Louisville Medicine Volume 69, Issue 11 | Page 22

AI LESSONS FROM RADIOLOGY Alexander Ding , MD
ARTIFICIAL INTELLIGENCE

AI LESSONS FROM RADIOLOGY Alexander Ding , MD

The hype cycle for artificial intelligence in medicine has come to a fever pitch . There is little doubt that it will overpromise and underdeliver , at least in the short term . However , it is highly plausible that in the medium to long term , AI will fundamentally change how medicine is practiced and transform health care .

The field of radiology prides itself in being technology-forward , innovative and early adopters of technology in medicine . Radiology was the first specialty to natively digitize with digital images ; first to telemedicine with the wide-spread practice of teleradiology ; first to use an electronic record with a picture archiving and communication system ( PACS ) system ; first to use advanced computation with CT scanners ; and the first to widely adopt natural language processing with dictation of reports . It is no surprise that radiology now leads the pack in the adoption of AI in regular clinical use . This article intends to share some of the lessons learned from radiology ’ s early experience with AI to share its promise , and its limitations and pitfalls .
There are pros and cons to being early adopters . The benefit of being an early adopter means the opportunity to not only gain experience and reap the value of this technology quicker and earlier , but also the opportunity to influence the development of the technology and tailor it closer to your specific needs . However , being an early adopter means dealing with the growing pains of the technology , stumbling over its limitations , expending time and resources through , at best , a buggy beta-test . Radiology ’ s experience with technology often serves as the “ canary in the coal mine ” as described by one of my former professors at the University of California San Francisco and it is these lessons in AI , I hope to convey through this article .
The first thing to emphasize is that the threat of AI coming to replace physicians is an empty threat . Medicine and health care will always need empathy and humanity , complex cognitive work and human dexterity that will not soon be replaced by AI . The better way to understand AI is as another tool in your arsenal . In fact , I would argue that the term AI should stand for augmented or assistive intelligence . From my experience in radiology , I often liken it to the introduction of the PACS system : a new technological tool that saw tremendous benefit for patients , referring physicians and radiologists alike in democratizing images , increasing efficiency and accuracy . But one which fundamentally changed the practice of radiology , such that we no longer had visitors in the reading room , and where some of our work could now be outsourced far and wide , commoditizing our interpretations . I predict that AI will also be a tool that will likely fundamentally change the way in which we practice , both with the positives and negatives associated with technological change .
According to the American College of Radiology , AI is already live in nearly a third of radiology practices . AI can be helpful in identifying abnormalities on imaging and is already used to identify ( with human over-read ) pneumothorax on chest radiograph , fractures on X-rays , suspicious lesions on mammography and pulmonary nodules on CT . It can increase study sensitivity , provide double-reads , and second opinions . It can also perform many of the tedious tasks that do not require extensive medical training , like comparing disease burden from one study to another over time , segmenting anatomy for surgical planning and quantitating and calculating clinical scores and measurements .
However , it is important to recognize that AI can be applied through the entire value chain in medical imaging , not just only at the interpretation phase . This includes identifying the best test to order for an indication for a specific patient at the time of order entry , reducing radiation requirements during image acquisition , providing feedback for quality control in image acquisition , smart triaging and prioritization of worklists based on predicted critical findings , auto-populating reports and providing “ peer ” review .
AI not only provides clinical value in improving accuracy and precision , but there are opportunities to provide value to the workforce and the business and practice . AI can remove tedious work like finding , counting , marking , measuring and comparing a zillion lesions between the last 10 studies over the past four years . And at a time of epic physician burnout , any opportunity to reduce the mundane , repetitive and wearying should be welcomed . Additionally , AI can provide a tremendous amount of business value with opportunities to increase efficiency and throughput , allowing increased work output in fewer hours and during this time of physician shortage , with fewer Full Time Equivalents ( FTEs ).
20 LOUISVILLE MEDICINE