Common childhood diseases as well as dangerous conditions can be diagnosed with an artificial intelligence system just about as accurately as an experienced pediatrician, scientists in San Diego and China report Monday.
The AI system processes large amounts of data from electronic health records, and also finds disease associations not previously identified, according to a study led by Kang Zhang, founding director of the Institute for Genomic Medicine at UC San Diego School of Medicine.
The work shows an across-the-board usefulness. Previous medical AI efforts were often focused on one disease or related conditions. For example, a study published in February of last year, also led by Zhang, was designed to screen patients who may have potentially blinding retinal diseases.
The AI system interpreted data from more than 1.3 million pediatric visits to a health care referral center in Guangzhou, China. Results were compared to a doctor’s initial diagnosis. It not only performed well in diagnosing common ailments, but also dangerous conditions such as bacterial meningitis, the study said.
For example, it achieved 85 percent accuracy with gastrointestinal diseases and 93 percent for bacterial meningitis. This can be useful for identifying patients who urgently need treatment.
“Although this impact may be most evident in areas where healthcare providers are in relative shortage, the benefits of such an AI system are likely to be universal,” said the study, published in Nature Medicine. It can be found at http://j.mp/aichildmed.
The study demonstrates the growing usefulness of artificial intelligence in medicine, said Giorgio Quer, Director, Artificial Intelligence at Scripps Research Translational Institute.
While the system in this paper can’t yet outperform an experienced physician, it could assist junior physicians, he said by email.
“An interesting application of this system - that would require new data and that should be clinically validated - would be in the emergency department,” Quer said. “This system could potentially analyze vital signs, basic history and notes from a physical examination by a nurse to help in the prioritization of patients who should be seen first by a physician.”
The AI system was trained on manually annotated information. Once trained, it was set loose on patient data in the electronic health records, including symptoms, physical exams and results of lab tests.
The system used natural language processing methods to extract pertinent data from doctors’ notes. It proceeded toward a diagnosis with a question-and-answer format just like doctors use.
“Examples of questions are ‘Is patient having a fever?’ and ‘Is the patient coughing?’” the study said.
Aside from identifying patients who need immediate attention, the study said the AI system might even be able to pick out patients who don’t need a doctor’s evaluation at that time. These patients can be referred to routine followup care at a later date.
“This diagnostic prediction would help to ensure that physicians’ time is dedicated to the patients with the highest and/or most urgent needs,” the study said.
The study also included researchers from VA Healthcare System in San Diego; Guangzhou Medical University, Guangzhou, China; Hangzhou YITU Healthcare Technology Co. Ltd, Hangzhou, China; Guangzhou Kangrui Co. Ltd, Guangzhou, China; and Guangzhou Regenerative Medicine and Health Guangdong Laboratory, Guangzhou, China.
Study funders included the National Key Research and Development Program of China; National Natural Science Foundation of China; Guangzhou Women and Children’s Medical Center, Guangzhou Regenerative Medicine and Health Guangdong Laboratory.