The research described in Nature Biomedical Engineering, found that the model was more effective at identifying problems such as pneumonia, collapsed lungs, and injuries than other self-monitoring AI models. In fact, it was similar in accuracy to human radiologists.
While others have tried to use unstructured medical data in this way, this is the first time a team’s AI model has learned from unstructured text and matched the performance of radiologists, and demonstrated the ability to predict multiple diseases from an X-ray determined with a high degree of precision, says Ekin Tiu, a Stanford undergraduate student and visiting researcher who co-authored the report.
“We are the first to do that and demonstrate it effectively in this field,” he says.
Code from the model has been made publicly available to other researchers in the hope that it can be applied to CT scans, MRIs and echocardiograms to help detect a broader range of diseases in other parts of the body, says Pranav Rajpurkar, a professor biomedical assistant. computer science at the Blavatnik Institute at Harvard Medical School, who led the project.
“Our hope is that people can immediately apply this to other chest X-ray and disease datasets that concern them,” he says.
Rajpurkar is also optimistic that diagnostic AI models that require minimal supervision could help increase access to healthcare in countries and communities where specialists are in short supply.
“It makes a lot of sense to use the richest training signal reported,” says Christian Leibig, head of machine learning at German startup Vara, which uses AI to detect breast cancer. “It’s quite an achievement to get to that level of performance.”