Does AI help medical artificial intelligence doctors get on the job? 【full text】

Security Exhibition Network Technology News One Minute Clinic originated in the United States as early as 2000, and it has taken shape in 2000. But at the time, the one-minute clinic was only able to treat pharyngitis, flu, pregnancy tests, cystitis, otitis, and sinusitis. In the clinic, a nurse practitioner or a physician assistant is the first patient. Doctor assistants and practicing nurses are different from ordinary nurses. They must have a master's degree and only obtain a license after rotation of relevant majors, such as the direction of a family doctor and the direction of pediatrics. After obtaining a license, you can see common diseases independently in more than half of the states, while in the remaining states you need to be treated under the supervision of a doctor. If the diagnosis is assisted by a practicing nurse or doctor, the preliminary judgment is that the clinic can treat the disease in one minute, and the patient can immediately undergo blood tests and other laboratory tests to confirm the diagnosis, and the prescription can be administered after the diagnosis. Because convenience clinics are usually opened in chain pharmacies or supermarkets that sell medicines, patients can easily take medicines. There is no medical doctor in the clinic, and patients will be recommended to go to the emergency department for problems that the clinic cannot handle.

Today, with the rapid development of technology, artificial intelligence has taken another step towards becoming an integral part of medicine in the 21st century.

Domestic one-minute clinic medical AI has been launched

When you walk into the consultation room, you can see a screen that provides man-machine dialogue, with a camera and microphone on it. Select "Start to ask the doctor" on the screen, and a voice announcement appears. Next, just like the doctor's consultation in the hospital, the voice broadcast questions one by one: "Hello, what is uncomfortable?" "How old is the patient?" "How long has it been uncomfortable?" Symptoms? "" How often does the attack happen? "... The question is very detailed, and there is a sense of temptation, just answer the microphone. After listening to the reporters answering in turn "uncomfortable throat", "some coughing", "no sputum", etc., a "General Medical \ Attending Doctor" appeared on the screen. After his judgment, both the screen and the voice indicated that they might have a cold , And gave advice on medication and rehabilitation.

By asking the employees of the company near the clinic, the employees said that they would not go to the hospital for diagnosis even if they were sick because the hospital was too far away.

American pediatric AI medical assistant is online

In the "Natural Medicine Express", it was proved that a natural language processing artificial intelligence is superior to novice pediatricians in diagnosing common childhood diseases. Like traditionally trained pediatricians, AI breaks down cases into major organ groups and infected areas (upper / lower respiratory tract, gastrointestinal tract, etc.), and then further subdivides them. It can then develop associations between various symptoms and organ groups and use these associations to improve its diagnosis. When comparing with human doctors, the study used 11,926 records from a group of unrelated children and gave a fair playing field comparison between MLC and 20 people. The results are clear: although the performance of the team of senior pediatricians is superior to artificial intelligence, the performance of artificial intelligence far exceeds that of junior pediatricians (pediatricians with 3 to 15 years of experience).

AI helps medical care, relying on "spectrum"

This "spectrum" is called the knowledge graph, which is to build the knowledge generated in the human world in the machine world, and then form a knowledge base that can support brain-like reasoning.

The construction of medical knowledge atlas must first represent the unstructured / semi-structured (Note: structured data is data that AI can directly use) data in the form of knowledge atlas, which includes transforming medical literature knowledge into medical knowledge atlas , Also includes transforming the empirical knowledge excavated in a large number of cases into a representation form that the machine can understand.

AI medical assistants in the United States, like Watson at IBM, use natural language processing in pediatric AI, essentially “reading” the written records of EHR books, just like human doctors viewing these records. This “assistant” has a knowledge base of 1,362,559 outpatient visits from 567,498 patients, which generated approximately 101.6 million data points for it to gain pediatric advantages.

And this AI doctor in China developed based on massive medical data, tens of millions of medical literature and clinical guidelines. Based on the Chinese medical knowledge map, at present, thousands of disease-assisted diagnosis models and adjuvant treatment models have been developed (among which hundreds of diseases support the recommendation of personalized medication programs). At the same time, it also provides auxiliary diagnosis from different perspectives, including auxiliary diagnosis based on the relationship between disease and symptoms; recommended inspection based on the relationship between disease and examination test; recommended medication based on the relationship between disease and drug and drug interaction; recommended medical treatment based on guidelines, literature and similar cases evidence.

Smart healthcare is closer to us, but still depends on us

Regardless of the final application, future AI doctors will gradually approach us. Research has proved that even in some complex and important decision-making processes, artificial intelligence can imitate the results of human deductive reasoning. Indeed, AI requires human input capabilities; the initial data points and the cases used to evaluate AI depend on the knowledge map written by the doctor. Although we make every effort to design a test mode to eliminate any signs of final diagnosis, some "data leaks" will inevitably occur.

In other words, when AI uses artificially created data, they will inherit human insight to some extent. However, advances in machine imaging, chatbots, sensors, and other fields have shown that this reliance on human input depends more on where we are now than what we might reach in the near future.

Looking to the future with data

There may also be some clear winners and losers in the near future. At present, these winners seem to be those that can capture and apply large data sets. With a rapidly digitized society collecting large amounts of data, China has a clear advantage. Combined with relatively loose privacy methods, it may continue to be one of the driving forces behind machine learning and its applications. Google / Alphabet will also conduct large-scale medical research. The data is the uranium in the AI ​​arms race. Everyone seems to be busy collecting more.

In a global community that seems to be increasingly aware of the potential problems arising from this need and reliance on data, it is good to know that it will also benefit. The technology behind AI medical assistants looks more and more mature, although we are still struggling to find the exact location, when and where and how this technology should be first popularized.

However, wherever we see our next efforts to make AI a standard tool in a real medical environment, we should have no doubt that it will greatly improve the lives of human patients. Today, AI doctors behave like a human counterpart with more than 10 years of experience. By next year or so, human competitiveness may take twice as long. Ten years later, the combination of all medical knowledge of human history may become a tool, as common as a stethoscope in the hands of a doctor.