Medical patient bilingual translation system based on medical terminology library and context verification

Abstract

The application discloses a doctor-patient bilingual translation system and method based on medical terminology library and context verification. The system includes a terminology library engine, a neural translation engine, a verification module, a symptom alignment module, and a fusion module. The terminology library engine performs accurate matching translation on terms from standard medical terminology libraries such as ICD-11, SNOMED CT, and ATC; the neural translation engine performs overall translation on text; the verification module generates translation confidence through back-translation; the symptom alignment module converts patient colloquial descriptions into medical professional expressions; the fusion module outputs the results of the two engines, and triggers manual confirmation when the confidence is below the threshold. This scheme overcomes the technical defects of high misinterpretation rate, lack of confidence evaluation, and difficulty in mapping colloquial symptoms in existing medical translation.

Description

Technical Field

[0001] This application relates to the interdisciplinary field of natural language processing and medical information technology, specifically to a bilingual translation system for doctors and patients based on a medical terminology database and context verification. Background Technology

[0002] With the development of cross-border medical services, there is an increasing number of overseas patients seeking medical treatment in China or domestic patients traveling overseas for medical treatment. Language barriers between doctors and patients have become a key pain point. General machine translation models have a high error rate when processing medical terminology and cannot recognize ambiguities in context, potentially leading to misdiagnosis or medication errors.

[0003] Existing technologies include medical translation methods that use neural machine translation models to translate medical texts. However, these methods have the following shortcomings: (1) they are not connected to a standard medical terminology database, resulting in a high rate of mistranslation of professional terms; (2) they lack a secondary verification mechanism for the translation results; and (3) they do not adequately support the conversion between patients' colloquial symptom descriptions and doctors' professional terminology.

[0004] Existing technologies also include medical document translation methods based on preset dictionaries / term lookup tables, but these methods have the following shortcomings: they only support dictionary-level matching and are insufficient in handling complex medical expressions and context-related meanings; the lookup tables are costly to maintain and are slow to update.

[0005] In summary, the existing technology has the following technical problems: First, the mistranslation rate of medical terminology is high, which may lead to medical accidents; second, there is a lack of assessment and feedback on the confidence level of the translation; third, it is difficult to accurately map patients' verbal symptoms to medical professional expressions. Summary of the Invention Technical issues

[0006] To address the problems of high mistranslation rates, lack of confidence assessment, and difficulty in mapping spoken symptoms in existing medical translation technologies, this application provides a bilingual doctor-patient translation system based on a medical terminology database and context verification. Technical solution

[0007] This application provides a bilingual translation system for doctors and patients based on a medical terminology database and context verification. The system is executed by a computer system and includes: A terminology database engine, connected to a standard medical terminology database, including at least two of ICD-11, SNOMEDCT, and ATC; the terminology database engine is used to perform precise matching and translation of medical terms in input text; A neural translation engine, based on a finely tuned large language model from the medical field, is used to perform a holistic translation of input text; The verification module, connected to the neural translation engine, is used to perform back-translation on the output of the neural translation engine, compare the semantic similarity between the back-translation result and the original text, and generate translation confidence. The symptom alignment module is used to convert patients' colloquial symptom descriptions into medical professional expressions through a pre-trained symptom-term mapping model. The fusion module, connected to the terminology database engine, the neural translation engine, and the verification module, is used to fuse the outputs of the two engines based on the confidence level. When the confidence level is lower than a first threshold, manual confirmation is triggered.

[0008] The core collaborative innovation of this application lies in the following: the terminology database engine ensures the accuracy of professional terminology, the neural translation engine ensures the fluency of context, and the back-translation verification mechanism provides quantitative confidence as a safety gate. The three work together to form a closed loop of "accuracy + fluency + controllability" in doctor-patient translation. The overall effect of this application cannot be achieved by using any one engine alone. Beneficial effects

[0009] By setting up a dual-channel translation mechanism, the translation of technical terms benefits from both the precise matching of the standard library and the contextual understanding of the neural model, thus solving the problems of stiff pure dictionary translation and mistranslation in pure neural translation.

[0010] By setting up a back-translation mechanism to quantify the confidence level of each translation, potential mistranslations can be automatically identified, thus avoiding medical accidents.

[0011] By setting up a symptom alignment module, patients' colloquial expressions can be accurately converted into medical professional terms, thus solving the communication gap between doctors and patients.

[0012] By triggering manual verification when the confidence level falls below a threshold, zero mistranslation is ensured in critical scenarios, meeting the high accuracy requirements of medical settings. Detailed Implementation

[0013] The embodiments of this application are described in detail below.

[0014] Example 1: The system in this embodiment includes a terminology database engine, a neural translation engine, a verification module, a symptom alignment module, and a fusion module.

[0015] The terminology database engine interfaces with the ICD-11 and SNOMED CT international standard databases, storing approximately 350,000 medical terms and their multilingual mappings.

[0016] The neural translation engine uses a large language model finely tuned for the medical field and is trained for Chinese-English medical translation scenarios.

[0017] For the input "Patient complains of chest pain and shortness of breath": 1. The terminology engine identifies "chest pain" as = chest pain (R07.4) and "dyspnea" as = dyspnea (R06.0). 2. Neural translation engine output: "The patient complains of chest pain accompanied by dyspnea." 3. The verification module translates the neural translation result back into the Chinese text "The patient complained of chest pain accompanied by shortness of breath," and calculates a similarity of 0.92 with the original text. 4. The fusion module outputs the final translation with a confidence level of 0.92, which is higher than the threshold of 0.85, requiring no manual confirmation. Regarding the patient's verbal description, "My chest feels like it's being crushed by a stone": 1. The symptom alignment module identifies "chest pressure pain" as the medical term "chest pressure pain". 2. After being processed by the dual-engine system, the output reads: "The patient describes chest pressure-like pain." Example 2: This embodiment handles low-confidence scenarios: when the back-translation similarity of a rare disease term by the neural translation engine is only 0.62 (below the threshold of 0.85), the fusion module automatically triggers the human physician confirmation process, pushing the original text, terminology database results, and neural translation results to the online physician for review, and only outputs the final translation after confirmation.

[0018] The above embodiments are merely illustrative of the technical solutions of this application and are not intended to limit the scope of the application.

Claims

1. A bilingual translation system for doctors and patients based on a medical terminology database and context verification, wherein the system is executed by a computer system, characterized in that, include: A terminology database engine that connects to a standard medical terminology database, which includes at least two of ICD-11, SNOMED CT, and ATC. The terminology engine is used to perform precise matching and translation of medical terms in the input text; A neural translation engine, based on a finely tuned large language model from the medical field, is used to perform a holistic translation of input text; The verification module, connected to the neural translation engine, is used to perform back-translation on the output of the neural translation engine, compare the semantic similarity between the back-translation result and the original text, and generate translation confidence. The symptom alignment module is used to convert patients' colloquial symptom descriptions into medical professional expressions through a pre-trained symptom-term mapping model. The fusion module, connected to the terminology database engine, the neural translation engine, and the verification module, is used to fuse the outputs of the two engines based on the confidence level. When the confidence level is lower than a first threshold, manual confirmation is triggered.

2. The system according to claim 1, characterized in that, The terminology database engine further includes at least one of the ATC drug classification coding library and the CPT treatment item coding library.

3. The system according to claim 1, characterized in that, The semantic similarity calculated by the verification module is the cosine similarity of the sentence embedding vectors, which are generated by a multilingual pre-trained model.

4. The system according to claim 1, characterized in that, The symptom-term mapping model of the symptom alignment module is trained using historical doctor-patient dialogue data, which is de-identified internal data of medical institutions.

5. The system according to claim 1, characterized in that, The first threshold value of the fusion module is 0.85; when the confidence level is lower than 0.85, the translation result and the candidate translation are submitted to the human physician for confirmation.

6. A method for bilingual translation between doctors and patients based on a medical terminology database and context verification, wherein the method is executed by the system described in any one of claims 1 to 5, characterized in that, Includes the following steps: T1: Receives input text, identifies whether it contains the patient's colloquial symptom description, and if so, converts it into medical professional expression through the symptom-term mapping model. T2: Feeds the text into the terminology database engine, where it performs precise matching and translation of medical terms; T3: Feed the text into the neural translation engine and perform overall translation; T4: Perform back-translation on the output of the neural translation engine, and calculate the semantic similarity between the back-translated result and the original text as the confidence score; T5: Merge the outputs of the terminology engine and the neural translation engine based on confidence level; When the confidence level falls below the first threshold, manual confirmation is triggered.