Artificial Intelligence (AI) and deep learning are a hot topic in radiology today. Recently, the Stanford Machine Learning Group published results of AI used for diagnosing pneumonia from chest X-rays. Turns out, machine learning is able to detect pneumonia at a level matching or even exceeding radiologists. Should radiologists feel threatened? Or should they embrace this new technology? We spoke to Erik Ranschaert, MD, PhD, CIIP and Vice President of the European Society of Medical Imaging Informatics (EuSoMII).
"AI is one of the fastest growing technologies and has proven its success already today. The University of Nijmegen, for example, already uses software to differentiate between malignant and benign breast cancer cells. It saves pathologists a tremendous amount of time and work. Fact is, AI is here to stay and to conquer. So we should look into how radiologists can use it to their benefit."
How can radiologists benefit from artificial intelligence?
AI might play an important part in the future of radiology. Radiologists today are drowning in work and often there’s a shortage of skilled medical professionals. In Europe, the UK is leading the way in AI research and development, which is not surprising when you know that radiologists there are seriously understaffed. Thanks to machine learning, radiologists will be able to free up time to focus on more complicated studies, reduce errors, and spend more time with patients.
Dr. Ranschaert believes there are four key domains in which AI can truly make a difference for radiologists (cf. diagram). “Next to image interpretation and treatment (think of the Stanford chest X-rays algorithm), AI can have a major impact on the radiologist’s workflow. Computer-aided image triage in the case of reporting, for example, or automated scoring of images, data incorporation, and translations could help radiologists deal with their ever-increasing workload.
He also sees opportunities in the field of communication, by automatically adapting the reporting style, for example, through Natural Language Processing (NLP). It means patients and physicians could receive customized reports.
Will artificial intelligence replace radiologists?
Although many tasks of the radiologist can be managed with machine learning, radiologists will be more difficult to replace than often suggested. “Sure, AI will be disruptive in the way we read and report on imaging exams. Imaging is becoming quantitative, and we will have to integrate all findings and analyses into our reports. AI can also help with this. It will definitely change the role of the radiologist, but it won’t make him or her unnecessary. You’ll always need someone who takes the final decision.”
"It’s like boarding an unmanned airline. Would you step into a plane without a pilot? I don’t think many people would. Even though technically, it’s possible, you’d still want the pilot to manage take-off and landing, especially in unusual and unexpected situations. You need someone who is in charge of the plane and takes responsibility for every decision made."
Challenges to overcome
Unleashing a disruptive technology such as AI into the clinical field doesn’t come without challenges. There are a few things that need to be considered first before AI can really kick off in the field of radiology. “Standardization, validation and integration are key,” Dr. Ranschaert continues.
Standardization is needed for acquisition and annotation of data, structured reporting and archiving... Standardization is also required to enable full integration of these data. At the same time, AI software needs to be tested and validated, again fueling the need for an internationally accepted standard or ecosystem.
"What’s more, for AI to be integrated in radiology, we’ll need to bring radiologists into the loop. We’ll need to provide them with incentives. And support from the government (and a reimbursement structure) is required to turn AI into the game-changer it is believed to be."