AI system 'can diagnose diseases like a doctor'

A team of Chinese and US researchers has developed an artificial intelligence system that can "read" health records and give diagnoses as accurately as experienced doctors.

In a paper published in Nature Medicine on Tuesday, the researchers said they have built an AI-based system that can automatically diagnose common childhood diseases after processing the patient's symptoms, medical history and other clinical data.

To train and validate the "smart" system, the scientists used it to study and interpret electronic health records collected from more than 1.36 million pediatric visits to a major referral center in China.

The system was highly accurate - comparable to experienced pediatricians - and may aid future doctors in diagnosing complex or rare diseases by providing more diagnostic predictions, the paper said.

In addition, it may be useful for triage patients, an important factor in areas that are short of medical resources.

"The study demonstrates that AI may help diagnose more diseases after systematically studying health records, but we still have much work to do," said Xia Huimin, co-author of the paper and a researcher at Guangzhou Women and Children's Medical Center.

"We need a wider range of data to validate the system's level of accuracy.

"The collection and analysis of high-quality data is a long process that requires close cooperation between algorithm engineers, clinicians and epidemiologists."