Kang Fuzi Zhang Chao: How is the "medical brain" made?

If you look at the doctor's day from the perspective of God, you will find that the doctor actually has to do a lot of repetitive work.

Repetitive work can be embodied in three areas: interaction, discovery, and decision-making, and these can all be machines:

At the interaction level: The doctor interacts with the patient at the time of the inquiry, and the patient's medication guidance, patient tracking, etc.

Level of discovery: mechanical repetitive labor of insurance review personnel in the examination of medical insurance documents, and proofreading of doctors' treatment plans;

Decision-making level: The doctor gives the diagnosis result according to the patient's symptom and auxiliary examination, and gives the treatment suggestion based on the diagnosis result.

On the whole, medicine is a knowledge-driven discipline. If we collect enough and reliable enough knowledge from multiple dimensions, we can play a huge role in assisting decision-making and free medical personnel from complicated and repetitive tasks. Go on to do more creative things.

If we can use artificial intelligence to help doctors correspond to symptoms and diseases in vast medical knowledge, and even assist doctors in making decisions about remedying the disease, it will undoubtedly improve the optimization efficiency of medical resources. Zhang Chao, the CEO of Kang Fuzi, did exactly this kind of “medical brain”—using artificial intelligence technology to automatically learn knowledge, construct knowledge maps, and implement knowledge reasoning from the medical literature, and then describe the general public’s description of symptoms and understanding of diseases. Accurately map on the serious medical plane.

Wallpaperup

Use AI to help users organize information

Beijing Kangfuzi Technology Co., Ltd. CEO Zhang Chao

Zhang Chao has worked for Baidu for five years and is a senior R&D engineer in the Natural Language Processing Department and responsible for text knowledge mining. Dealing with netizens' search behavior has become the norm in his work.

By chance, Zhang Chao found that Internet users often look at the results of many pages on Baidu, but the contents presented are actually mixed. As a result, Zhang Chao produced artificial intelligence to replace the effective information of netizens in order to improve the user experience.

At that time, Zhang Chao aimed at a population of pregnant women, the group most valued by nutrition, to build knowledge maps and introduced a function of “analysis of dietary records”. Through the daily diet records of pregnant women, the user can determine whether the nutritional intake is balanced and then obtain more scientific dietary guidance.

The idea is very good, but it has little effect in practice. Zhang Chao told Lei Fengwang (search for "Lei Feng Net" public concern) that after the product went online, less than 20% of pregnant women would try to record their own diet, and less than 8% of pregnant women could record continuously for 1 week.

Zhang Chao learned a lesson from this venture: Although pregnant women's demands for nutrition are highly concerned, at this stage, netizens are still at the stage of “Ye Gong Hao Long” for nutrition. "Nutrition is not just needed. To a certain extent, it is also characterized by anti-humanity."

Some B-side companies learned about the knowledge map that Kang Fuzi is doing and hope to use their services in a paid form. Taking into account the relative "nutrition", "medical" pain points more prominent, but also let Zhang Chao began to want to do just need to want to do "we are the best thing."

All this starts with the repetitive work of optimizing medical behavior.

How is the "medical brain" made?

In March this year, Kang Fuzi began to provide knowledge and technical services to the B-side, and they also turned from the original "catering advice provider" to "medical artificial intelligence technology, knowledge, and data providers."

The knowledge map is composed of two parts at the data presentation level: firstly, the entity node, and secondly, the relationship between the entities. For example: disease name, symptom name, drug name, and laboratory test data are physical nodes. Each node has a one-to-one relationship with the node, such as the relationship between disease and symptoms, and what kind of disease needs to be used. What tests do drugs and diseases need to do? In the actual diagnosis, more correspondences will be involved.

Kang Dafu's knowledge maps are mainly taught by the church computer, after reading a large amount of text, automatically giving a description of the rules of writing a certain knowledge, and carrying out large-scale knowledge extraction (Information Extraction).

Data extraction is mainly divided into four steps:

1. Data preprocessing, for example, tasks such as data cleaning, word segmentation, proper name recognition, and referential elimination;

2. Pattern learning, based on hundreds of millions of texts, automatically discovering the writing patterns of these texts;

3. Knowledge extraction, second extraction of hundreds of millions of texts in a pattern that continues to be learned;

4. Enhanced learning + Bootstrapping, based on some annotation data and judgment criteria, repeat points 2 and 3 continuously, and give result data with an accuracy rate of more than 99%.

With a complete knowledge map, it is still far from enough. The system must have reasoning ability to realize intelligent diagnosis. In the Confev subsystem, this is reflected in two aspects. The first is the representation of knowledge vectorization. This step is also a key step in building a bridge for colloquial medicine and literature terms. That is, "translating" popular sentences described by patients into professional terms (such as the corresponding relationship between "cramps" and "sputum"); secondly, judging the weights of multiple symptoms under the condition of comprehensiveness and disease, and coordinating diagnostic models to improve diagnosis. The hit rate.

What's different from the construction of other knowledge maps is that the two characteristics of the scattered and unstructured medical knowledge make it more difficult to construct the medical map. For example, in the field of entertainment, from the entertainment sites and encyclopedic knowledge, it is possible to rapidly discover the works of stars and their spouses and other knowledge relations. However, in the medical field, it is necessary to read a large number of authoritative documents and extract them in order to achieve the desired results.

Another point is the logic application. Doctors are immersed in the complexity of knowledge barriers and medicines. In the case that diagnosis cannot be comprehensive, Kang Kang can interact and further judge the illness according to the patient's answer.

The main business content of Kangfuzi mainly includes API service and clinical assistant decision-making. According to Zhang Chao, Kang Fuzi has completed the construction of a knowledge map of drugs and currently covers nearly 300,000 drugs. The accuracy of the typical symptoms of 100 common diseases has exceeded 90%, and the typical symptoms of 4000 hot diseases are The hit rate is also more than 80%.

Although the diagnostic effect has surpassed most of the general practitioners, the Canf subsystem still cannot complete the diagnosis independently. In addition to the ethical factors, the diagnostic complexity of the actual scenario also makes artificial intelligence technology currently only able to handle specific tasks. This also made Zhang Chao's idea of ​​serving the B-side. "We don't directly serve patients, but patients can get services through our B-side partners."

In Zhang Chao's understanding, as a 60-year subject, artificial intelligence applications for specific tasks will be integrated into everyone's life in the next 3-5 years. The logic drive represented by autopilot and Go, and the knowledge driven represented by medical care are the two “engines” of artificial intelligence. Of course, in addition to continue to improve the knowledge map in the field of intelligent diagnosis, the Chao also hopes to knowledge-driven medical intelligence to better serve humanity in logic level, such as virtual assistants, medication mining, intelligent diagnostic program, so that the doctor really Freed from complicated labor.

Lei wears the public number of Lei Feng Net and focuses on sports health products and technologies. If you want to talk about your own business story, you can add Xiaobian Wechat 417423625, or send an e-mail to.

Silver-zinc Rechargeable Battery

Silver Zinc Rechargeable Battery,Battery For Aircraft Standby Power,Professonal Silver Zinc Battery,22.5V 45Ah Silver Zinc Battery

Henan Xintaihang Power Source Co.,Ltd , https://www.taihangbattery.com