There is much concern over the potential of Artificial Intelligence (AI) and job security in all fields, and when it comes to medicine, radiology seems to particularly be in the crosshairs. This uncertain territory, as well as the unfamiliar jargon, only adds to the fear. Terms such as "deep learning," "machine learning," "big data" and so on induce more than just a flutter of unrest. As with all things unfamiliar to us—including AI—the best way to curb irrational fears is to get the facts and learn. AI has the potential to transform the practice of radiology in the modern era, and instead of being left behind, we must use our expertise as master diagnosticians and imaging experts to properly guide the AI revolution and positively impact radiologic care. Not only can AI improve our workflow efficiency, productivity, accuracy, and satisfaction, but it can most importantly help improve the quality, outcomes, and delivery of our imaging patient care.
Several practical implementations of AI have already been studied and put to use, providing a glimpse of our future. A team at The Ohio State University led by Prevedello et al. developed an AI algorithm which automatically evaluated head CT studies as they were scanned for critical findings to help appropriately order the studies by severity on the radiologist's worklist. The algorithm was successfully able to identify acute hemorrhage, mass effect, hydrocephalus and suspected acute infarcts, and flag these studies to be read first by the radiologists in that busy, high-volume practice (1). This work demonstrated great promise for further developing similar algorithms to make the initial assessment and guide the radiologist to review studies with potentially critical findings first to deliver timely patient care. Rather than "competing" with radiologists, AI hence can aid radiologists to provide an even better service to patients.
Other future avenues of improving the radiology workflow include AI delving into the sometimes hundreds of medical notes and previous radiology reports to help the radiologist generate a more clinically relevant report efficiently. AI can potentially mine the patient's electronic medical record for pertinent medical and surgical history, labs, and reported physical exam findings. This automatization alone could provide the radiologist with helpful clinical history leading to better interpretations for our increasingly complex patients. Likewise, AI can increase our accuracy and productivity by automatically measuring lesions and correlating size changes with prior studies, so we can focus on our roles as interpreters of the findings and as consultants to our colleagues and patients—an emphasis of the ACR's Imaging 3.0 initiative. In addition to helping provide better imaging care, this role of AI will likely improve our work environment, job satisfaction, and reduce burnout as we get to focus more on our patients and the true application of our expertise in patient care while reducing the mundane, repetitive aspects of our work.
In addition to gains in improved finding detection and efficiency, the potential role of AI in prognostication and personalized therapies is enormous. AI applications such as texture analysis can have implications in detecting whether a lesion is benign or malignant and further predicting long-term outcomes, acting essentially as a non-invasive digital biopsy. A recent study by Choi et al. demonstrated the potential role of this analysis in predicting outcomes of pancreatic adenocarcinoma evaluated by MRI (2). Another work by Digumarthy et al. showed successes in reliably predicting well- and poorly differentiated lung malignancies evaluated by CT (3). As we embrace our role as experts in imaging informatics, we can use such tools to provide better consultations to our patients and our colleagues. In fact, such synergistic tools can solidify our role at the center of healthcare delivery, rather than replace or outcompete us.
Despite the positive potential impact for radiologists and radiology, the perceived fear of AI within the field is real. A recent survey showed that this fear is palpable not only among radiologists, but even more so among trainees. In the current crossroads, many second guessed their decision of entering radiology with half of the respondants indicating either yes or a maybe to the question: Would their existing knowledge of AI have changed their decision to enter radiology? (4). The only true way to dispel this fear and ensure a bright future in radiology is to equip ourselves to lead this revolution and shape the future applications of AI. Together, human and machine can create a better patient experience and better clinical outcomes than either can alone. We have always successfully adapted whenever a technological advancement rolled out or rapid change faced us, and AI is no different. The path to success in the AI revolution is to own it now.
1. Prevedello LM, Erdal BS, Ryu JL, et al. Automated critical test findings identification and online notification system using artificial intelligence in imaging. Radiology. 2017;285(3):923-931.
2. Choi SY, Kim SH, Kang TW, et al. Differentiating mass-forming autoimmune pancreatitis from pancreatic ductal adenocarcinoma on the basis of contrast-enhanced MRI and DWI findings. AJR Am J Roentgenol. 2016;206(2):291-300.
3. Digumarthy SR, Padole AM, Logullo R, et al. CT texture analysis of histologically proven benign and malignant lung lesions. Medicine. 2018;97(26):1-3.
4. Collado-mesa F, Alvarez E, Arheart K. The role of artificial intelligence in diagnostic radiology: A survey at a single radiology residency training program. J Am Coll Radiol. 2018;17:1440-1546.