A team of researchers has investigated the potential for artificial intelligence (AI) to detect diseases in medical images as well as humans can. After conducting the first systematic review of published studies related to this topic, the experts found that humans and machines have similar levels of accuracy.
The researchers explained that medical imaging is one of the most valuable sources of diagnostic information, but the need for diagnostic images is outpacing the capacity of available specialists, particularly in low-income and middle-income countries.
“Deep learning offers considerable promise for medical diagnostics,” wrote the study authors. “We aimed to evaluate the diagnostic accuracy of deep learning algorithms versus health-care professionals in classifying diseases using medical imaging.”
Out of more than 20,000 studies of applications using AI for medical imaging, the team found that only 14 produced good quality data. Collectively, these 14 studies suggest that deep learning systems correctly detect a disease 87 percent of the time, while healthcare professionals are accurate 86 percent of the time.
Study co-author Professor Alastair Denniston is an eye specialist at the University Hospitals Birmingham . He explained that while the results are encouraging, the study is a reality check for some of the hype about AI.
“There are a lot of headlines about AI outperforming humans, but our message is that it can at best be equivalent,” said study lead author Dr Xiaoxuan Liu.
The experts noted that there were few high-quality studies that involved a direct comparison or evaluated the performance of AI in a real clinic setting, which could significantly change the results of the meta-analysis. Because of these limitations, the researchers said the true diagnostic power of AI “remains uncertain.”
Professor Paul Leeson is a professor of Cardiovascular Medicine at the University of Oxford, who was not involved with the study.
“This paper summarizes the current state of research testing how well artificial intelligence identifies disease in a medical image compared to a clinician. The review has been performed very carefully but real challenges were encountered trying to get a useful result,” said Professor Leeson.
“The authors had to lump together findings from completely different medical problems and types of imaging, including research performed at a very early stage in the development.”
“Importantly the work highlights a new phase of research is needed, using more detailed trials, to work out the best ways to use artificial intelligence in healthcare.”
Image Credit: Shutterstock/ sdecoret