Artificial intelligence (AI) in radiology is more than just a buzzword. Countless developers are creating tools to optimize imaging analytics in healthcare and each advancement has unimaginable potential. So how is AI being deployed in radiology right now? And perhaps more importantly, where should AI in radiology be headed? I sat down to talk with one of Segmed’s medical advisors, Matthew Lungren, MD MPH, to get his thoughts on these very questions.
Lungren is a trained radiologist and an associate professor at Stanford University. He is also the co-director of Stanford’s Center for Artificial Intelligence for Medicine and Imaging. The center brings together teams of clinicians, computer scientists, biostatisticians, engineers, and legal experts to develop and support AI methods that advance patient health. As one of Segmed’s medical advisors, Lungren says he helps the team “come up with ways to ethically, responsibly make patient data available so that we can see advancements actually make it into the clinic and have the impact we know they can have.”
A lot of people have grand visions for what AI can do for medicine. Daydreams of computers detecting cancer come to mind. So it may come as a surprise to learn that AI is already being employed in radiology. Technology like fastMRI are finding ways to leverage machine intelligence to make medical imaging faster without losing image quality. Companies that make medical imaging scanners, including Siemens, Philips, and GE, are using this to reinvent the way they create images.
Lungren points out that this has huge implications for patient satisfaction and hospital economics. CTs that can be completed with a fraction of the traditional radiation dose protect patients from excess radiation exposure. MRIs that can be completed in ten minutes allow patients and clinicians bonus time to do other tasks. They also allow hospitals to scan more patients each day, which helps offset the cost of an improved scanner.
Complicated algorithms that can detect cancer or perform risk stratification are the futuristic ideas that often overshadow important behind-the-scenes AI applications. Lungren thinks that development companies should create algorithms to optimize triage and workflow operations before tackling more lofty challenges like diagnostics.
“It’s probably the unsexy things that need to be fixed first,” he laughed. Right now, if a patient gets an x-ray that isn’t adequate for diagnosis, the radiologist has to track down the patient and rescan them. Lungren thinks that unoptimized tasks like this, ones that involve evaluating images immediately, would be very quick and straightforward to improve with an algorithm. Addressing workflow sore spots would have potential to lighten the workload of medical professionals and make healthcare run more seamlessly.
Radiologists excel at interpreting imaging scans and communicating patient care to other clinicians. However, there are quantitative aspects of their job (like making precise measurements, performing perfect calculations, and conducting quantitative comparisons) that are very difficult for humans to do consistently. Humans are just “subject to so many different biases that limit our capacity to do these things,” Lungren highlights. Opportunities to standardize some of these quantitative tasks could revolutionize radiology and how we care for patients.
Lungren thinks AI models that allow radiologists to do what humans do best (namly contextualizing their medical training and emotionally connecting with patients) while enhancing their jobs with computers doing what they do best (performing quantitative analysis on data) is the ideal road forward.
When looking outside the Bay Area perspective that Lungren, Segmed, and many other tech companies find themselves in, there may be a different path forward for medical AI. Around the world there is a “huge spectrum of what a radiologist does and whether there is even a radiologist available,” Lungren points out.
There are opportunities to leverage existing technology that wasn’t initially developed for clinical practice (think smart phones) that can empower non-radiologist medical professionals to more effectively care for patients. One major application for this could be digitizing scan interpretation processes instead of having clinicians examine a hard copy of an image (remember when x-rays had to be held up to a light?).
AI that can provide assistance in interpreting scans has potential to increase the sheer number of people impacted around the world. Our blog post, The Transformative Power of Medical AI in Rural Parts of Developing Nations, discusses this potential further.
All of these visions for the future of radiological AI are exciting, but how do developers ensure that the solutions they’re building are prioritizing the patients and clinicians they’re intended to help? Lungren emphasizes that informed consent needs to be very transparent; the patient needs to understand that their data could potentially be used for the research and development of new technologies. “Understanding that these are not just data points, these are people and they need to have their privacy preserved” is of the utmost importance in this field, stressed Lungren.
When acquiring data from parts of the world with large economic disparities, Lungren wonders if there is an opportunity to see the patient benefit from sharing their data with development companies. “Are there opportunities to see the patients reimbursed in some fashion?” he muses. Questions like this are important checks against whether companies are putting patients and clinicians above their own profit goals.
One of the reasons Lungren wanted to be a medical advisor for Segmed was because he thought the founding team emphasizes the ethics of medical data acquisition. “Ultimately, I think having the leadership that is focused on the clinical endgame as opposed to just getting something out to market is really important,” he professed.
Total transparency is one of Segmed’s central beliefs that is incorporated throughout our company. We run our own security checks to make sure that the data we receive can never be traced back to the patient, physician, or institution it came from. We want to empower developers to work on the problems that are most needed clinically, and patient protection is at the forefront of that.
At the end of the day, healthcare is about the people involved. What are we doing to make sure that people around the world are getting access to quality, responsible healthcare? That question is at Segmed’s core and we encourage everyone in this field to keep it in mind.
.svg)
