Thursday, 13 June, 2024
HomeTechnologyWill AI make radiologists redundant?

Will AI make radiologists redundant?

The diagnostic power of artificial intelligence (AI) is rapidly rising and changing healthcare along the way, with experts cautioning that if medical professionals try to resist, or don’t catch up, they’ll simply be left behind.

AI has added a new term to the medical lexicon: “previvor” – someone in whom new technology can pre-diagnose a serious or life-threatening condition even before symptoms are revealed.

No doctor on their own can do so with such accuracy.

Most doctors will concede they know only fragments of medicine well. But take the entire internet, Wikipedia, a few hundred billion books, feed all that data into AI computation, and you’ve got GPT4 – and it knows a lot of medicine, writes Chris Bateman in Financial Mail.

Used in healthcare, AI programs now conduct clinical diagnoses with a better accuracy rate than human doctors. And if medical professionals are resistant to precision medicine and AI, well, “the train has already left the station”, says Deshen Moodley, associate professor at the University of Cape Town (UCT) and the first occupant of the South African Research Chairs Initiative chair in AI systems.

“Physicians will have to scramble to catch up. If they don’t adjust, they’ll be left behind,” he told the Financial Mail. “People have access to a whole lot more knowledge today, and patients will be the future drivers of change.”

With smartphones, computers and digital technology helping with healthcare, there’s a shift away from doctors, which brings with it the potential to halve healthcare budgets within 20 years, Moodley adds.

Besides open-source access to healthcare knowledge, the tools are already here to help doctors with better predictive diagnoses and preventive treatments.

Take AI-powered technology from US company Prenuvo. It uses magnetic resonance imaging to flag potential health problems, “seeing” complications that a radiologist or cardiologist cannot. Before symptoms emerge, the technology predicts the condition, enabling physicians to figure out prevention methods.

It analyses every part of the body to detect all forms of pathologies, from cancer to liver stones and kidney diseases. There are no invasive procedures or radiation, and the results are made available to the patient within 48 hours.

AI-powered coronary artery scan technology, developed by the Fountain Life health technology company in the US, detects heart pathologies, especially heart attacks, a decade before symptoms surface. The AI coronary artery scan (CAS) is similar to the coronary computed tomography angiography used for years to diagnose heart disease.

The difference is that CAS uses the analyses of radiologists and cardiologists, then adds AI to scrutinise the results. This ensures plaques previously concealed from the human eye are seen.

David Jankelow, a cardiologist practising out of the Linksfield Park Medical Centre in Johannesburg and a past president of the South African Heart Association, is working on AI with the Mayo Clinic. The US clinic has developed an algorithm that works with an ECG machine to predict cardiac disease before it clinically manifests.

The clinic is seeing amazing results predicting atrial fibrillation, the biggest cause of stroke, as well as aortic valve disease, cardiomyopathy and other heart dysfunctions.

“The clinic took a cohort of about 53 000 patients and fed their data into an algorithm paired with an ECG and ultrasound,” says Jankelow.

“Then they took a different data set of patients and fed their ECG readings in and asked who had a weak heart pump. The accuracy was 93%. They followed all the computer false positives and the truly AI ECG-tested patients for several years. The former had a fourfold chance of developing a weak heart and the latter remained well.”

This, says Jankelow, is how the term “previvor” was coined.

It’s not just in the realm of predictive medicine and early treatment that AI has valuable application; it’s a real-time decision-making tool for disease outbreaks, with the ability to play out all the possible response models in pseudo-reality, says Moodley.

“It will not just follow an outbreak in real time but see it spatially and (give) you a completely different real-time decision-making tool. You can rewind and see what happened in the past, scrutinise (it), and say ‘This led to that’, whether it be Covid or any communicable disease.

Professor Athol Kent, a UCT-based medical education innovator and obstetrician/gynaecologist, writing on the Journal Article Summary Service site, observes: “AI will move us forward, and Luddite responses will merely discredit us. We – the doctors – will have to figure out how AI works to our advantage and stay informed as to how this is best achieved.”

He believes the sector needs both more and different types of health-care professionals, “and maybe these will be a mix of humans with communication skills and AI technology”.

“We need better ways reaching and responding to our patients, more time with those we care for better systems to run our practices and hospitals. All of these possibilities can be facilitated by AI to the clinical benefit of us all,” Kent writes.

Jankelow adds: “There’s a large gap, which some call the digital divide, from research and development of this technology to implementation in clinical practice. But it’s hotting up.”

His dream is the democratisation of healthcare. “There are areas in South Africa where there are no doctors, nursing sisters or ECG machines. As far back as 2016, we had one cardiologist per 260 000 people. Brazil had one in 23 000, 10 times more than us. That hasn’t changed much.”

Tracing the recent evolution of AI, Harvard University’s Isaac Kohane, a pioneer in the AI field and a paediatric endocrinologist who leads Harvard’s bioinformatics department, says things began changing in 2012.

“We were already impressed with (AI), but in about 2018 we started seeing, in the medical literature, the consequences of these convolutional neural networks – that they could actually detect changes in images that were perhaps imperceptible to humans, which would allow them not only to diagnose retinopathy in the back of the eye, for example, but could tell you the BP, the sex of the patient, their age and what other diseases they had.”

What changed, Kohane says, was that the new programs were trained for specific purposes – “diagnosing retinopathy (and) predicting time to readmission”, he says. “These were programs that not only were purpose built but, because of that, you could easily evaluate them and assess their accuracy for a specific task.”

Jankelow cites several studies predicting that by 2040 there will be 640m diabetics worldwide, with a third of them women, suffering some form of diabetic eye disease. The global shortage of ophthalmologists, narrowed down to South Africa, comes in at six for every 1m people.

“Some form of computer analysis will hugely ease the burden on the healthcare system – and that’s just one discipline,” he says.

As a natural language processing tool, GPT allows you to have a personalised conversation with an AI bot. It’s the fastest uptake of any technology in the history of humankind, says Jankelow, with 100m users globally already.

“It passed the US medical licensing exams with no cramming, unlike human students,” he says. “We must remember it’s not a replacement for specialised knowledge, critical thinking and ethical consideration in the practice of medicine. But it’s the next step in the fourth industrial revolution, where all processing in terms of search tools is going. Every search engine will have this.”

As “the art of making machines smart”, AI will soon render radiologists redundant, in Jankelow’s view.

Speaking about redundancy in the profession, he paraphrases Eugene Stead, who paved the way for cardiac catheterisation in the 1940s and led the concept of a computerised textbook for medicine: “If you’re a radiologist, you’re already over the edge of the cliff – you just haven’t looked down yet. There’s no ground underneath. In five years, deep learning is going to do the job better than radiologists. They should stop training them now.”

Still, he doesn’t believe AI is “some kind of magic bullet”.

“The answer is ‘no’ in the short term, but ‘likely’ in the medium to long term. Imagine a complete overview of patient data and the reduction in medical errors. Solving medical problems and recommending individualised treatment will become the norm.”

Asked how AI could help South Africa, as the most unequal country in the world, Moodley says: “There are two big issues. One is access to AI, beyond just health. That drives our AI group at UCT. We need open-source platforms, otherwise the cost goes up. We need to mobilise the use of AI for open-source applications and reference implementations and platforms to reduce cost.

“The other is that if you look at our current systems, we’re only using about 10% of them in terms of innovative AI possibilities on the preventive side. Google Search, ChatGPT and GPT4 have knowledge that can empower individuals to start managing their health.

“The only difference between me and entrepreneurs is that our public health platforms will be freely available whereas they want to make money from it.”


Financial Mail – Will AI make radiologists redundant? (Restricted access)


See more from MedicalBrief archives:


SA’s radiologists flocking to write equivalency exams in UK, Australia and NZ


AI system accurately detects key findings in chest X-rays of pneumonia patients within 10 seconds


AI can’t replicate this crucial aspect of practising medicine








MedicalBrief — our free weekly e-newsletter

We'd appreciate as much information as possible, however only an email address is required.