6 pieces of advice from an AI early adopter
Dr. Francis Schlueter shares some of his experiences as an early adopter of AI in radiology and advises what it will take to advance AI adoption across healthcare.
The use of advanced analytics and AI in healthcare continues to evolve rapidly. A recent survey found one-third of healthcare organizations already use AI in medical imaging, with another one-third of respondents saying they would within the next two years.1
TriHealth, a leading healthcare provider in the Cincinnati area, is an early adopter of AI. TriHealth partnered with IBM Watson Health for this journey. Dr. Francis Schlueter, Radiology Systems Chief and President of Tristate IMG (Imaging Medical Group), has been one of the clinical leaders pioneering this work at TriHealth.
I recently talked with Dr. Schlueter about his experiences. He offered great insights about applying AI in clinical practice, including these:
1. The next 1-2 years are critical for AI adoption.
The pandemic created a sense of urgency around digital transformation. One survey across industries found 59% of respondents indicated the pandemic accelerated digital transformation and 66% completed initiatives that had previously encountered resistance.2 This period of disruption may accelerate AI adoption, though the journey will remain challenging.
“The next couple of years will be dynamic for AI in radiology, and adoption will continue to be challenging” Dr. Schlueter predicted.
Brilliant and dedicated teams can often imagine numerous possibilities for the use of AI in medicine, but Dr. Schlueter advises a practical approach in the near term. “Some teams focus on the grander picture of AI in healthcare without focusing on just getting the process started,” Dr. Schlueter said. “Think about the problems we can solve in the next one or two years instead of thinking too conceptually about the future.”
2. Strive for integration.
Throughout our conversation, Dr. Schlueter noted integration with workflow is important to encourage clinicians’ use of AI. He observed that both he and many fellow radiologists have hit the limit when it comes to adding new technologies.
“I’m already working with four monitors. There’s literally no more room for another tool,” he said. “AI apps need to be integrated into our daily lives. I believe that’s where we are headed…unless apps are integrated in a seamless way, no one will use them.”
3. Don’t slow the radiologist down.
“Radiologists are impatient by nature,” Dr. Schlueter said. “If you offer a solution that slows us down, we won’t use it.”
We talked through an example of when radiologists are assessing individual applications, such as lung nodule detection. Radiologists must initially look at a series of images, go back and compare to previous images, and then make a final assessment. In the short term, we need to find a way for AI to help make that interaction as efficient as possible. Ultimately, this technology will improve accuracy, make comparisons more seamless and quantify how disease has changed over time.
“We’re good at what we do, so AI needs to convince radiologists that it will enhance our workflow,” Dr. Schlueter said.
4. Beware alarm fatigue.
“Many studies are imperfect, which leads to one of my biggest concerns: alarm fatigue,” Dr. Schlueter said. “If apps alert radiologists dozens of times, and they have to click through multiple notifications, they will eventually just tune it out.”
This concern is at the crux of a much larger issue for those developing AI algorithms: should the algorithm tune for sensitivity and potentially find too much, or should it tune for specificity and potentially miss something? These are familiar areas of concern – not just in the application of AI or other technologies – but as clinicians strive to deliver high-quality, evidence-based standards of care.
5. AI automation is the only way forward.
In my opinion, the near future for AI in radiology is not as much in detection as it is in workflow efficiency. The field is currently experiencing a transition like when Picture Archiving and Communication Systems (PACS) revolutionized standard film interpretation. Radiologists can now review hundreds of images at a time, instead of only one at a time.
“When I was in training, I read 12-16 CT scans a day, and that was a good day. Now, I’m doing that in an hour or so,” Dr. Schlueter said. “Ultimately, we’re going to need to be reading upwards of maybe 200 cross-sectional exams per day. The only way we’ll get there is with automation and AI support.”
I also believe AI will take us to that next level of efficiency. The technology will share information in a way that’s easier for radiologists to comprehend and extract relevant information.
6. Demonstrate that AI adds value.
“The application must do its job well. It will also need to present information to the radiologist in a simple and meaningful way,” Dr. Schlueter said. “It’s ultimately about clinicians seeing how AI can help them deliver quality, service, and value to our patients.”
For Dr. Schlueter, the value of AI is clear. Though he has seen some radiologists ready to give up because AI has had its challenges, he plans to continue to work toward integration of AI into his work.
“All radiologists need to be early adopters of this technology, especially for the sake of their patients,” he said. “Begin the journey and expect, like with many planned journeys, there will be some tangents along the way. However, in the long run, it will be worth it.”
Kudos to Dr. Schlueter for his work as an AI early adopter, and for taking important steps toward advancing the promise and possibility of AI in healthcare.
- “One-Third of Orgs Use Artificial Intelligence in Medical Imaging” by Jessica Kent, Health IT Analytics Jan. 28, 2020
- “COVID-19 and the future of business” IBM Institute for Business Value