Introduction by Croakey: When Prime Minister Anthony Albanese made his AI in Australia’s interests speech last week about the economic possibilities of designing, making and building AI, he might have pointed to the health and medical sectors as key sectors for investment.
As the CSIRO’s new AI Trends for Healthcare report shows, AI technologies are moving quickly beyond research settings into real-world clinical settings, providing ways to improve diagnosis, prognosis and patient care, support better clinical decision-making, increase workflow efficiency, and help health services meet demand.
But there are also myriad challenges for professionals in managing AI technology procurement and adoption, as well as its ethical, effective, equitable and accountable use (not to mention the wider public health issues at stake, as recently covered at Croakey).
These dilemmas, and inevitable fears of job losses and de-skilling, will be on the minds of those attending #Intelligence26 in Sydney this week, a conference exploring the impacts and implications of rapid AI adoption in radiology and radiation oncology.
Run by the Royal Australian and New Zealand College of Radiologists (RANZCR), #Intelligence26 will showcase the profession’s multi-faceted response to being a leading sector for machine learning innovation.
Indeed, as Marie McInerney reports below, radiologists are at the global forefront of AI use – and the potential downsides of its misuse. This article is the first in a Croakey Conference News Service series from #Intelligence26.
Marie McInerney writes:
A decade ago, Dr Geoffrey Hinton, a British-Canadian computer scientist and artificial intelligence (AI) pioneer, famously declared that AI would soon kill the radiology profession.
The Nobel Prize laureate, dubbed the “the Godfather of AI“, didn’t mince his words, dramatically comparing radiologists to Wile E. Coyote from the legendary Road Runner cartoon.
“If you work as a radiologist, you’re like the coyote that’s already over the edge of the cliff, but hasn’t yet looked down, so doesn’t realise there’s no ground underneath him,” he said.
Yet so far the opposite has happened, as noted last year by Assistant Minister for Productivity, Competition, Charities and Treasury, Dr Andrew Leigh, among others, pointing to radiology salaries having instead risen and positions chasing applicants.
“Far from being replaced, radiologists have become more productive and more valued,” Leigh said.
Still radiology and radiation oncology offer many lessons and insights to the broader health sector from their place on the frontline of AI transformation, as this week’s Royal Australian and New Zealand College of Radiologists (RANZCR) dedicated AI conference will highlight.
#Intelligence26 will bring together clinicians, researchers and international experts to discuss governance, ethics and patient consent; workforce readiness and changing clinical roles; data quality and bias; and practical applications such as decision support, screening, triage and risk prediction, offering insights and lessons for the broader healthcare sector.
Transforming radiology
According to the US Food and Drug Administration (FDA) approved devices list, radiology is the main focus for medical AI innovation.
It was a relatively slow start: between 1995 and 2015, the FDA approved 33 AI radiology devices for use. However in 2023 alone, 221 got the nod.
As at 30 March 2026, of a total 1,524 FDA-cleared AI algorithms, the majority – 76 percent of approvals – were for radiology, with 68 of those new algorithms cleared in the first three months of 2026.
Much of the innovation comes from generative AI’s visual processing capacity, where it can map images against massive datasets, applying spatial reasoning and providing real-time analysis of data.
“AI’s big breakthrough was in analysing images, so radiology was the immediate first candidate,” said Dr Daniel Pinto dos Santos, Managing Senior Physician and Section Chair for AI and Imaging Data Science of the Department of Radiology at the Johannes Gutenberg-University Mainz in Germany, who will be a keynote speaker at the RANZCR conference.
He told Croakey that means radiology has “had a couple of years now to move beyond the hype” and to “discuss AI more rationally — requiring evidence, being aware of psychological factors, return on investment, regulations, etcetera.”
The conference comes amid hopes and signs that AI in healthcare can improve diagnostic accuracy, enhance clinical decision-making, personalise treatment plans, increase access to care, provide earlier detection of diseases and relieve workplace pressure by taking some tasks off radiologists.
But there are also big issues to resolve around bias, consent, efficacy, and evidence, not to mention AI’s huge environmental costs, and concerns that Australia has been “lagging” in healthcare AI development, deployment and governance.
Healthcare concerns
Whether those concerns have been allayed to some extent by Prime Minister Anthony Albanese’s announcement last week of a new Office of AI and national framework may depend on how closely the healthcare sector is involved in those developments.
The Lowitja Institute, Australia’s national Aboriginal and Torres Strait Islander health research institute, also recently published a timely policy and scoping review on AI, calling for Aboriginal and Torres Strait Islander leadership, perspectives, and Indigenous Knowledge Systems to be embedded throughout the AI lifecycle.
The report – Preparedness for Artificial Intelligence in Aboriginal and Torres Strait Islander Health: workforce policy, perspectives and future directions – says AI offers opportunities in the Aboriginal and Torres Strait Islander health context to strengthen prevention, diagnosis, service planning, population health monitoring, and culturally responsive models of care.
But it warns the benefits are not guaranteed and depend on how AI systems are designed, governed, and deployed.
“Gaps at any stage of the AI lifecycle can create or lead to unintended consequences, including biased decision-making, exclusion, privacy breaches, inappropriate use of data, reduced trust in health services, and the perpetuation or amplification of existing inequities.”
The Royal Australasian College of Physicians also issued a warning last week, that doctors have been left exposed to responsibility for patient harm cause by AI, as technological change outpaces legal protections.
Amid such concerns, interest in the RANZCR #Intelligence26 event has been so strong that organisers have had to move it in recent weeks to a bigger venue, at the Hyatt Regency in Sydney.
Trust and accountability
Trust is also a big issue for clinicians, as well as consumers, particularly for example in radiology where refusing consent for AI use might mean patients going without treatment.
“In some cases, AI is embedded within clinical care in a way that cannot be switched off, for example MRI scans use AI for reconstructing and optimising images,” said Dr Farhannah Aly, a radiation oncologist researcher with Sydney’s Ingham Institute for Applied Medical Research, who is one of four convenors of the conference.
“This is a very different scenario to the use of AI for transcribing patient notes, where it can be switched off,” she told Croakey. “This is why it is important for patients to be informed about the different uses of AI in their care and to understand how clinicians check the AI outputs.”
How those processes could best happen is the focus of new research by the Australian Centre for Health Engagement, Evidence and Values (ACHEEV) at the University of Wollongong.
It recently convened a citizens’ jury for a national deliberative process to capture public attitudes to AI in healthcare. It is analysing that data alongside stakeholder interviews, service user dialogue groups and choice experiments to understand what people think should happen with diagnostic and screening AI technology.
In 2024, ACHEEV’s Director Professor Stacey Carter reported a very strong emphasis from the citizens’ jury on the need for good AI governance.
“The jurors called for a national Charter overseen by an independent committee, high quality research to underpin AI tools, ongoing evaluation and monitoring to make sure AI performs the way its developers and vendors claim, and close scrutiny by regulators and health professional groups to make sure introducing AI technologies does not degrade the performance of health systems,” she said.
“There was also a strong focus on fairness, justice, rights and engagement: ensuring everyone can benefit from AI, preferring systems built on data from Australia to reflect our multicultural community, ensuring patient rights are upheld and engaging the community to understand AI and participate in its oversight.”
Investigating AI bias
Some of these concerns will be explored at the RANZCR conference by Daniel Pinto dos Santos.
He told Croakey he believes AI can, and hopefully will, deliver good for patients and the workforce. But he is urging radiologists to have an “honest and scientific discussion” about its implications and drawbacks.
He will speak at #Intelligence26 on the issue of bias — selections that skew training data or AI algorithms, leading to distorted outputs and potentially harmful outcomes, including underrepresentation of various communities.
He said that most machine learning concerns focus on biases in the data used to train AI models and the data those models are applied to.
For example, he said, it is quite hard to get image data from German hospitals, because data protection regulations are so strict that it is hardly feasible for startups to get enough well-curated information, raising questions about how clinically useful it can be.
But he is concerned also about the “much less explored phenomenon” of users’ psychological reliance on AI to do the right thing, known as automation bias, and the risks of that uncritical trust for “de-skilling or even never-skilling” within the profession.
“We expect the clinician/radiologist to be the one ultimately responsible for any medical decision. However, we know that the mere knowledge of the AI’s results/suggestion/prediction can significantly influence their decision to the point where they are not able to counter any wrong suggestion from the AI,” he said.
Evidence before adoption
In another keynote, Pinto dos Santos will also deliver a call to arms on ‘evidence before adoption’, highlighting a lack of randomised prospective studies of AI application that show evidence with patient-relevant outcome measures.
He believes there is only one convincing trial so far, the 2023 Swedish MASAI trial, which found that an AI-supported mammography screening resulted in a similar cancer detection rate compared with standard double reading.
“All other use cases until now fail to provide convincing evidence,” he said.
“On the contrary, some evidence at the moment for use cases like prostate MRI (magnetic resonance imaging) and brain haemorrhage rather show that using AI did not have any beneficial impact and the respective institutions stopped using it again.”
His worry is that unlike pharmaceutical companies, AI startups just don’t have the resources required to run clinical trials.
“And for some use cases (for example, fracture detection, chest X-ray reading) I even doubt it would be possible to generate evidence, because there are just so many confounders beyond the radiological image read,” he said.
That makes him keen to hear two presentations at the conference on trials, including in the radiation oncology stream, the award-winning ASTuTE clinical trial ,which is investigating how an AI-based precision medicine test can guide shared treatment decision making for men diagnosed with prostate cancer.
In a concurrent session, Melbourne radiologist Associate Professor Helen Frazer will outline the development of an AI algorithm used to both detect breast cancer in screening images and estimate a woman’s risk over the next four years more accurately than traditional factors such as age, breast density, and family history.
Breast cancer screening
Frazer, who is State Clinical Director for BreastScreen Victoria and Associate Professor at the University of Melbourne, told Croakey that the tool offers “the potential for the most significant step change in breast cancer mortality since screening was introduced in Australia in the early 90s”.
The detection and risk prediction tools were developed by the BRAIx program – a partnership, funded by the Medical Research Future Fund, based at St Vincent’s BreastScreen Melbourne with researchers from St Vincent’s Hospital Melbourne, St Vincent’s Institute of Medical Research, The University of Melbourne, The University of Adelaide, and BreastScreen Victoria.
According to the population cohort study published in April in The Lancet Digital Health, the BRAix risk score was developed using mammograms from nearly 400,000 women. It was then tested on data from almost 96,000 women from Australia and confirmed in a separate Swedish population of over 4,500 women.
That’s now led to a prospective randomised controlled trial, begun last December, using the BRAIx detection algorithm to work alongside radiologists in the population screening program.
“It’s probably the first AI randomised controlled trial for a high consequence healthcare decision in Australia, aimed at improving outcomes for women,” Frazer said of the work she will present on at the conference (more to come at Croakey in the coming weeks).
Analysing overlooked information
Associate Professor Hyun Soo Ko, also an #Intelligence26 convenor, will speak at the conference on opportunistic screening and the potential for AI to improve population health outcomes, believing that AI can unlock the untapped potential of everyday imaging “to prevent major health events and reduce inequities”.
Every year, she said, millions of computed tomography (CT) and MRI scans are performed in hospitals and clinics for trauma, cancer staging, and routine diagnosis.
“These scans contain rich, often overlooked information: bone density and vertebral fractures, coronary and aortic calcification, emphysema and lung nodules, liver fat, and body composition,” she said.
Ko said her presentation will showcase practical radiology and radiation oncology use cases that can flag patients at higher risk for morbidity and mortality, and can enhance cancer survivorship care by identifying downstream side effects of radiation therapy.
Developed and monitored “in a sound ethico-legal and practical framework”, she said these new tools can create opportunities to improve general population health outcomes including equity for often underserved Aboriginal and Torres Strait Islander peoples, Māori and Pacific communities, and rural and remote populations.
“This is not a vision of a distant future. It is a conversation about what radiology and radiation oncology can start doing differently right now,” she said.
Follow Croakey’s #Intelligence26 posts at LinkedIn: @CroakeyHealthMedia
You can listen to this episode of RANZCR’s podcast, The Scan, on how AI is reshaping radiology and cancer screening with Associate Professor Helen Frazer and #Intelligence26 co-convenor Dr Martin Gunn.
Declaration: Marie McInerney was involved in editing the Lowitja Institute’s AI report.
Bookmark this link to follow Croakey Conference News Service’s coverage of #Intelligence26.
Add Comment