New subspecialty or close ally emerging on the radiology horizon: Artificial Intelligence

Artificial Intelligence (AI) is becoming a hot and evolving topic in the radiological sciences nowadays. Its implementation, challenges, and future directions on the radiology horizon are discussed in this letter. AI will likely find a place as a subspecialty in the radiological sciences in the near future.

Perhaps even Professor Wilhelm Conrad Roentgen himself did not believe his historic invention when he first discovered X-rays. This was a milestone in medicine, and he may not have realized the impact it would have on medical history when he published his work (1).

Replacing a medical doctor with a robot has remained a mainstay in science fiction for decades (2). AI in radiology is a bit different, as there is no individual inventor; there were lots of contributors, including the software known as convolutional neural networks (CNNs). In horse races, nobody is interested in the jockeys and their names are not well-known, but the horse’s names and unique features stick in people’s minds. You see the same attitude in robotics; nobody remembers the inventors of intelligent robots, just the robots themselves—such as Sofia (3), a robot can that can hold meaningful conversations, and more importantly, is designed to learn and communicate. Some people may find this idea scary; others might find it thrilling.

As a radiologist, if you are good at IT, then you are likely a fan of AI. It is easy to foresee a new subspecialty or close-knit ally coming into practice in the radiological sciences. The current millennium looks fertile regarding advances in medicine and radiology after the implementation of AI. In a few decades, radiologists may be categorized as AI-capable or AI-friendly and perhaps this will be a feature that may help you find your dream job. In the AI era, we need to learn about this unique implementation in radiological sciences, rather than remain in denial. Even if you are in denial, you may be using AI without realizing it. Nobody would deny that if they felt hesitant about a diagnostic challenge, they would seek help from the so-called Doctor Google (4). Of course, similar to the delivery of a new baby, the process will be a bit painful, but concentrating on the benefits rather than the complications is much more sensible, beneficial, and productive. The most important thing is how we are controlling AI in terms of quality and safety of the patients and those who use it, such as data scientists, data engineers, and doctors.

There were several talks and presentation sessions on the subject during last year’s Radiological Society of North America (RSNA) annual meeting. Dr Vijay M Rao, the 2018 president of the RSNA, stated that AI applications are already proving themselves extremely useful in radiology, and that these applications provide tools to make radiologists more efficient. She emphasized that the time has come for radiologists to provide leadership in this field (5).

AI research in radiology has increased since 2015. Now Radiology: Artificial Intelligence, a journal under the RSNA, has emerged and is accepting AI-related papers from scientists, physicians, and radiologists.

Data scientists and radiologists both contribute to this innovative field. The radiomics of CNNs will evolve but will remain a bit mysterious, like a black box, as the interactions in these neural networks would not be fully understood, exactly like those in the human brain. The neural network can be multi-layered and could easily contain hundreds of hidden layers (Fig. 1).

Fig. 1.

CNN working diagram.

This author believes that learning one of most common machine learning AI CNN languages—Python—would facilitate our understanding of what is going in the mind of a CNN and create common ground with our future data scientists, who will be our colleagues and allies in AI-enhanced radiology departments. AI data science master’s and PhD programs will eventually be rolled out for radiologists.

Whether AI is a friend or a foe will remain a dilemma in most radiologists’ minds, but I believe AI is a reliable, hard-working friend rather than a foe and should be included in radiology training curricula in the near future (6).

Doctors and radiologists are far from protected against the coming of AI. We physicians, in particular, have for centuries enjoyed an unwarranted sense of guarantee that is now changing dramatically. The rollout of evolving AI is likely to be incremental but is absolutely assured. Patients and their loved ones will see the benefits in the government budgets and perhaps the macroeconomy. As a matter of fact, it is hard to think of hospitals without imaging. Admittance to the hospital for both outpatients and inpatients would involve some sort of imaging. Radiologists should anticipate and be receptive to these changes and make a good plan of survival. It is clear that AI would be quite sufficient for both diagnosis and triage and would be a partial substitute for a radiologist. Therefore, within a decade or so, we will be witnessing job cuts to radiologists or even radiographers, especially in the private healthcare sector. On the other hand, AI will be helpful in countries that have been suffering for decades due to unfilled radiology jobs, such as the UK, which has the most severe shortage of radiologists in Europe. Vacancies in the UK were 10.3% according to the Royal College of Radiologists (RCR) census report (7).

The need for active engagement with AI at the radiology trainee level was made clear by leaders in the field at the RSNA’s annual scientific meeting in 2017. They suggested that radiology training in AI may become as essential as radiology training in physics (8).

In conclusion, AI will become an ally of radiologists and will be soon accepted as a core subject in the radiology curriculum. It will likely find a place as a subspecialty in the radiological sciences in the near future.