Artificial Intelligence 'Trained' to Diagnose Medical Conditions

"In some situations, physicians cannot consider all the possibilities. This system can spot-check and make sure the physician didn’t miss anything."
"The sheer size of the population [in China] — the sheer size of the data [advantages researchers in China in collecting and assessing data] — is a big difference."
Dr. Kang Zhang, chief of ophthalmic genetics, University of California

"You have go to multiple places. The equipment is never the same. You have to make sure the data is anonymized."
"Even if you get permission, it is a massive amount of work."
Dr. George Shih, associate professor of clinical radiology, Weill Cornell Medical Center

"Medicine is a slow-moving field."
"No one is just going to deploy one of these techniques without rigorous testing that shows exactly what is going on."
Ben Shickel, researcher, University of Florida

Doctors competed against A.I. computers to recognize illnesses on magnetic resonance images of a human brain during a competition in Beijing last year. The human doctors lost.  CreditMark Schiefelbein/Associated Press

 

 

A collaboration between researchers in the United States and China has resulted in the testing of a potential corrective for the frequent inability of humans to find solutions, to adequately assess data, and to produce correct diagnoses. This is where artificial intelligence comes in, in its ability to do all these things free of human frailties, depending only on an elite mechanistic influence that calibrates and assesses and reaches logical conclusions.

System have been built by scientists capable of automatically diagnosing common conditions afflicting children after the patient's symptoms, history, lab results and other clinical data has been processed; the system producing highly accurate conclusions. A system that holds out great promise as one day being viewed as sufficiently dependable to act as a huge assist to doctors in diagnosing complex or rate conditions whose symptoms elude many doctors.

The records of close to 600,000 Chinese patients who had visited a pediatric hospital over a period of a year and a half were drawn upon for the data useful in training the new system. The very reality of the most populous country on the planet producing logistics to enable this type of research speaks volumes of the advantages to Chinese researchers using their own population numbers as a base for their research.

Not only is the Chinese population enormous, but the state itself fails to involve itself in the kind of human rights concerns exerted in Western democracies; the protection of health and other types of personal information; with fewer restrictions on sharing digital data, it becomes simpler for Chinese researchers to engage in designing "deep learning" systems for health care which in turn advantages China in general in the global race to achieve the pinnacle of artificial intelligence.

Systems are being developed by many groups to analyze electronic health records to flag medical conditions like osteoporosis, diabetes and heart failure. Technologies with similar aims are being built for the purpose of detecting signs of illness in X-rays, M.R.I.s and eye scans, the systems reliant on neural networks, a type of artificial intelligence capable of 'learning' tasks independently through analyzing vast tracts of data.

Dr. Kang Zhang, chief of ophthalmic genetics at University of California, San Diego, has been involved in building systems to analyze eye scans for the presence of hemorrhages, lesions and related signatures of diabetic blindness with a view to serving as a first line of defence, screening patients and identifying who among them would require additional attention.

A newer system developed by Dr. Zhang's laboratory is one capable of diagnosing a wider range of conditions through the recognition of text patterns, not merely in medical images alone. A report by the scientists involved was published in the journal Nature Medicine.

Medical records of close to 600,000 patients at the Guangzhou Women and Children's Medical Center in southern China were utilized for the experimental system to analyze, in the process 'learning' to associate common medical conditions with specific patient information gathered by doctors, nurses and allied technicians. Initially, physicians annotated hospital records, identifying information by labels related to specific conditions which the system then analyzed.

New information was then provided including patients' symptoms, enabling the device to make connections between records and symptoms. The software, when tested on unlabeled data, rivaled the performance of experienced physicians, its diagnoses of asthma over 90 percent accurate, while physician accuracy ranged from 80 to 94 percent.

Neural networks can be extremely powerful in their capacity to recognize patterns in data which human intelligence might never identify on their own, but experts too, experience difficulty understanding why the systems make specific decisions and how they 'teach' themselves.