Method for realizing cross-mechanism physical examination report examination item standardization based on large model

Abstract

The invention discloses a method for realizing standardization of examination items of a cross-mechanism physical examination report based on a large model. The examination item names in the physical examination reports of different institutions are reasoned and judged through the large language model, and effective understanding and standardized mapping of medical text data from different sources are achieved. Finally, the unification of the same index names of different hospital institutions can be realized, the unification is not limited to the knowledge base range and the dependence on codes, part of manpower is replaced by a large language model, and the labor cost is saved.

Description

Technical Field

[0001] The present invention relates to a method for standardizing text data of examination items in physical examination reports of different institutions based on a large language model, and belongs to the field of artificial intelligence natural language processing. Background Art

[0002] With the aggravation of the aging problem and the increasing attention paid to health by people, it has become particularly urgent to build a medical and health big data platform, and the standardization of the names of the examination items in the cross-institutional physical examination reports is an extremely critical step. Different hospitals have multiple ways of expressing the names of the examination items, and the coding of the same examination items is not completely consistent, resulting in prominent data integration problems. This problem seriously hinders the circulation and sharing of data, making it difficult to effectively carry out subsequent big data analysis and application, affecting the decision-making efficiency and business development of the medical industry. Standardization of examination item names not only helps to promote data sharing between hospitals, improve the authority and systematicness of physical examination reports, and ensure the accuracy and consistency of physical examination results, but also makes it possible to compare the physical examination report results of the same patient in different institutions horizontally. In addition, standardized physical examination data can also promote the construction of laws and regulations and standards systems, improve the availability and reliability of data, and provide a basis for the standardization of health and medical big data products and service processes.

[0003] Traditional manual methods of data standardization are too inefficient and costly. Methods based on database similarity functions are also limited in efficiency and accuracy, and it is difficult to integrate expert experience. In recent years, large language models (LLMs) have provided a new way to standardize the names of examination items in physical examination reports due to their outstanding performance in natural language processing (NLP) tasks. Through pre-training, the large language model's ability to understand medical texts has been improved, and it has cross-domain generalization. The multi-task learning and knowledge transfer capabilities of the large language model are crucial for unifying the names of examination items in different physical examination reports and test reports. In addition, the instruction-following and contextual learning capabilities of the large language model enable it to accurately identify and standardize the names of examination items, thereby improving the efficiency and accuracy of data integration and promoting data sharing and analysis among medical institutions.

[0004] The patent with application number CN202410402366.5, which was published on June 4, 2024, involves an AI-based diagnostic text standardization method. First, the platform's diagnoses and surgeries are stored in different mapping knowledge bases according to data types. Then, the fast recall method that integrates the Trie dictionary tree + BM25 fast document retrieval algorithm is used to quickly obtain a recall set that is very similar to the semantics of the diagnostic text to be standardized from the platform's standardized diagnostic text. The purpose is to quickly and automatically match the longest prefix and retrieve all ICDs and diagnostic names under the prefix to determine the standardized diagnostic text on the platform. This method is limited to the scope of the knowledge base and its reliance on ICD codes. If the ICD code is missing, randomly filled in, or incorrectly filled in, it may affect the accuracy of the standardized mapping.

[0005] The patent with application number CN202311372303.1 published on January 19, 2024 involves a method for standardizing clinical information text. First, the Chinese text of the clinical information to be converted is converted from Chinese to English and text processed to obtain a set of medical English vocabulary. Then, the medical English vocabulary set is retrieved using the target hash table to obtain the first target medical English vocabulary with successful string matching and the corresponding first standard human phenotype ontology term. The purpose is to achieve the standardization of Chinese text of clinical information. This method is limited to the accuracy of the pre-trained language model and the coverage of the target hash table, which will affect the accuracy and comprehensiveness of the standardization. Summary of the invention

[0006] The technical problem to be solved by the present invention is: the limitations in handling the task of standardizing the names of examination items in cross-institutional physical examination reports, such as the ability to recognize and standardize specific medical terms is limited by the reliance on existing knowledge bases and coding coverage, as well as reliance on manual resources.

[0007] In order to solve the above technical problems, the technical solution of the present invention is to disclose a method for standardizing examination items in cross-institutional physical examination reports based on a large model, which is characterized by comprising the following steps:

[0008] Step 1: Design prompt words according to specific standardized task requirements, including:

[0009] Explicitly prompt the large language model to standardize the names of inspection items;

[0010] Explicitly specify the output format;

[0011] Clarify the reasoning method of large language models according to requirements;

[0012] Step 2: Input the large language model for inference and judgment, prompting the large language model to output the final standardized result, which specifically includes the following steps:

[0013] Large language model reasoning: The data to be standardized and the prompt words are input into the large language model, and the preprocessed inspection item names are inferred and judged to generate the most correct standardized names;

[0014] Extract standardized names: Clean the output of the large language model, formulate regular expressions based on the design of prompt words to extract standardized inspection item names, and unify the format.

[0015] Preferably, after step 2, the method further includes:

[0016] Step 3: Further improve the effect through the auxiliary enhancement module, which specifically includes: accessing the standard mapping database, recalling several names with the closest semantics as options through text similarity, and combining them with the prompt words and inputting them into the large language model to select the most correct name as the final result.

[0017] Preferably, step 3 comprises the following steps:

[0018] Standardized mapping database: Establish a standardized mapping database to store standardized examination item names and corresponding department information, which can be connected to the large language model and cross-institutional medical big data platform;

[0019] Similarity calculation: Use the text similarity algorithm to traverse and compare the names to be standardized with the names in the inspection item name standard mapping database to find the TopK most similar standard inspection item names;

[0020] Combined with prompt words: The recall result and the prompt words are combined and input into the large language model for reasoning, and finally the most likely standardized name is generated.

[0021] The present invention proposes a method for standardizing the names of examination items in cross-institutional physical examination reports based on a large language model, which realizes intelligent matching and standardization of the names of examination items in physical examination reports, aims to improve the consistency and interoperability of medical information and promote the sharing and analysis of medical data.

[0022] By utilizing the large language model's in-depth understanding and processing capabilities of medical texts, compared to the traditional database similarity function method, the present invention can more accurately identify and standardize the names of examination items in physical examination reports of different institutions, improving the accuracy and efficiency of standardized mapping. Through text similarity evaluation, the present invention can recall the K examination item names with the highest semantic similarity to the data to be standardized, and input them into the large model for selection and judgment, further enhancing the effect. Compared with methods that rely on specific encodings, this method reduces dependence on encoding integrity, improves the robustness of standardization, and is more flexible and generalizable.

[0023] The technical solution of the present invention does not rely on the scope of the existing knowledge base and the limitations of coding, improves the accuracy, coverage, robustness, flexibility and reliability of standardization, while reducing labor costs, improving data processing efficiency, and providing strong technical support for the construction of a medical and health big data platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0025] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0026] The present invention proposes a technical solution for standardizing the names of examination items in physical examination reports based on a large language model, aiming to improve the efficiency and accuracy of standardization through a large language model while reducing labor costs. The core of the present invention is to use the large language model's ability to understand and process medical texts to achieve intelligent standardization of examination item names.

[0027] Specifically, the technical solution disclosed in the present invention first designs prompt words according to the specific standardization task requirements, and then inputs the large language model for reasoning and judgment, prompting the large language model to output the final standardization result. The effect can be further improved through the auxiliary enhancement module, which specifically includes accessing the database of standard mapping, recalling several names with the closest semantics as options through text similarity, and combining them with the prompt words and inputting them into the large language model to select the most correct name as the final result.

[0028] Specifically, the present invention comprises the following steps:

[0029] 1. Design tips:

[0030] The present invention uses a large language model to standardize the names of examination items in physical examination reports from different medical institutions. In order to enable the large language model to clarify the standardized mapping process, prompt words need to be designed according to specific task requirements and meet the following requirements.

[0031] (1) Clearly prompt the large language model to standardize the names of inspection items;

[0032] (2) Clearly specify the output format, for example, output the final standardized results in JSON format and mark them with specific symbols to facilitate post-processing extraction;

[0033] (3) Clarify the reasoning method of the large language model based on needs, such as directly asking or providing a range for selection.

[0034] 2. Reasoning and judgment of large language models:

[0035] Reasoning and standardized judgment are performed through large language models, which are not limited by existing knowledge bases and encodings.

[0036] (1) Large language model reasoning: The data to be standardized and the prompt words are input into the large language model, and the preprocessed inspection item names are inferred and judged to generate the most correct standardized names.

[0037] (2) Extracting standardized names: Clean the output of the large language model, formulate regular expressions based on the design of prompt words, and extract standardized inspection item names in a unified format.

[0038] 3. Auxiliary enhancement module:

[0039] Depending on the complexity of the task, by selectively using auxiliary enhancement modules, the large language model can convert standardized questions about the names of examination items in the physical examination report into multiple-choice questions to improve the accuracy.

[0040] (1) Standardized mapping database: A standardized mapping database is established to store the standardized examination item names and corresponding department information, which can be connected to the large language model and the cross-institutional medical big data platform.

[0041] (2) Similarity calculation: Use text similarity algorithms, such as cosine similarity or Jaccard similarity, to traverse and compare the names to be standardized with the names in the standard mapping database of inspection item names, and find the top K most similar standard inspection item names. For example, if there are 10,000 standard names in the database, obtain the top 50 standard names with the highest text similarity to the data to be standardized.

[0042] (3) Combined with prompt words: The recall result and prompt words are combined and input into the large language model for reasoning. For example, the above K standard inspection item names are used as options. When designing prompt words, the large model is prompted to answer multiple-choice questions and choose from the K options, and finally the most likely standardized name is generated.

[0043] The present invention uses a large language model to infer and judge the names of examination items in physical examination reports of different institutions, and realizes effective understanding and standardized mapping of medical text data from different sources. Ultimately, the present invention can unify the names of the same indicators of different hospitals and institutions, without being limited to the scope of the knowledge base and dependence on coding, and replaces part of the manual work with a large language model, saving labor costs.

Claims

1. A method for standardizing examination items in cross-institutional physical examination reports based on a large model, characterized in that: The following steps are involved: Step 1: Design prompt words according to specific standardized task requirements, including: Explicitly prompt the large language model to standardize the names of inspection items; Explicitly specify the output format; Clarify the reasoning method of large language models according to requirements; Step 2: Input the large language model for inference and judgment, prompting the large language model to output the final standardized result, which specifically includes the following steps: Large language model reasoning: The data to be standardized and the prompt words are input into the large language model, and the preprocessed inspection item names are inferred and judged to generate the most correct standardized names; Extract standardized names: Clean the output of the large language model, formulate regular expressions based on the design of prompt words to extract standardized inspection item names, and unify the format.

2. A method for realizing cross-institutional physical examination report inspection item standardization based on a large model as claimed in claim 1, characterized in that: After step 2, the method further includes: Step 3: Further improve the effect through the auxiliary enhancement module, which specifically includes: accessing the standard mapping database, recalling several names with the closest semantics as options through text similarity, and combining them with the prompt words and inputting them into the large language model to select the most correct name as the final result.

3. A method for realizing cross-institutional physical examination report inspection item standardization based on a large model as claimed in claim 1, characterized in that: The step 3 comprises the following steps: Standardized mapping database: Establish a standardized mapping database to store standardized examination item names and corresponding department information, which can be connected to the large language model and cross-institutional medical big data platform; Similarity calculation: Use the text similarity algorithm to traverse and compare the names to be standardized with the names in the inspection item name standard mapping database to find the TopK most similar standard inspection item names; Combined with prompt words: The recall result and the prompt words are combined and input into the large language model for reasoning, and finally the most likely standardized name is generated.

Images