Recently, in order to fully implement the Healthy China Strategy and implement the "Opinions of the General Office of the State Council on Promoting the Development of "Internet + Medical Health"", the National Medical and Medical Administration issued the "Informationization Construction of Medical Institutions with the Core of Electronic Medical Records as the Core" Notice of work.

The "Notice" mentioned that it is necessary to play the role of clinical diagnosis and treatment decision support. Encourage medical institutions to embed clinical pathways, clinical diagnosis and treatment guidelines, technical specifications and medication guidelines into information systems in the construction of electronic medical records, to improve the standardization of clinical diagnosis and treatment. And pointed out that by 2020, it is necessary to achieve a graded evaluation of level 4 or above, that is, to achieve information sharing in the hospital, and to have medical decision support functions.

To develop clinical decision-making, the digitization and intelligence of electronic medical records have become inevitable. Among them, in order to realize the structural calcification of data, the importance of natural language processing tasks such as Named Entity Recognition (NER) can be highlighted.

It is reported that the NER of the electronic medical record is to identify and extract the physical reference related to the medical clinic through the plain text document of the established electronic medical record, and combine the content and characteristics of the data source "current medical history record" into the predefined category.

It is not only the first step in text mining, but also an important tool in the biomedical field, which can be applied in many fields, such as medical literature, online medical communities, and electronic medical records. Therefore, the construction of the open dataset of the electronic medical record NER can achieve the effect of “one arrow and many sculptures”—helping to structure and standardize the medical entity, and completing the tasks of medical entity relationship extraction and medical knowledge mapping construction.

Internationally, there have been a number of NER public evaluation and annotation data sets for English electronic medical records, including I2b2, ShARe/CLEF eHealth and SemEval, but in China, this evaluation is still blank.

In order to promote the development of Chinese electronic medical record related research, to fill the gap in the domestic electronic medical record NER evaluation competition and the annotation data set, Zhiduyun and Tsinghua University Knowledge Engineering Laboratory and Harbin Institute of Technology jointly organized the "name for Chinese electronic medical records." Physical identification" project evaluation.

According to Yan Jun, chief artificial intelligence scientist at Medical Duyun, under normal circumstances, medical data can be directly used in clinical applications without being in the hospital. The original intention of clinical medical record design is record-oriented rather than research-oriented. In other words, the doctor will record all the clinical conditions originally, but there is no processing for research and application.

As a result, most of the clinical electronic medical records are natural language. And this kind of text information can't be calculated in any way in the computer, so the first thing to do is data structuring.

There are many manufacturers of domestic information systems, not only the standards adopted, but also the writing habits and expression habits of doctors in each hospital. The name of some diseases, there are even hundreds of expressions in the hospital. Therefore, to present data from the entire paragraph of natural language text, if there is no technical support, it requires huge manpower input.

"Medical Cloud held the evaluation of the "Named Entity Recognition for Chinese Electronic Medical Records" project, the original intention is also here." Yan Jun said.

In order to overcome the difficulties, Zhidu Cloud first carried out the “roughing and refining” of the data: on the one hand, directly in the pre-defined categories, the key points of this evaluation—the medical entity mentioned, the starting and ending position identification and the predefined category. Combing, make the "station" of massive data in an orderly manner; on the other hand, in the details of "excellence", there are five aspects in the predefined categories: independent symptoms, symptom descriptions, anatomy, drugs and surgery. The specific data information is as follows:

In order to ensure the professionalism and authority of the evaluation, Zhidu Cloud also organized a professional doctor team to back up the data collation and labeling of the project evaluation. In order to ensure data security, the 118 teams that signed up for the competition clearly indicated that the data is limited to the CCKS 2018 competition evaluation.

Why is such an evaluation going to be carried out in China today? For this problem, Jiao Zengtao, a natural language processing expert in the medical and artificial cloud laboratory, also explained: "This is not unrelated to the technical difficulties. There are two difficulties in the following: First, most of the symptoms are due to the symptom type entities. Structured form; Second, some unique expressions of medical terms make the computer often "catch" in the identification and reading, which makes it difficult to organize and classify medical terms."

Therefore, if we can overcome the above difficulties and solve the problem of the lack of open resources available for the current Chinese electronic medical record NER, the value can be more reflected. Although the evaluation task is based on the sensitivity of medical data, all data are simulated by professional doctors, but the scientific analysis of sensory data simulation and statistical significance data has been strictly tested.

Behind the NER review reflects the importance of standardization of medical data. In addition to naming issues, quality control of data is equally important.

In the process of structured processing of hospital texts, Medical Duyun found that the quality of medical data is actually not high. Although the medical and labor cooperation hospitals are among the top 150 hospitals in China, their data still has many unqualified and inaccurate places.

In order to solve these problems, Zhidu Cloud spent three and a half years focusing on the development of a highly integrated "Medical Data Intelligence Platform" (DPAP), which can turn raw scattered uncalculated data into high-quality computable data. The platform has a large collection of knowledge maps, more than 300 intelligent processing modules and more than 20 specialized disease banks.

By integrating data from scattered data, DPAP builds a patient's timeline module to model disease data and complete disease data modeling. From a disease perspective, DPAP can also provide disease data models. Whether it is the disease data model or the patient diagnosis and treatment model, this is the basis for clinical research, path mining, efficacy evaluation, and auxiliary diagnostic applications.

In this process, Zhidu Yun also carries out strong quality control on the production of the entire data. By establishing a knowledge base of medical common sense and normalization processing, different statements in the hospital are mapped to the same standard.

But the question is, how is the standard set?

At this stage, organizations in the medical industry , including government, academia, civil society, and businesses, are trying to define standards. But the challenge is how to get the industry to follow the standards once they are developed.

Yan Jun said: "Medical cloud is not willing to wait for the standard to be produced, so we try to set data standards with many experts and hospitals. In addition, we think that the more effective way is to not promote the standard through the market. Instead, we work with hundreds of top three hospitals to help each of their hospitals improve their data quality."

This means that no matter which hospital, which standard is used, it will form a mapping with the doctor's own standards. As long as the mapping relationship exists, Zhiduyun can realize clinical multi-center research, and only need to open an interface under the premise of hospital authorization, all cooperative hospitals can cooperate on one platform.

Only by doing a good job of data services will there be an opportunity to form a standard that everyone recognizes in the subtle.

The definition of the quality of the standard, Yan Jun believes that can be seen from two ways: whether one can really bring the actual value of the landing, and second, whether anyone is willing to follow. Zhiduyun hopes to be able to put its own specialized products in the field of scientific research, not only to empower the clinical department, but also to promote consensus and resonance between the hospital information department and the clinic.

In addition, from the perspective of talent structure, if you really want to form a standard, Yan Jun believes that medical experts and computer experts must be included.

Taking knowledge map as an example, the knowledge and experience of experts is the basis of knowledge map, and the law of data is the category of machine learning. Therefore, the integration of the two universities is not only the talents pursued by big data and artificial intelligence enterprises. The combination of structures is also the inevitable integration of the final formation of industry standards.

The result of standardization is to help hospitals produce high-quality disease-specific data, which is also an important value of big data companies like Zhiduyun. In the process of building a special hospital for many hospitals, Zhiduyun does not provide data for the hospital, but only delivers it as a processing processor for the data.

At work, I will experience the production of many specialized disease banks and experience many iterations. Medical Duyun must first process and produce according to the doctor's interpretation of the data. In the end, the output data should be docked and corrected with the doctor. The production of high-quality disease data requires a lot of processes. “Doctors have a deeper understanding of the data, and the company has learned a lot of medical knowledge from this process. This is a process of common progress.” Yan Jun said.

In many projects, the medical cloud has to do more. For example, a specialized disease database has a large number of papers in medical academia. In addition to evaluating the quality of data and the degree of structuralization, Medical Duyun depends on whether this data can reproduce some of the previous research papers and achieve its stated effect. In this way, the doctors will verify the delivery level.

“In these years, the core of the medical cloud is the construction of the 'medical brain', on the one hand, artificial intelligence technology, and on the other hand, the construction of medical knowledge maps. Artificial intelligence is inseparable from real world data and the latest medical paper research results. The support, the combination of the formation of knowledge points, is the key to building a medical brain." Medical doctor cloud CTO Xu Jiming told reporters.

It is understood that currently Duduyun has established strategic cooperation with more than 700 medical institutions, including 100 top medical institutions in the top150 nationwide, integrating medical data integrating more than 300 million patients and 1.3 billion people in hospitals. In addition, Zhidu Cloud has established nearly 30 high-quality special disease banks, and it is still increasing every year.

In the country, Medical Duyun has produced nearly 20 domestic/international journal articles with medical institutions. Through powerful technology to process data, Zhidu Cloud provides a theoretical basis for cooperative institutions to assist clinical decision-making and improve efficiency by establishing standards, integrating data, cultivating medical brains, and providing scientific research inspiration.

Protein Powder

Protein Powder,High Protein Powder Supplement,Natural Protein Supplements,Vegetable Protein Powder

Shaanxi Changsheng Industrial Co., Ltd. , https://www.sxcsmalehealth.com