This article is part of our special report Digital Transformation in Healthcare.
Radiology is at the forefront of artificial intelligence (AI) in the healthcare sector, as it can help enhance the quality of diagnosis on the basis of knowledge acquired from other patients, said medical professor Boris Brkljačić.
Three years ago, Canadian physicist Geoffrey Hinton shocked the medical world when he said radiology will be replaced by artificial intelligence (AI) in five year’s time. Radiologists should not be trained anymore because of that, he argued.
Others beg to differ, however. “It’s not really the case,” said Boris Brkljačić, Professor of radiology and Vice-Dean at the University of Zagreb School of Medicine in Croatia, who is also President of the European Society of Radiology (ESR).
In an interview with EURACTIV, he said artificial intelligence has a huge potential in helping radiologists do their work, for instance by allowing better interpretation of imaging and, consequently, diagnosis. However, he believes the technology won’t replace radiologists altogether.
“But it can also help in many other ways, even in the workflow with patients and integrating clinical decision with support systems,” he said.
He pointed out that lots of imaging data are currently stored in different hospitals from different countries with no purpose of research nor other use.
“This data can actually be integrated for clinical use, and then, together with the so-called radiomics, AI can help to extract some features automatically with computers, which humans cannot do,” he added.
The potential of AI is not only in better diagnosis but also in making complex relations linking imaging with genetic liability and even lifestyle to predict certain outcomes. “In precision or personalised medicine, this will play a major role for sure,” Brkljačić said.
Although feasible in the long run, dealing with these aspects requires more work, and entails training radiologists and students in this new area. One of the priorities of the ESR, Brkljačić’s association, is indeed to integrate AI and data science into training curricula.
Despite all the enthusiasm for AI, plenty of questions remain when it comes to the link between AI and ethics in the field of healthcare. This is particularly the case with regards to patients’ protection of privacy and data.
Among those is the use of patient data for commercial purposes, Brkljačić says, as some AI products will eventually deal with sensitive data belonging to patients.
“There are some countries like China where maybe these issues might not be so important,” he said, but this is not the case in Europe where the GDPR law protecting data privacy is crucial, he said.
Another issue involves the accountability for potential mistakes in diagnosis when an AI algorithms assisting a radiologist would come to wrong conclusions.
“Somebody has to be legally responsible. Is it the software developer? Or is it really the radiologist who signed the report?” he wondered.
An aspect to avoid as well is what he called the “black-box phenomenon,” when human medical skills are completely overcome by computers.
“You wouldn’t like to be the patient who’s told by a doctor: you had to go to surgery but I can’t tell you why because the computer told me so,” he said.
A technical limit that still needs to be addressed is the crucial and arduous process of labelling RAW imaging data to train algorithms with examples in order to deliver a diagnosis.
“It has to be done by somebody who knows very well radiology and those data have to be representative,” Brkljačić said. That could be an issue particularly for very rare diseases where the number of cases are insufficient nto train the algorithm properly.
Standards are another potential problem, as AI software are supposed to function in different types of scanners manufactured by different companies.
The European Society of Radiology recently published a paper to get a glimpse of AI’s real implementation in daily practice, mentioning several uses like in neuroradiology brain, multiple sclerosis and prediction of Alzheimer.
“But I don’t think that the technology is still ready for full-time clinical use,” Brkljačić warned.
Asked if AI could increase inequalities between hospitals that have the technology and those that don’t, Brkljačić said it could go both ways. If properly used, AI might in fact reduce costs of healthcare systems, and lower inequalities, he said.
However, he also admitted that the technology will first be implemented in the wealthiest hospitals that can afford it. “It’s true that, in the first glance, AI will be implemented in some major academic centre in most developed EU countries,” he said.
But in the long run, Brkljačić believes AI has the potential of making the healthcare systems more homogeneous, for instance by integrating pathology data from all over Europe, to the benefit of patients.
[Edited by Frédéric Simon]