AI is central to many large technology companies such as Facebook and Google, and may soon have a similar role in the medical imaging world, argued a group of radiologists in a new editorial published Aug. 31 in Clinical Imaging.
Four big names in radiology--Julie Sogani, MD, of Weill Cornell Medicine; Bibb Allen Jr., MD, of Grandview Medical Center in Birmingham, Alabama; Keith Dreyer, DO, PhD, with the ACR’s Commission on Informatics; and Geraldine McGinty, MD, MBA, part of the ACR’s Board of Chancellors--combined to define AI in radiology, entertain the idea of an AI imaging ‘ecosystem’ and describe how the field will need to overcome challenges to use the technology in clinical practice.
Below are three takeaways from the report:
1. AI can touch many aspects of radiology
Radiology, and healthcare as a whole, is laboring toward the transition from volume to value-based care and AI could help with this, and many other areas of inefficiency.
A radiologist with AI tools can bring more information to a patient interaction and enhance their role on a healthcare team. And aside from image interpretation and analysis (a monumental undertaking in itself), AI can help schedule imaging exams at appropriate times, lower radiation dose and improve patient safety.
“…AI algorithms can benefit every step of the imaging process,” the researchers wrote. “Integration of AI into clinical practice may not only improve radiologists' diagnostic performance but can also help our departments and institutions become more efficient – improving patient satisfaction and decreasing costs.”
2. Radiology must develop an AI ‘ecosystem’
Medical imaging, like tech giants such as Apple, must cast a wide net and incorporate all players in both healthcare and software development in order to implement AI and realize the changes mentioned above, the authors wrote.
In radiology, it is “crucial” to establish and maintain an “interconnected ecosystem” which includes government regulators, hospital chief information officers and technology support staff, individual radiologists and professional societies along with the radiology vendor community and, of course, patients.
“With these key players working in tandem, only then can AI be effectively and safely integrated broadly into radiology practices,” the authors wrote.
3. Challenges remain
Discussing the many promises of AI in radiology is one thing, the group wrote, but implementing the technology into clinical practice is another.
High-quality datasets are needed to create algorithms, which is not only time consuming for radiologists, but costly. Those methods must be put through serious validation and testing; if completed, clinical practice, legal and ethical challenges must also be considered.
If an AI algorithm makes a mistake, who is at fault? Can engineers create algorithms that overcome their “black box” limitation and allow radiologists to understand how a technology came to a decision? These are only a few questions ahead for the future of AI.
“It is crucial we build an ecosystem of key players in technology, research, radiology, and the regulatory bodies who will work together to effectively and safely integrate AI into clinical practice,” the researchers concluded. “As a result, adoption of this technology will expand our efficiency and decision-making capabilities, leading to earlier and better detection of disease and improved outcomes for our patients.”