Large model-based admission record generation method and device

Through the large-scale admission record generation method and fine-tuning technology, the problem of inefficient entry of hospital admission records is solved, and the rapid and accurate admission record generation is achieved, which improves work efficiency.

CN119007912APending Publication Date: 2024-11-22SHANDONG LANGCHAO YUNTOU INFORMATION TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411002565.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-25
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The entry process of hospital admission records depends on the doctor's professional knowledge and a lot of time, resulting in a long and inefficient entry cycle, which becomes a bottleneck in the doctor's work.

Method used

The admission record generation method based on the big model is adopted, and the admission record is converted into text through doctor-patient dialogue, and the admission record is generated using the big model fine-tuning technology, and the model is optimized through the scoring evaluation mechanism until satisfactory results are achieved.

Benefits of technology

It realizes the rapid and accurate generation of professional admission records, saves doctors and patients' time, and improves the work efficiency of the entire process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119007912A_ABST
    Figure CN119007912A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of medical treatment, and particularly provides a hospital admission record generation method and device based on a large model, and the method comprises the following steps: S1, doctor-patient dialogue and inquiry generation; s2, marking a dialogue, and executing fine tuning; s3, verification and scoring evaluation are executed; and S4, judging whether the evaluation effect is satisfied or not, and deciding to put into operation or continue fine adjustment. Compared with the prior art, the time of doctors and patients can be saved, and the working efficiency of the whole process is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and specifically provides a method and device for generating admission records based on a large model. Background Art

[0002] Large models and large speech models usually refer to deep learning models with a large number of parameters, while fine-tuning refers to retraining the pre-trained large models for tasks in specific fields, such as generating admission records based on doctor-patient conversations in the medical field.

[0003] Large Models Parameter scale: Large models are composed of billions or even hundreds of billions of parameters, which enable the models to capture and learn complex patterns and features in the data.

[0004] Building structure: They are usually based on deep neural networks and contain multiple layers or modules, each with a large number of neurons and connections.

[0005] Generalization ability: Due to the exposure to massive amounts of data during training, large models have strong generalization capabilities and can make accurate predictions on a variety of tasks.

[0006] Application areas: Large models are used in natural language processing, computer vision, speech recognition, and recommendation systems.

[0007] Fine-tuning of large models for performance optimization Fine-tuning is to further train the pre-trained large model using data sets in a specific field to improve the model's performance in tasks in that field.

[0008] Customized functions: A general large model may not perform well in a specific field. Fine-tuning can give the model more customized functions to adapt to the needs and characteristics of a specific field.

[0009] Domain knowledge learning,During the fine-tuning process, the large model can learn domain-specific knowledge and language patterns,,which helps to achieve better results on specific tasks.

[0010] Hyperparameter Tuning,Fine-tuning is not only the adjustment of model parameters, but also the optimization,of hyperparameters to obtain the best performance.

[0011] The difference between the two is that the large model focuses on improving the overall performance and generalization ability through large-scale data and complex structures, while fine-tuning further refines and optimizes the performance of the model on specific tasks based on the capabilities of the large model.

[0012] In short, large models are applicable to a wide range of fields due to their strong generalization ability and high flexibility, while fine-tuning techniques enable these models to better adapt to specific application scenarios and task requirements.

[0013] The current hospital admission records include: chief complaint, current medical history, past medical history, personal history, marital history, menstrual and reproductive history, family history, physical examination and auxiliary examinations.

[0014] The admission record business occurs during the doctor-patient conversation. Currently, after the consultation, the doctor usually recalls the content of the doctor-patient conversation or uses voice conversion technology to obtain the conversation text in the doctor-patient vernacular, and then manually organizes and fills out the admission record based on medical professional knowledge. After being signed and confirmed by both the doctor and the patient, it becomes legally effective.

[0015] Currently, this method requires the expertise of doctors and a lot of time for analysis and collation. Doctors need to receive many patients every day and also conduct ward rounds, perform surgeries, etc. Therefore, the entry cycle of admission records is currently very long and delayed, resulting in low work efficiency for doctors, which has become a bottleneck in their daily work. Summary of the invention

[0016] The present invention aims at addressing the above-mentioned deficiencies of the prior art and provides a method for generating admission records based on a large model with strong practicality.

[0017] A further technical task of the present invention is to provide a device for generating admission records based on a large model that is reasonably designed, safe and applicable.

[0018] The technical solution adopted by the present invention to solve the technical problem is:

[0019] A method for generating admission records based on a large model comprises the following steps:

[0020] S1, doctor-patient dialogue, consultation generation;

[0021] S2, mark the dialogue and perform fine-tuning;

[0022] S3, execution verification, scoring and evaluation;

[0023] S4. Determine whether the evaluation results are satisfactory and decide whether to put the product into operation or continue to make fine adjustments.

[0024] Furthermore, in step S1, the doctor's consultation with the patient converts the spoken content of both parties into a text of the doctor-patient dialogue through a recording conversion device. The doctor should ask questions based on the patient's symptoms, using professional knowledge and work experience to guide the patient to obtain the original dialogue content that is written into the hospital admission record.

[0025] Furthermore, the original conversation content includes questions and answers related to the chief complaint, questions and answers related to the present medical history, questions and answers related to the past medical history, questions and answers related to the personal history, marital history, female menstrual and reproductive history, family history, physical examination and auxiliary examination.

[0026] Furthermore, in step S2, it further includes:

[0027] S2-1, the doctor writes the admission record based on the doctor-patient dialogue, and obtains an admission record and the corresponding dialogue;

[0028] S2-2. The big model engineer needs to collect a large number of the above data pairs and mark them as input data sets required for fine-tuning the big model;

[0029] S2-3. The large model engineer divides the labeled data set into training data set, validation data set and test data set according to 96:2:2;

[0030] S2-4. The big model engineer generates and constructs professional prompt words for hospitalized cases, and uses the training data set to perform fine-tuning on the big model to obtain the fine-tuned big model.

[0031] Furthermore, in step S3, it further includes:

[0032] (1) The big model engineer uses the previous prompt words to execute the validation data set on the fine-tuned big model and obtains the admission record generation results of the validation data set;

[0033] (2) Doctors used the admission record generation results of the validation data set to compare with the manually completed admission record, and scored each item of the generated admission record. The scoring dimensions included terminology usage and content conformity, with 50 points each.

[0034] Furthermore, the individual admission record score and individual scores of the admission record were calculated based on the following data;

[0035] Chief complaint score: 50 points for terminology application + 50 points for content compliance = 100 points;

[0036] Current medical history score: 50 points for terminology application + 30 points for content compliance = 80 points;

[0037] Past history score: 50 points for terminology usage + 20 points for content compliance = 70 points

[0038] Personal history: 40 points for terminology + 40 points for content compliance = 80 points

[0039] Marital history: 40 points for terminology usage + 50 points for content compliance = 90 points

[0040] Female menstrual and reproductive history: 40 points for terminology usage + 50 points for content compliance = 90 points

[0041] Family history: 50 points for terminology + 40 points for content compliance = 90 points

[0042] Physical examination: 40 points for terminology usage + 30 points for content compliance = 70 points

[0043] Auxiliary examination: 40 points for terminology application + 40 points for content compliance = 80 points;

[0044] The score of a single admission record = (chief complaint score + current illness history score + past history score + personal history score + marriage history score + female menstrual and reproductive history score + family history score + physical examination score + auxiliary examination score) / 9;

[0045] Average score of a single admission record = SUM (single admission record score) / number of verified admission records;

[0046] Average score of single admission record = SUM (single admission record score) / number of verified admission records

[0047] Finally, the average score of a single admission record and the average score of each item in the admission record were obtained.

[0048] Furthermore, in step S4, based on the set threshold, it is determined whether the result of this fine-tuning meets expectations. If a single score does not meet expectations, the training data set, validation data set and test data set are adjusted, and the next round of fine-tuning is re-entered until satisfactory results are achieved.

[0049] A device for generating admission records based on a large model, comprising: at least one memory and at least one processor;

[0050] The at least one memory is used to store a machine-readable program;

[0051] The at least one processor is used to call the machine-readable program to execute a method for generating admission records based on a large model.

[0052] Compared with the prior art, the method and device for generating admission records based on a large model of the present invention have the following outstanding beneficial effects:

[0053] The present invention combines the large model fine-tuning method with the item-by-item scoring and evaluation mechanism to continuously fine-tune and optimize the model's ability to generate admission records based on doctor-patient conversations, ultimately achieving accurate generation of professional admission records based on doctor-patient conversations, which are finally confirmed by doctors and patients to take effect, thereby saving time for doctors and patients and improving the work efficiency of the entire process. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0055] Attached Figure 1 It is a flowchart of a method for generating admission records based on a large model;

[0056] Attached Figure 2 It is a flowchart of the doctor-patient dialogue and consultation generation based on the large model admission record generation method;

[0057] Attached Figure 3 It is a flowchart of labeling dialogues and performing fine-tuning in a large model-based admission record generation method;

[0058] Attached Figure 4 It is a flowchart for performing verification and scoring evaluation in a large model-based admission record generation method;

[0059] Attached Figure 5 The present invention is a flowchart for evaluating whether the effect of the large-model-based admission record generation method is satisfactory. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention is further described in detail below in conjunction with specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] A best embodiment is given below:

[0062] like Figure 1 As shown, a method for generating admission records based on a large model in this embodiment has the following steps:

[0063] S1, doctor-patient dialogue, consultation generation;

[0064] When a doctor consults a patient, the recording conversion device will convert the spoken language of both parties into text of the doctor-patient conversation. The doctor should ask questions based on the patient's symptoms, using his professional knowledge and work experience, and guide the patient to obtain the original conversation content that is written into the hospital admission record.

[0065] like:

[0066] Questions and answers related to the chief complaint, such as the main symptoms and duration of the visit;

[0067] Questions and answers related to current medical history, such as: symptom causes, accompanying symptoms, medication information, patient status, etc.

[0068] Questions and answers related to past medical history, such as: past history of chronic diseases, history of infectious diseases, history of surgeries, history of blood transfusions, history of drug allergies, history of vaccinations, etc.

[0069] Questions and answers related to personal history, such as: whether born and raised in the place of origin, history of long-term residence in other places, history of contact with infected water, history of smoking and drinking, history of contact with industrial dust, poisons and radioactive substances, history of traveling, etc.

[0070] Marital history: when did you get married, how many sons and daughters you have, the health of your spouse and children. Female menstrual and reproductive history: how your periods appear, whether your periods were regular in the past, etc.

[0071] Family history: health status of parents and siblings, family history of genetic diseases.

[0072] Physical examination: Information about the physical examination performed on the patient by the doctor on the spot.

[0073] Auxiliary examination: The results of medical examinations done on the patient before admission to the hospital.

[0074] S2, mark the dialogue and perform fine-tuning;

[0075] Further including:

[0076] S2-1, the doctor writes the admission record based on the doctor-patient dialogue, and obtains an admission record and the corresponding dialogue;

[0077] S2-2. The big model engineer needs to collect a large number of the above data pairs and mark them as input data sets required for fine-tuning the big model;

[0078] S2-3. The large model engineer divides the labeled data set into training data set, validation data set and test data set according to 96:2:2;

[0079] S2-4. The big model engineer generates and constructs professional prompt words for hospitalized cases, and uses the training data set to perform fine-tuning on the big model to obtain the fine-tuned big model.

[0080] S3, execution verification, scoring and evaluation;

[0081] (1) The big model engineer uses the previous prompt words to execute the validation data set on the fine-tuned big model and obtains the admission record generation results of the validation data set;

[0082] (2) Doctors used the admission record generation results of the validation data set to compare with the manually completed admission record, and scored each item of the generated admission record. The scoring dimensions included terminology usage and content conformity, with 50 points each.

[0083] The individual admission record scores and individual scores for the admission record were calculated based on the following data;

[0084] Chief complaint score: 50 points for terminology application + 50 points for content compliance = 100 points;

[0085] Current medical history score: 50 points for terminology application + 30 points for content compliance = 80 points;

[0086] Past history score: 50 points for terminology usage + 20 points for content compliance = 70 points

[0087] Personal history: 40 points for terminology + 40 points for content compliance = 80 points

[0088] Marital history: 40 points for terminology usage + 50 points for content compliance = 90 points

[0089] Female menstrual and reproductive history: 40 points for terminology usage + 50 points for content compliance = 90 points

[0090] Family history: 50 points for terminology + 40 points for content compliance = 90 points

[0091] Physical examination: 40 points for terminology usage + 30 points for content compliance = 70 points

[0092] Auxiliary examination: 40 points for terminology application + 40 points for content compliance = 80 points;

[0093] The score of a single admission record = (chief complaint score + current illness history score + past history score + personal history score + marriage history score + female menstrual and reproductive history score + family history score + physical examination score + auxiliary examination score) / 9;

[0094] Average score of a single admission record = SUM (single admission record score) / number of verified admission records;

[0095] Average score of single admission record = SUM (single admission record score) / number of verified admission records

[0096] Finally, the average score of a single admission record and the average score of each item in the admission record (such as chief complaint score, current illness history score, personal history score, etc.) were obtained.

[0097] S4. Determine whether the evaluation results are satisfactory and decide whether to put it into operation or continue fine-tuning;

[0098] Based on the set threshold, determine whether the fine-tuning result meets expectations;

[0099] For example: The average score of a single admission record = 90, the threshold is 90 points, this item passed!

[0100] The average score of the chief complaint = 90, the threshold is 90 points, this item passed!

[0101] Average score of current medical history = 80, threshold 80 points, this item passed!

[0102] If the score of a single item does not meet expectations, we need to adjust the training data set, validation data set, and test data set, and re-enter the next round of fine-tuning until satisfactory results are achieved.

[0103] Based on the above method, a large model-based admission record generation device in this embodiment includes: at least one memory and at least one processor;

[0104] The at least one memory is used to store a machine-readable program;

[0105] The at least one processor is used to call the machine-readable program to execute a method for generating admission records based on a large model.

[0106] The above-mentioned specific implementations are only specific cases of the present invention. The patent protection scope of the present invention includes but is not limited to the above-mentioned specific implementations. Any technical solutions that comply with the claims of the present invention and any appropriate changes or substitutions made by ordinary technicians in the relevant technical field shall fall within the patent protection scope of the present invention.

[0107] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for generating admission records based on a large model, It is characterized in that The steps are as follows: S1, doctor-patient dialogue, consultation generation; S2, mark the dialogue and perform fine-tuning; S3, execution verification, scoring and evaluation; S4. Determine whether the evaluation results are satisfactory and decide whether to put the product into operation or continue to make fine adjustments.

2. A method for generating admission records based on a large model according to claim 1, It is characterized in that In step S1, the doctor interviews the patient and converts the spoken content of both parties into a text of the doctor-patient conversation through a recording conversion device. The doctor asks questions based on the patient's symptoms, using his professional knowledge and work experience to guide the patient to obtain the original conversation content that is written into the hospital admission record.

3. A method for generating admission records based on a large model according to claim 2, It is characterized in that The original conversation content includes questions and answers related to the chief complaint, questions and answers related to the present medical history, questions and answers related to the past medical history, questions and answers related to the personal history, marital history, female menstrual and reproductive history, family history, physical examination and auxiliary examination.

4. A method for generating admission records based on a large model according to claim 3, It is characterized in that In step S2, further comprising: S2-1, the doctor writes the admission record based on the doctor-patient dialogue, and obtains an admission record and the corresponding dialogue; S2-2. The big model engineer needs to collect a large number of the above data pairs and mark them as input data sets required for fine-tuning the big model; S2-3. The large model engineer divides the labeled data set into training data set, validation data set and test data set according to 96:2:2; S2-4. The big model engineer generates and constructs professional prompt words for hospitalized cases, and uses the training data set to perform fine-tuning on the big model to obtain the fine-tuned big model.

5. A method for generating admission records based on a large model according to claim 4, It is characterized in that In step S3, further comprising: (1) The big model engineer uses the previous prompt words to execute the validation data set on the fine-tuned big model and obtains the admission record generation results of the validation data set; (2) Doctors used the admission record generation results of the validation data set to compare with the manually completed admission record, and scored each item of the generated admission record. The scoring dimensions included terminology usage and content conformity, with 50 points each.

6. A method for generating admission records based on a large model according to claim 5, It is characterized in that The individual admission record scores and individual scores for the admission record were calculated based on the following data; Chief complaint score: 50 points for terminology application + 50 points for content compliance = 100 points; Current medical history score: 50 points for terminology application + 30 points for content compliance = 80 points; Past history score: 50 points for terminology usage + 20 points for content compliance = 70 points Personal history: 40 points for terminology + 40 points for content compliance = 80 points Marital history: 40 points for terminology usage + 50 points for content compliance = 90 points Female menstrual and reproductive history: 40 points for terminology usage + 50 points for content compliance = 90 points Family history: 50 points for terminology usage + 40 points for content compliance = 90 points Physical examination: 40 points for terminology usage + 30 points for content compliance = 70 points Auxiliary examination: 40 points for terminology application + 40 points for content compliance = 80 points; The score of a single admission record = (chief complaint score + current illness history score + past history score + personal history score + marriage history score + female menstrual and reproductive history score + family history score + physical examination score + auxiliary examination score) / 9; Average score of a single admission record = SUM (single admission record score) / number of verified admission records; Average score of single admission record = SUM (single admission record score) / number of verified admission records Finally, the average score of a single admission record and the average score of each item in the admission record were obtained.

7. A method for generating admission records based on a large model according to claim 6, It is characterized in that In step S4, based on the set threshold, it is determined whether the result of this fine-tuning meets expectations. If the score of a single item does not meet expectations, the training data set, validation data set and test data set are adjusted, and the next round of fine-tuning is re-entered until satisfactory results are achieved.

8. A device for generating admission records based on a large model, It is characterized in that include: at least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is used to call the machine-readable program to execute a method for generating an admission record.