France's strong emphasis on the ethical and humanistic aspects of artificial intelligence (AI) and its practical impact on patient care have attracted praise and support from leading U.S. radiologist Dr. Paul Chang from the University of Chicago.
It's refreshing that the medical profession in France is asking, "What are the ethics of using AI correctly and what is the consequence to our patient care?" he told delegates on 2 June at a daylong seminar held in Lyon on artificial intelligence and radiology, titled Artificial Intelligence: The radiologist's dream or nightmare?
"Those of us in the states have been more driven; I would hate to say, by self-preservation. The perspective tends to be: What would AI do to me as a practicing radiologist? Is it going to help me? ... Make me more efficient, reduce variability, and improve quality? Or is it a threat to me? And I think a lot of the discourse ... tends to center around the effect of AI in the practice of radiology," Chang noted in a video interview.
Dr. Paul Chang.There are marked differences between the views of the French radiology community and that of the U.S. on artificial intelligence, he added.
Ethical and human factors aren't ignored in the U.S., but radiologists tend to take a more pragmatic, and perhaps a more self-centered approach, with respect to AI practices, one that is combined with entrepreneurial and commercial considerations, as well as doctors' impatience in procuring the technology. The U.S. professional and industrial community concentrates on U.S. Food and Drug Administration (FDA) approval, and the key question asked is: How can the technology be commercialized and used?
This is in contrast to the French attitude, which in addition to being pragmatic, places ethics and patients at the top of the agenda.
"Many of the discussions today were dealing with principles ... In fact, the congress was started with a philosopher, which I find incredibly refreshing," noted Chang, pointing to the key question at the morning session: What are the philosophical and ethical considerations with respect to how this technology is going to impact our patients?
He doubts this approach will slow down the adoption of the technology in France, although he conceded that other factors may act as a brake on the adoption of AI.
Growing sophistication of systems
The Lyon seminar investigated the impact of any disruptive technology, not just AI as it stands now, and pointed to necessary vigilance as today's basic AI systems grow more sophisticated.
"Deep learning systems are actually quite primitive. We're nowhere near singularity; we're nowhere near what many folks are concerned about. But we may eventually get there," Chang noted. "The progress will increase significantly, I suspect, and I don't think it's too early to talk about these [ethical and philosophical] considerations."
Furthermore, he considered that establishing an ethical framework showed that France wanted to remain in the driver's seat with regard to AI, and suggested it was crucial for all radiologists to get behind the wheel.
"If not, we'll be in the trunk of the car, being driven by other motivations. I think being driven by humanistic, ethical, philosophical orientations is a very important safety net," Chang said.
The technology will not replace specialists, but they need to understand and prioritize how best to use it to improve patient care. His personal belief is that AI will allow radiologists to minimize all the "busy" work.
"I would make the argument that we are more like machines now, before AI. I think appropriate use of artificial intelligence will make us less machine-like so we can spend more time with our patients," he added.
Chang pointed to how before PACS, radiologists spent time with clinical colleagues, and with patients. However inefficient this was, this practice was more human and more ethical. Imaging now has entered an adolescent period where because the technology is so immature, radiologists are spending too much time being like machines and less time talking to patients.
"The real goal of this is not to replace us, the goal is to get rid of what de-humanizes us as physicians and allow us more time to be more human," he said.
During his presentation, he noted how adoption of AI in radiology was a relatively slow process. The imaging community tends to buy into new technology hype early, but takes a long time to consume it in practice.
He described how modern imaging trends such as the significant increase in dataset size and complexity, the transition to functional, physiological, and molecular imaging, the transition from qualitative to quantitative imaging, as well as the shift toward actionable phenotypic characterization (radiogenomics) and from isolated image interpretation to multidisciplinary synthesis presented challenges for system development. Advanced IT could serve as a critical enabler -- or a barrier, he noted.
Chang also warned attendees at the Lyon meeting about complacency. The belief that imaging informatics has been solved is a misconception, he noted, and he stressed that existing IT offerings are still relatively immature. This IT immaturity only allows radiologists to demonstrate commodity-level service, when they require significantly more capable and agile IT solutions to provide measurable value to patient care in a significantly more complex, demanding, and competitive environment, according to Chang.
Finally, advanced informatics tools such as AI, big data, and deep learning are neither the "horrible threat" nor "promised savior," he explained. These tools hopefully will be used appropriately to help improve both efficiency and quality to enhance the value proposition to patients. He urged delegates to stay calm because deep learning was nothing new: Radiology has always redefined itself when incorporating new, potentially disruptive technology, usually to the benefit of both clinicians and patients.
Chang noted it was probably too early to "pick a winner." Instead, departments should concentrate on preparing their IT infrastructures and strive for "deep integration" with workflow, for the ultimate goal of data-driven, optimized, human-machine cybernetic workflow orchestration.
The reality check from Chang came ahead of the joint announcement made on behalf of France's professional radiology council (G4) by the seminar's organizer, the national union of private radiologists (Fédération Nationale des Médecins radiologues, FNMR) and its partner, the national radiology society (Société Française de Radiologie, SFR) about its collaborative AI imaging ecosystem project for developing a complete AI diagnostic imaging tool.
