The role of Radiologist is continually changing and looks to be on the precipice of another paradigm shift that may result in the traditional role of a Radiologist becoming obsolete.
As a specialty that is heavily dependent on technology, this should be expected; from advancements in X-ray production and digitisation, to the use of radioactive tracers and Positron Emission Tomography, the last 50 years have grown and formed a unique subset of medical professionals that provide an essential step in the overarching pathway for patient care.
Technological advancements have also changed workflow: many consultants have dedicated workstations to work from home, and teleradiology has boomed, filling reporting gaps and backlogs, as well as providing out of hours services that link rural communities with diagnostic services. It seems perfectly normal for NHS CT scans performed at night in the UK to be routinely live reported by colleagues in Australia.
The introduction of Artificial Intelligence (AI) and Machine Learning (ML) is the next big significant change, and appears to be almost inevitable in Radiology. However, care must be taken as the challenges posed by such technologies are still being evaluated. For the uninitiated, AI and ML in this form, are computer systems that can emulate human reasoning by applying logic to data or knowledge, often iterating and learning to improve performance (RSNA, 1996). This has been attempted at various levels in a theoretical and research capacity for the last couple of decades, but now, due to recent advancements in memory and processing capacity, this is likely to be rolled out clinically within the next decade.
Radiology is a relatively unique specialty in that the processes involved in image interpretation are complex from a human perspective; identifying patterns, shapes, mosaics, often moving images in various levels of contrast and colour, within an anatomical and clinical context, takes many years to master. However, if the correct software can be designed and supervised with archival images, a neural network could focus on specific data values and be trained to interpret diagnoses relatively quickly. The ML component would be the use of previous data to learn how to apply it to new data sets, such as patient imaging. In 2011, a team in Toronto trained a computer neural network to review chest radiographs (RSNA), and the results demonstrated a maximum sensitivity of 95%. More recently, ML has been shown to be 30 times faster in detecting breast cancer in mammograms than humans (Science Alert, 2016), and effective at spotting malignant lung nodules (RSNA).
Dr. Nick Bryan provides a nice introduction to the technical aspects of this process, and he claims that there will be a point where the machines are “better” than Radiologists at image interpretation (RSNA, 2016). They will be able to look at more images, with a higher accuracy, in a shorter period of time without getting tired. They won’t have satisfaction of search bias, and will possibly be more genial to junior doctors that request scans. Emphasis on the ‘possibly.’
But before you start planning to leave Radiology due to the impending rise of the robots, here are a few things to consider: firstly, the infrastructure is not yet in place to deliver this universally, in an affordable way. There are private companies that have the software and processing capabilities, but this also raises another question regarding the ethics of mass data collection, consent and data use: is altruism the only goal for these cloud service corporations? Who else will be able see the metrics and analysis generated and at what cost to the patient?
On a positive note, there may be multiple benefits for Radiologists. The role of the job will change from bulk image processing to more complex oversight and interventional medicine. People will also still be needed in practical and interventional Radiological procedures, which are a long way off automation at this point in time. Somebody will also have to teach, monitor and adjust the eventual neural networks, and human Radiologists will continue to act as gatekeepers, although perhaps in a reduced number.
For patients, the arrival of ML should be welcomed warmly; care systems are stretched to capacity and ML will help doctors regain some control in the battle for public health. In the near future, AI and ML should be seen as effective tools to aid diagnosticians. However, to ignore the likely future path of automation may be a luxury that Radiologists can ill afford.
Check out some of the latest Radiology cases below:
Developed by doctors, MedShr is the easiest and safest way for medical professionals to discover, discuss and share clinical cases and medical images.