Sorting by


The new AI technology that could save your life

Radiologists can evaluate complex brain scans much faster and more accurately when they use artificial intelligence, or AI, as a ‘second pair of eyes’, a study has found.

Researchers found that radiologists could interpret head CT scans for brain hemorrhages 32 per cent more accurately and 11 per cent faster when they were assisted by artificial intelligence.

The NHS is using the technology at several of its hospitals, although the research was carried out in Australia.

The study, published in the European Journal of Radiology, compared how accurately 30 radiologists were able to detect ‘clinical findings’ on 2,848 complicated CT brain scans with and without the aid of a deep-learning system created by Annalise, an Australian AI company with offices in the UK.

The researchers found that the AI was particularly beneficial in helping radiologists to interpret the most difficult brain scans.

Quinlan Buchlak, of the University of Notre Dame Australia in Sydney, said the AI assistance “significantly improved radiologist detection accuracy across a wide range of clinical findings and demonstrated the potential to improve interpretation [of complex brain scans]”.

“The benefits were most pronounced when aiding radiologists in the detection of subtle findings,” he said.

“Although some findings proved difficult for radiologists to detect, subtle abnormalities were generally present on the CT scan that allowed detection by the AI model. The considerable improvement of the radiologists in detecting these when assisted by the model suggests that the findings were visible to the human eye even though they were often missed in the unassisted arm of the study,” he said.

Annalise’s chief medical officer, Rick Abramson, said: “The results have important implications for patients and clinicians.”

A ‘subarachnoid’ hemorrhage means there is bleeding in the space that surrounds the brain, which is life-threatening. And so the quicker and more accurately it is identified the faster emergency surgery can be performed.

Radiologists often use a liquid, known as a contrast, that is taken orally or through an injection that allows the organ being examined to be seen more clearly.

However, many scans do not involve contrast because it can cause kidney failure in some people with kidney problems, while others are allergic to the liquid. Meanwhile, people are not allowed to eat for thee hours before contrast is used, meaning that in emergency cases where the patient has recently eaten, a contrast cannot be used.

The lack of contrast means that ‘non-contrast’ computed tomography (CT) scans of the brain can be difficult to interpret, making the assistance of AI particularly valuable, experts say.

Nick Woznitza, a consultant radiographer at University College London Hospitals and clinical academic at Canterbury Christ Church University, who was not involved in the research, welcomed the findings.

“This well-conducted study adds another incremental piece of evidence that suggests artificial intelligence clinical support tools are accurate and improve radiologists when reviewing CT scans of the brain,” Dr Woznitza told i.

“Crucially, the additional information provided by the AI to the radiologists did not appear to increase the time taken to review the scans.

“What is yet to be determined is if these promising initial results are replicated in routine practice and if they improve outcomes for patients. Work like this paves the way to conducting the further rigorous clinical evaluations that are required to decide how best to use this promising technology and for which patients,” he said.

The study was based on radiologist readings in Australia by researchers from several universities there and led by the University of Notre Dame Australia, Sydney.

The deep learning model being examined was trained using 212,484 scans drawn from a private radiology group in Australia.

Source link

Related Articles

Back to top button