New research suggests that AI (Artificial Intelligence) is just as effective as the human mind when it comes to making medical diagnoses based on imagery.
Much excitement has been generated around the advances that AI can bring to the healthcare industry.
Greater use of technology and artificial intelligence in the medical profession is often seen as paving the way to earlier detection, treatment and even prevention of deadly cancers and diseases.
Advocates say AI will free up more time for patient-doctor interaction and ease the strain on healthcare resources and waiting lists.
Now, a study in Lancet Digital Health has found that AI models are just as effective, if not more, at giving a correct diagnosis compared to human medical experts when assessing images and scans.
However, the suggestion that machines are on a par with the expertly-trained medical mind comes with a caveat.
The study’s authors believe it comes off the back of years of poor-quality research into the role and influence AI could have on healthcare development.
Writing in Lancet Digital Health, Dr Xiaoxuan Liu and Prof Alastair Denniston filtered through around 20,000 studies which had been published on the subject since 2012.
Of those, they found that only 14 were based on good quality data.
Building on those findings, they concluded that deep learning systems in AI models correctly detected a disease 87% of the time – compared with 86% for expert healthcare professionals.
The AI model correctly gave an all-clear 93% of the time, compared with 91% for human experts.
However, given the volume of perceived poor-quality research they had to trawl through, author Denniston said reaching their findings served as a reality-check over the hype surrounding AI and healthcare.
However, Denniston remained optimistic, saying the models still provide a diagnostic tool and could help tackle backlogs of scans and images quicker than a human could.
Meanwhile, Liu said AI assessments could prove useful in regions and countries where there are no experts to interpret those images.