadopoulos, an NEA Venture Partner with more than 30 years of experience in the tech industry including as Chief Technology Officer of Sun Microsystems, instilled a sense of excitement among attendees with real-life examples of AI applied outside of healthcare, demonstrating that AI will permeate our lives.
Friday’s program included plenty of examples of AI being applied in radiology, albeit in relatively early stages. We heard from a range of product developers from both industry and academia on ML applications:
- Dr. Elliot Fishman, a professor of radiology at Johns Hopkins Medicine, is applying machine learning for earlier detection of pancreatic cancer
- GE is building an ML algorithm development platform to help streamline the development process, from image annotation to programming
- Aidoc is using ML to achieve more timely detection of abnormalities on CT scans in acute care and shared some exciting preliminary data on outcome improvement
- Enlitic is working on a lung cancer screening solution providing nodule detection and characterization with clinical decision support and an AI second reader solution that would evaluate both the image set and report to look for potential discrepancies. Kevin Lyman, CEO of Enlitic, also presented their approach to creating a radiologist labeled image truth set to be used for future algorithm development
- Subtle Medical is looking to use ML to create high-quality images while shortening image acquisition times and reducing radiation dose
- DeepHealth is applying deep learning to create the next generation of mammography screening support tools
All of these cases come with their own sets of challenges, the common themes being data availability and diversity, regulatory hurdles, clinical adoption, economics, and workflow integration.
In Silicon Valley, where it’s easy to get swept away by the wave of enthusiasm, we left the Summit with a sense of pragmatic optimism, built on three key insights for our practice:
- As we continue to build our IT infrastructure, AI will be an integral part of that process.
- We must develop AI use cases that resonate with our radiologists and add value. The true value of AI will only be captured if radiologists play a leading role in its development.
- Scale will matter more than ever, both in terms of access to data to train and improve algorithms and to support the substantial investments that will be necessary to develop, integrate and operationalize AI within radiology practices and health systems.
As we continue to learn about AI and prepare our practice for this new wave of innovation, one thing is certain, AI will be enhancing not replacing radiologists. But to quote Stanford radiologist and computer science expert Dr. Langlotz, “radiologists who use AI will replace radiologists who don’t.”