Medical record information extraction system based on AI large model

Abstract

The invention belongs to the technical field of case information extraction, and particularly relates to a case information extraction system based on an AI large model, which comprises an input module, a processor, an AI screening model, a case database and a display, the processor is connected with the AI screening model, the processor converts input case information into an instruction recognized by the AI screening model, the AI screening model is connected with the case database, and the AI screening model carries out case screening in the case database based on the instruction. Cases of a patient are stored in a case database, the cases in the case database are browsed and screened based on an AI screening model, medical record texts are pre-trained based on a large-scale pre-training language model, then fine adjustment is carried out on a specific medical record data set, and by means of the powerful language understanding ability of the model, the medical record data set can be obtained. And the accuracy and recall rate of named entity recognition are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of case information extraction, and particularly to a medical record information extraction system based on an AI large model. Background Art

[0002] A medical record (case history) is a record of the medical activities of medical staff regarding the occurrence, development, and outcome of a patient's disease, including examinations, diagnoses, treatments, etc. It is also a medical health record of the patient written in a prescribed format and requirements after summarizing, organizing, and comprehensively analyzing the collected data. The medical record is both a summary of clinical practice work and a legal basis for exploring disease patterns and handling medical disputes, and is a valuable asset of the country. Medical records play an important role in medical treatment, prevention, teaching, scientific research, hospital management, etc.

[0003] In modern times, a medical record refers to the sum of information such as words, symbols, charts, images, and sections formed by medical staff during medical activities. It is mainly completed by clinical physicians and medical staff such as nurses and medical technicians. They complete the medical record based on the data obtained from medical activities such as medical history taking, physical examination, auxiliary examinations, diagnosis, treatment, and nursing, through induction, analysis, and organization. The medical record not only records the condition but also records the process of the physician's analysis, diagnosis, treatment, and nursing of the condition, the prognosis estimate, and the opinions of physicians at all levels during ward rounds and consultations. As the original record of the entire diagnosis and treatment process of the patient, the medical record records the true situation of the patient's admission to the hospital, including the description of the onset process by the patient or accompanying person, the diagnosis, treatment, and physical and chemical examinations of the patient by medical staff, until the patient is discharged or dies. Therefore, the medical record is both an actual record of the condition and an embodiment of medical and nursing quality and academic level.

[0004] Currently, for the examination of patient cases, it is usually necessary to manually move the patient cases over and flip through each case until the corresponding patient case is found, which is time-consuming and labor-intensive. Summary of the Invention

[0005] The purpose of the present invention is to provide a medical record information extraction system based on an AI large model to solve the problem that for the examination of patient cases, it is usually necessary to manually move the patient cases over and flip through each case until the corresponding patient case is found, which is time-consuming and labor-intensive as mentioned in the above background art.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A medical record information extraction system based on an AI large model, comprising:

[0007] An input module, a processor, an AI screening model, a case database, and a display;

[0008] Among them, the output end of the input module is connected to the processor. Based on the case information input to the processor by the input module, the processor establishes a connection with the AI screening model. The processor converts the input case information into instructions recognizable by the AI screening model. The AI screening model is connected to the case database, and the AI screening model performs case screening in the case database based on the instructions to obtain target cases;

[0009] The output end of the processor is connected to the input end of the display, and the display is used to display the obtained target cases.

[0010] Preferably, the input module uses an input keyboard, a mouse, or an input touch screen.

[0011] Preferably, the AI screening model needs to be trained. The specific training steps include: clarifying the screening target, data collection, data preprocessing, selecting an AI model, model training, model evaluation, and model optimization.

[0012] Preferably, the specific steps of training the AI screening model are as follows:

[0013] Clarify the screening target: Clearly define the purpose of the screening task;

[0014] Data collection: Collect data from multiple channels according to the screening target;

[0015] Data preprocessing: Clean the collected data to remove duplicate, incorrect, and missing value data;

[0016] Select an AI model: Select a model according to the screening task and data characteristics;

[0017] Model training: Divide the preprocessed data into a training set, a validation set, and a test set;

[0018] Model evaluation: Evaluate the trained model with the test set;

[0019] Model optimization: According to the evaluation results, if the model performance does not meet the standard, adjust the hyperparameters.

[0020] Preferably, when the input module is an input touch screen, the functions of the input touch screen are integrated on the display.

[0021] Preferably, the case database includes an identity information storage unit, a case storage unit, and a patient feature storage unit, and the identity information storage unit, the case storage unit, and the patient feature storage unit are interconnected.

[0022] Preferably, the identity information storage unit is used to store the user's name, ID number, medical record number, and home address.

[0023] Preferably, the case storage unit corresponds one-to-one with the identity information storage unit. The case storage unit is used to store the case information of the corresponding user, and record the patient and diagnosis and treatment information in chronological order.

[0024] Preferably, the patient feature storage unit is associated with the case storage unit, and stores the patient features shown by the patient based on the patient information recorded in the case storage unit.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] In this solution, the cases of patients are stored in the case database. Based on the AI screening model, the cases in the case database are browsed and screened. Based on a large-scale pre-trained language model, the medical record text is pre-trained, and then fine-tuned on a specific medical record data set. Utilizing the powerful language understanding ability of the model, the accuracy and recall rate of named entity recognition are improved.

[0027] Moreover, the AI model combines multiple tasks such as named entity recognition and relation extraction to construct a complete medical record information extraction system. An end-to-end deep learning architecture can be adopted to directly extract structured information from the original medical record text.

[0028] The large AI model provides a unified language understanding foundation for system integration. Through the method of multi-task learning, the model simultaneously learns multiple information extraction tasks, and utilizes the correlation between tasks to promote each other and improve the performance of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is the system logic block diagram of the present invention;

[0030] Figure 2 is the system logic block diagram of the case database of the present invention;

[0031] Figure 3 is the training flow chart of the AI screening model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0033] In the description of the present invention, it is necessary to understand that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0034] Embodiment 1:

[0035] See also Figures 1-3 The present invention provides a technical solution: a medical record information extraction system based on an AI big model, comprising:

[0036] Input module, processor, AI screening model, case database and display;

[0037] The output end of the input module is connected to the processor. Based on the case information that the input module requires the processor to input, the processor establishes a connection with the AI ​​screening model. The processor converts the input case information into instructions recognized by the AI ​​screening model. The AI ​​screening model establishes a connection with the case database. The AI ​​screening model performs case screening in the case database based on the instructions to obtain the target case.

[0038] The output end of the processor is connected to the input end of the display, and the display is used to display the acquired target case.

[0039] Analysis of the above content: This solution has many application scenarios, as shown below:

[0040] For example, it is used at the recruitment site. Since the work content has physical requirements for the applicant, a cooperative relationship is established between the recruiting company and the hospital. The recruiting company inputs the applicant's identity information based on the input module, and the AI ​​screening model searches the hospital's case database for the case of the user corresponding to the identity information based on the identity information. Based on the case, it determines whether the applicant's physical condition does not meet the job requirements.

[0041] For example, in hospitals, when retrieving user medical records, there are currently both electronic medical records and paper medical records, and the efficiency is low during retrieval. Even if electronic medical records are used, duplicate names and case information acquisition problems lead to low efficiency. This solution is based on AI screening models to screen cases, directly obtain case content, obtain patient conditions, and provide accurate answers.

[0042] Embodiment 2:

[0043] See also Figures 1-3 The present invention provides a technical solution based on the first embodiment: the input module adopts an input keyboard, a mouse or an input touch screen.

[0044] Analysis of the above content: Input keyboards, mice, or input touchscreens are common input devices, and here they are selected according to the convenience of input information and application habits.

[0045] Example 3:

[0046] Please refer to Figures 1-3 , based on Example 1, the present invention provides a technical solution: The AI screening model needs to be trained, and the specific training steps include: clarifying the screening target, data collection, data preprocessing, selecting an AI model, model training, model evaluation, and model optimization. The specific training steps of the AI screening model are as follows:

[0047] Clarify the screening target: Clearly define the purpose of the screening task. For example, in the recruitment scenario, screen for suitable candidates, in the medical field, screen for potential disease patients, and in the e-commerce field, screen for high-value customers, etc. Only by clarifying the target can the subsequent required data and evaluation indicators be determined.

[0048] Data collection: According to the screening target, collect data from multiple channels. For example, in medical screening, collect patient medical records, examination reports, etc.; in recruitment screening, collect candidate resumes, interview records, etc. Data sources include databases, file systems, web crawlers, etc.

[0049] Data preprocessing: Clean the collected data, remove duplicate, incorrect, and missing value data. For example, delete duplicate candidate resumes. Standardize the data, such as unifying the different formats of educational background information. Feature engineering is also required to extract key features, such as extracting symptoms, indicators, etc. from medical records and converting them into a numerical form that the model can process.

[0050] Select an AI model: Select a model according to the screening task and data characteristics. For simple binary classification screening, logistic regression can be used; for complex pattern recognition such as image screening for cancer cells, use the deep learning model convolutional neural network; for processing sequential data such as text screening, use recurrent neural networks or Transformer architectures.

[0051] Model training: Divide the preprocessed data into a training set, a validation set, and a test set. Use the training set to train the model and adjust the parameters to make the model perform best on the validation set. During the training process, use optimization algorithms such as stochastic gradient descent to update the model parameters and monitor the loss function and evaluation indicators.

[0052] Model evaluation: Use the test set to evaluate the trained model. Common indicators include accuracy, recall, F1 value, precision, etc. For example, in medical disease screening, focus on recall to avoid missing patients; in recruitment screening, comprehensively consider accuracy and recall.

[0053] Model Optimization: Based on the evaluation results, if the model performance does not meet the standards, adjust the hyperparameters, such as changing the number of neural network layers and the learning rate. You can also reprocess the data, increase the data volume or improve the feature engineering. You can also try to integrate multiple models, such as combining the results of multiple classifiers to improve performance.

[0054] Furthermore, the following steps can also be carried out:

[0055] Deployment and Application: Deploy the optimized model to the actual environment, such as an online recruitment system or a medical diagnosis platform. Process new data in real time and screen the results according to the model output, such as automatically screening resumes or assisting doctors in diagnosis.

[0056] Continuous Monitoring and Updating: After the model is deployed, continuously monitor its performance and regularly evaluate it with new data, such as evaluating the accuracy of the medical screening model monthly. According to the new data and business changes, update the model in a timely manner and retrain it to adapt to the new situation, such as updating the medical screening model as the disease characteristics change.

[0057] Example 4:

[0058] Please refer to Figures 1-3 , based on Example 1, the present invention provides a technical solution: when the input module is an input touch screen, the functions of the input touch screen are integrated on the display.

[0059] Analysis of the above content: After the functions of the input touch screen are integrated on the display, a touch display screen is obtained. The following two types of touch display screens are used here:

[0060] Capacitive Touch Screen: It works using the capacitance characteristics of the human body. Its structure is usually to coat a layer of transparent ITO conductive material on the glass surface to form a capacitance array. When a finger touches the screen, due to the action of the human body electric field, the capacitance value at the touch point will change, and the touch position is determined by detecting the change in capacitance. Capacitive touch screens have a fast response speed and are touch-sensitive, and dominate in consumer electronic devices such as smartphones and tablets.

[0061] Surface Acoustic Wave Touch Screen: Ultrasonic transmitters and ultrasonic receivers are respectively installed at the four corners of the glass screen. When there is no touch, the ultrasonic waves emitted by the transmitter propagate on the screen surface and are received by the receiver. When an object touches the screen, it will absorb part of the ultrasonic energy, causing a change in the signal received by the receiver, and the touch point position is determined by detecting this change. Surface acoustic wave touch screens are not affected by environmental factors such as temperature and humidity, have good stability, and are commonly used in public information query devices, bank self-service terminals, etc.

[0062] Example 5:

[0063] Please refer to Figures 1-3, the present invention provides a technical solution based on Embodiment 1: The case database includes an identity information storage unit, a case storage unit, and a patient feature storage unit, and the identity information storage unit, the case storage unit, and the patient feature storage unit are interconnected with each other.

[0064] The identity information storage unit is used to store the user's name, ID number, medical record number, and home address. The case storage unit corresponds one-to-one with the identity information storage unit, and the case storage unit is used to store the case information of the corresponding user, and record the patient's illness and diagnosis and treatment information in chronological order. The patient feature storage unit is associated with the case storage unit and stores the patient features exhibited by the patient based on the patient information recorded in the case storage unit.

[0065] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic features of the present invention, the present invention can be implemented in other specific forms; therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced by the present invention, and any reference signs in the claims should not be regarded as limiting the claims involved.

[0066] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A medical record information extraction system based on a large AI model, characterized in that, Including: An input module, a processor, an AI screening model, a case database, and a display; Among them, the output end of the input module is connected to the processor. Based on the case information required by the input module for the processor, the processor establishes a connection with the AI screening model. The processor converts the input case information into instructions recognizable by the AI screening model. The AI screening model is connected to the case database. The AI screening model performs case screening in the case database based on the instructions to obtain target cases; The output end of the processor is connected to the input end of the display. The display is used to display the obtained target cases.

2. The medical record information extraction system based on the AI large model according to claim 1, characterized in that: The input module uses an input keyboard, a mouse, or an input touch screen.

3. The medical record information extraction system based on the AI large model according to claim 1, characterized in that: The AI screening model needs to be trained. The specific training steps include: clarifying the screening target, data collection, data preprocessing, selecting an AI model, model training, model evaluation, and model optimization.

4. The medical record information extraction system based on the AI large model according to claim 3, characterized in that: The specific training steps of the AI screening model are as follows: Clarifying the screening target: Clearly defining the purpose of the screening task; Data collection: Collecting data from multiple channels according to the screening target; Data preprocessing: Cleaning the collected data to remove duplicate, incorrect, and missing value data; Selecting an AI model: Selecting a model according to the screening task and data characteristics; Model training: Dividing the preprocessed data into a training set, a validation set, and a test set; Model evaluation: Evaluating the trained model with the test set; Model optimization: According to the evaluation results, if the model performance does not meet the standard, adjusting the hyperparameters.

5. The medical record information extraction system based on the AI large model according to claim 2, characterized in that: When the input module is an input touch screen, the functions of the input touch screen are integrated on the display.

6. The medical record information extraction system based on the AI large model according to claim 1, characterized in that: The case database includes an identity information storage unit, a case storage unit, and a patient feature storage unit. The identity information storage unit, the case storage unit, and the patient feature storage unit are connected to each other.

7. An electronic medical record information extraction system based on an AI large model according to claim 6, characterized in that: The identity information storage unit is used to store the user's name, ID number, medical record number, and home address.

8. The medical record information extraction system based on the AI large model according to claim 6, characterized in that: The case storage unit corresponds one-to-one with the identity information storage unit. The case storage unit is used to store the case information of the corresponding user, recording the patient and diagnosis and treatment information in chronological order.

9. The medical record information extraction system based on the AI large model according to claim 6, wherein: The patient feature storage unit is associated with the case storage unit and stores the patient features shown by the patient based on the patient information recorded in the case storage unit.

Images