An AI program developed in China that outlines test results, health papers, and even handwritten notes diagnoses childhood illnesses just like doctors, researchers said on Monday.
From flu and asthma to life-threatening pneumonia and meningitis, the system consistently coincides or transmits pediatricians to primary care, they report in Nature Medicine.
Dozens of studies in recent months have shown how AI revolutionized the detection of diseases, including cancer, genetic diseases and Alzheimer's disease.
AI-based technology learns and improves in a way similar to humans but has virtually unlimited capacity for data processing and storage.
"I believe he will be able to perform most of the positions the doctor does," says senior author Kang Zhang, a researcher at the University of California.
"But AI will never replace a doctor," he added, comparing the link to an autonomous car that remained under the supervision of a human driver.
"It will simply allow doctors to work better in less time and at lower costs."
The new technology, Zhang said, is the first one in which AI absorbs unstructured data and "natural language" to imitate the process by which the physician judges what is wrong with the patient.
"He can imitate a human pediatrician to interpret and integrate all kinds of medical data – patient complaints, medical history, blood and image tests – to diagnose," he said.
The system can be easily transferred to other languages and settings, he added.
By comparing hundreds of information about a patient with a huge amount of knowledge, technology reveals links that previous statistical methods – and sometimes physicians of flesh and blood – overlook.
At the last moment
To train the Concept Proofing System, Zhang and a team of 70 scientists injected more than 100 million data points from a 1.3 million patient visit to pediatrics at a large targeting center in Guangzhou, China.
The AI program diagnoses respiratory infections and sinusitis – a common sinus infection – with 95% accuracy.
Even more surprising, Kang said, he did so with less common illnesses: acute asthma (97%), bacterial meningitis and varicella (93%) and mononucleosis (90%).
Such technologies can only come in no time.
"Disease coverage, diagnostic tests and treatment options have increased exponentially in recent years, making the decision-making process for doctors more complicated," Nature said in a press release.
Experts who did not participate in the study said the study was yet another proof of the growing role of AI in medicine.
"Work has the potential to improve health care by helping physicians make swift and accurate diagnosis," said Duck Pham, a professor of engineering at Birmingham University.
"The results show that on average the system works better than junior doctors."
"But that will not replace clinicians," he added.
Machine training – which forms common rules from specific training examples – "can not guarantee 100% correct results, no matter how many training examples they use."
AI-based diagnostic tools abound, especially for the interpretation of machine-generated imaging such as MRI and CAT scan.
A method revealed last month in the United States to detect lesions that can lead to cervical cancer found precancer cells with 91% accuracy, compared with 69% for physician studies performed by doctors and 71% for conventional laboratory tests .
Similarly, an AI-based mobile phone application has performed experienced dermatologists to distinguish potentially cancerous skin lesions from benign ones, according to a study in the Annals of Oncology.