Dear MARNMP members,

​

It’s 4th April 2030.

 

Traffic is heavier than usual and I arrive 10 minutes late at my office in the medical imaging department. I place my index finger on the keypad next to the office door and the lock is released. As I walk into my office, the light dims and my workstation switches on. I take my jacket off and the workstation virtual assistant goes through the usual cordialities, “good morning Dr Mizzi, I hope you had a good weekend “. I reply with a “thank you Majel. Weekend was great. What’s new?”. As I stand on the treadmill in front of the workstation, Majel briefs me on my upcoming morning schedule and updates me on the previous week’s departmental workflow, number of requests, pending vetting issues, departmental output and efficiency...I ask her a few questions; which she answers succinctly. She opens the previous night’s emergency work list...10 accident and emergency CTs...All have been pre reported by the recently installed state of the art C3P0 software. C3P0 talks me through the scans, outlining its findings as it presents the images on the screen in an interactive fashion. I double check C3PO’s findings with the help of the mousepad on the treadmill handlebar, and using my voice recognition facility, I make diagnostic clinical conclusions to the reports and approve them. There’s not a single discrepancy. “Well done CRP0. You’ve been really great”. By now I have finished my morning work out. The treadmill makes way to my comfortable chair and a fresh espresso is waiting for me on my desk. Majel opens my main work list. There are elective CTs and MRIs from the weekend; all pre-reported by CRPO. Again, I’m taken through all the positive findings with automated comparison with previous imaging when necessary. CRP0 has a couple of unresolved observations related to two of the scans. In another two examinations, it has made interesting correlations with the patient’s blood results and in another couple, it has found important discrepancies in historical reports. In the next hour and a half or so I go through all exams and finalize and authorize all pre-reports made by C3PO. My work list is now cleared.

​

Majel tells me that she has taken three phone calls on my behalf, during the time when I was reporting. One by one, she calls back the clinicians and I reply to them, answering their queries. All calls are webcam supported, making the calls feel more like real office consultations. It’s time for a coffee break! When I return to my office, Majel brings up the respiratory multidisciplinary meeting patient list. All relevant examinations have been duly downloaded and all relevant images and reconstructions are ready for viewing. C3P0 runs me through the cases and relevant findings and I add some annotations through voice recognition. It’s half past twelve and I’m ready for the MDT meeting in the department seminar room. All participants are comfortably seated. C3P0 brings up all the cases, one by one, and projects the images alongside any relevant histology slides on the seminar room screen, whilst allowing all participants to make their comments. C3P0 picks upon the clinical conversations and summarizes the discussion, keeping a concise digital record of the points discussed and decisions taken on the patient’s virtual file. Meeting finishes at 2pm. C3P0 organizes 4 CT guided lung biopsies, including vetting. Emails are automatically sent to referring clinicians and radiologist and an appointment date is set on RIS, automatically informing bed management and sending appointment email/letter to the patient with all instructions.  It’s time for lunch....and back to the office for an afternoon tutorial.

​

 

On 28th March 2019, Professor Matthias Goyen gave a lecture on artificial intelligence to the Maltese Association of Radiologists and Nuclear Medicine Physicians. In his inspiring talk, professor Goyen outlined the concepts behind AI and how these will influence our future practice in Medicine, including Medical Imaging.

​

Artificial intelligence was the subject of a special two day conference organized by the European Society of Radiology on the 5th and 6th of April 2019. The conference was free to online participants. The keynote guest speaker was professor Toby Walsh, from Sydney, Australia, an artificial intelligence expert with 40 years experience in the field. He attributes the recent explosion in interest in artificial intelligence to four exponential growths: (a) the first is computer power; the number of transistors that one can place on a single chip, has been consistently doubling every two years (Moore’s law); (b) secondly, there has been an exponentially growth in funding of AI research; (c) the third factor is the exponential growth in computer data and (d) finally, the last exponential is particularly relevant to imaging. The ability for computer algorithms to recognize an image has also grown exponentially. The number of images misclassified by AI back in 2010 was 1 in 6. Now it's around 1 in 30. There's been a doubling in performance in the error rate every two years. 1 in 30 is better than human. The latter attribute has far reaching applications ranging from a self driving car recognizing pedestrians to medical imaging software recognizing pneumothorax on a chest Xray. Professor Walsh gave very much the same lecture to engineering students at UNSW Sydney, Australia on 9 May 2018, which you can read here.

​

The European Society of Radiology has recently published a white paper on Artificial Intelligence in Radiology. Various aspects of AI are explained, including deep learning, radiomics, etc. Ethical issues are raised, such as ownership of data, imaging biobank regulation and medicolegal issues, such as who will be medico-legally responsible for radiology reports issued solely by computers. AI is the star topic of the moment, and it will surely come our way, perhaps faster than we ever thought. However, it is not without difficulties. Like any major breakthrough in medical and other scientific fields, it will require regulation. Regulation of AI will necessitate multiple facets, including legislation specific to medical data ownership and data mining. Would AI software corporations channel money back to patients and hospitals that owned the clinical data that they used to develop their software? There are issues related to training radiologists of the future, alongside our deep learning machines. Would administrators prefer to channel money towards Radiology training or AI software? Would radiology trainees still need to acquire skills in plain film reporting if AI can do this task with a great degree of accuracy? Would general radiology training still be feasible or would it make more sense to train doctors in specific areas of imaging, e.g. breast imagers, cross-sectional imagers, and the like. Like with many other innovative scientific fields, there is often an initial hype driven by business in search of funding, that is often followed by a slow down in enthusiasm as reality meets hype. This would inevitably slow down the speed of the AI train. Finally, AI may be very good at some tasks but not so much at others. Try asking a robot to fold a towel. Presently, this job takes several minutes to achieve by the best available technology. Interventional Radiology jobs seem safe, at least for the time being!

​

 

Radiologists and Nuclear Medicine Physicians will inevitably see artificial intelligence coming their way. There will be change. Some of it will be good. Other aspects will be more painful and will require adaptation. The speed with which things are changing is fast. Some aspects are predictable. Others are not. Whilst there’s no doubt that we have to safeguard our specialities, we must always keep patient care at the centre of all our future decisions. Patients deserve best care and if AI improves patient care, then we must embrace it. Patient data should be safeguarded and any money generated from patient data by AI corporations should be channeled back into patient care.

​

To the question “Will radiologists be replaced by AI?”, the white paper answers very swiftly: “The simple answer is: NO. However, radiologists’ working lives will undoubtedly change in this era of artificial intelligence.” Radiologists and Nuclear Medicine Physicians should become familiar with the various concepts of artificial intelligence. “The real challenge is not to oppose the incorporation of AI into the professional lives (a futile effort) but to embrace the inevitable change of radiological practice, incorporating AI in the radiological workflow”. AI concepts should be introduced into our training curricula. Our trainees need to understand the concepts behind AI and should be conversant with the aspects of AI that are relevant to medical imaging. There will be challenges and turf issues ahead of us. If we resist change we’re doomed. If we adopt and change, we will survive and evolve, and stay at the forefront of patient care. Radiologists and Nuclear Medicine Physicians are encouraged to collaborate “with researchers to ensure it (AI) is deployed in a useful, safe, and meaningful way, and ensuring that its use is always directed primarily towards the patient benefit. In this way, AI can enhance radiology, and allow radiologists (and Nuclear Medicine Physicians) to continually improve their relevance and value”.

 

 

I look forward to my day at the office....in 2030....

 

​

Adrian Mizzi
President of the Maltese Association of Radiologists and Nuclear Medicine Physicians.