Community medical terminal triage decision intelligent generation system based on deep learning

Abstract

The invention relates to the technical field of data analysis, and discloses a community medical terminal triage decision intelligent generation system based on deep learning, which comprises a training module for performing model training by using training data of different triage departments to obtain a triage model; the optimization module is used for screening the questionnaire data of the target object by utilizing preset training data, judging the triage model by utilizing the conformity degree and the similarity degree of a preset triage department and the questionnaire data, obtaining the triage capability, and determining the triage defect degree of the to-be-supplemented disease in combination with the non-triage times and the triage capability; on the basis of the similarity between the questionnaire data of the to-be-supplemented disease and the questionnaire data of different triage departments, the feature supplement necessity degree of the to-be-supplemented disease is determined in combination with the triage defect degree; supplementing training data of a preset training model based on the feature supplement necessity degree; the triage accuracy can be effectively improved, and the workload of medical care is effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of data analysis, specifically to a community medical terminal triage decision intelligent generation system based on deep learning. Background Technology

[0002] Current medical triage systems primarily rely on human experience or rule-based automated systems for initial triage. Traditional methods use prominent disease features as training data to acquire triage models, but this approach has significant drawbacks: First, it fails to adequately consider the similarity between prominent features of similar diseases, leading to triage confusion when faced with certain questionnaire answers; second, existing systems lack mechanisms to identify potential misselected options in patient questionnaire data; third, traditional models cannot dynamically assess their triage capability for specific questionnaire data; and finally, the system lacks mechanisms for automatically identifying and supplementing defects in training data.

[0003] These problems are particularly prominent in community healthcare settings. Patient-completed questionnaires may contain misunderstandings or incorrect selections, and traditional systems cannot effectively identify and correct these errors. Furthermore, due to the complexity and overlap of disease symptoms, models often face triage difficulties, but current technologies cannot accurately assess these situations and take appropriate measures. In addition, with the emergence of new disease types and changes in disease characteristics, the training data for traditional models often cannot be updated in a timely manner, leading to a decline in triage accuracy.

[0004] While deep learning technology can automatically learn the complex nonlinear relationships between massive combinations of options and diseases within a department, it still faces many challenges in practical applications. For example, the model has limited ability to handle atypical or overlapping symptoms, insufficient judgment on the reliability of patient questionnaire data, and a lack of dynamic evaluation mechanisms for its own triage capabilities. These problems seriously affect the accuracy and reliability of the triage system, increase the workload of medical staff, and may also lead to delays in patient treatment.

[0005] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0006] To address the existing technical problems, the present invention aims to provide a community medical terminal triage decision intelligent generation system based on deep learning, which has the advantages of dynamically identifying incorrect questionnaire options, real-time evaluation of model triage capabilities, and automatic supplementation of training data defects, thereby improving triage accuracy and effectively reducing the workload of medical staff.

[0007] To address the aforementioned technical problems, this application provides the following technical solution: A deep learning-based intelligent triage decision generation system for community healthcare terminals, comprising: a training module for training a model using training data from different triage departments to obtain a triage model; an optimization module for filtering questionnaire data of target subjects using preset training data; judging the triage model based on the conformity and similarity of preset triage departments and the questionnaire data to obtain triage capability; determining the degree of triage deficiency for diseases to be supplemented by combining the number of times they cannot be triaged with the triage capability; determining the necessity of supplementing features for diseases to be supplemented based on the similarity between the questionnaire data of the diseases to be supplemented and the questionnaire data of different triage departments, combined with the degree of triage deficiency; and supplementing the training data of the preset training model based on the necessity of supplementing features.

[0008] In one embodiment of the present invention, the training process of the training module is as follows: designing a questionnaire, the questionnaire including at least: symptom characteristics, medical history information and basic information; combining questionnaire options for all symptoms, and assigning them to the corresponding departments according to the combination of options, and obtaining labeled data of suspected symptom types; using the labeled data as the training data to train the model to obtain the triage model.

[0009] In one embodiment of the present invention, the optimization module includes: a screening module, which uses preset training data to screen questionnaire data of target objects; a triage capability judgment module, which uses the conformity and similarity of preset triage departments and the questionnaire data to judge the triage model and obtain triage capability; a triage defect judgment module, which combines the number of times the disease cannot be triaged and the triage capability to determine the degree of triage defect of the disease to be supplemented; a data supplementation judgment module, which determines the degree of feature supplementation necessity of the disease to be supplemented based on the similarity between the questionnaire data of the disease to be supplemented and the questionnaire data of different triage departments, combined with the degree of triage defect; and a determination module, which supplements the training data of the preset training model based on the degree of feature supplementation necessity.

[0010] In one embodiment of the present invention, the screening module includes: obtaining the number of common symptom options between a first questionnaire option combination filled in by the target object and a second questionnaire option combination of each training data; obtaining the severity difference of each option between the first questionnaire option combination and the second questionnaire option combination; determining a first selection similarity between the first questionnaire option combination and the second questionnaire option combination using the number of common symptom options and the severity difference; determining the necessity of removing the first questionnaire option combination using the severity difference and the first selection similarity; and removing the target option from the first questionnaire option combination in response to the necessity of removal being greater than a second preset threshold.

[0011] In one embodiment of the present invention, determining the necessity of eliminating the first questionnaire option combination by using the severity difference magnitude and the first selection similarity further includes: obtaining the maximum value of the severity difference magnitude among all options; and determining the necessity of eliminating the target option in the first questionnaire option combination by using the maximum value of the severity difference magnitude and the first selection similarity.

[0012] In one embodiment of the present invention, the triage capability judgment module makes the following judgments: obtaining the degree of conformity of the first questionnaire option combination filled in by the target object to different triage categories, and obtaining the minimum value of all the conformity degrees; obtaining the average value of the conformity degrees of all triage categories; and using the conformity degrees, the minimum value of the conformity degrees, and the average value of the conformity degrees, determining the triage capability of the triage model.

[0013] In one embodiment of the present invention, determining the triage capability of the triage model using the degree of conformity, the minimum value of the degree of conformity, and the mean value of the degree of conformity further includes: obtaining the total number of triage categories; using the total number of triage categories, the degree of conformity, and the minimum value of the degree of conformity, determining the overall degree of underestimation of the degree of conformity of the first questionnaire option combination for all triage categories; using the mean value of the degree of conformity, the total number of triage categories, and the degree of conformity, determining the similarity of the degree of conformity of the first questionnaire option combination for all triage categories; using the similarity and the overall degree of underestimation, determining the low triage capability performance of the first questionnaire option combination; and, in response to the low triage capability performance being greater than a third preset threshold, determining that the triage model has low triage capability and performing preset triage on the first questionnaire option combination.

[0014] In one embodiment of the present invention, the triage defect judgment module includes: obtaining the proportion of the number of times the triage model cannot triage the target symptom in the total number of times the target symptom is identified; in response to the target symptom being untriageable, obtaining the number of different times between the triage department corresponding to the maximum degree of conformity and the preset triage department, and obtaining the magnitude of low triage ability performance corresponding to the target object questionnaire option combination for each target symptom; and using the magnitude of low triage ability performance, the number of different times, and the proportion, determining the degree of triage defect of the triage model for the target symptom.

[0015] In one embodiment of the present invention, the data supplementation judgment module makes the following judgments: obtaining the number of first disease types and the number of triage departments for each triage department, and obtaining the number of second disease types corresponding to all triage departments; obtaining the difference between the degree of triage defect of the target symptom and the maximum value of the degree of triage defect; and using the difference, the number of second disease types, the number of triage departments, and the number of first disease types, determining the degree of feature supplementation necessity of the triage model.

[0016] In one embodiment of the present invention, supplementing the training data of the preset training model based on the necessity of feature supplementation includes: obtaining a second selection similarity between the third questionnaire option combination corresponding to the target symptom and the fourth questionnaire option combination of other triage departments; in response to the second selection similarity being greater than a fourth preset threshold, obtaining the number of the first disease types and the corresponding number of triage departments corresponding to the fourth questionnaire option combinations of other triage departments; supplementing the target symptom with corresponding feature training data using the number of the first disease types and the second feature supplementation necessity being greater than a fifth preset threshold, and optimizing the triage model with the disease to be supplemented.

[0017] The beneficial effects of this invention are as follows: It provides a community medical terminal triage decision intelligent generation system based on deep learning, which builds a triage model through a training module, realizes dynamic optimization of data screening, capability assessment and feature supplementation through an optimization module, and provides visualized results through a display module. It has the advantages of dynamically identifying incorrect questionnaire options, evaluating the model's triage capability in real time, and automatically supplementing training data defects, which can effectively improve the triage accuracy and effectively reduce the workload of medical staff. Attached Figure Description

[0018] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the structure of the first embodiment of the community medical terminal triage decision intelligent generation system based on deep learning provided by the present invention; Figure 2 This is a schematic diagram of the second embodiment of the community medical terminal triage decision intelligent generation system based on deep learning provided by the present invention. Detailed Implementation

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a deep learning-based intelligent generation system for triage decisions in community healthcare terminals proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0022] In existing technologies, community healthcare triage systems typically rely on human experience or fixed rules for initial diagnosis. Traditional methods construct triage models based on explicit symptom features, but fail to adequately consider the overlap in symptom presentation between similar diseases. When patients submit incorrectly selected options or their symptom descriptions are atypical, the system is prone to misclassifying cases with overlapping symptoms into a single department, leading to a decrease in triage accuracy. Especially when dealing with complex cases involving multiple departmental affiliations, existing models struggle to dynamically identify data biases and effectively correct for deficiencies in the training data.

[0023] To address these issues, the research team discovered a correlation between the degree to which incorrectly selected options interfere with triage results and the differences in features among similar training data. By analyzing the differences in severity between questionnaire options and historical cases, abnormal options can be identified and corrected. Simultaneously, the model's triage capability can be quantitatively assessed through the distribution differences in the degree of conformity across departments; when the degree of conformity becomes converging, it indicates the existence of triage blind spots. Time-based triage defect statistics provide a dynamic basis for data supplementation and optimization, forming a closed-loop improvement mechanism.

[0024] This application proposes a deep learning-based intelligent triage decision generation system for community medical terminals. The system constructs a triage model through a training module, dynamically optimizes the identification of incorrect options, assesses capabilities, and supplements features through an optimization module, and provides visualized results through a display module. It has the advantages of dynamically identifying incorrect questionnaire options, evaluating the model's triage capabilities in real time, and automatically supplementing training data defects. It can effectively improve triage accuracy and reduce the workload of medical staff.

[0025] The main objective of this invention is to: eliminate incorrectly selected options based on the differences between the patient's questionnaire choices and the options in the most similar training data; assess the current model's triage capability based on the differences in the degree of agreement between the questionnaire choices and different triage options, as well as the similarity between the choices and the triage data; and filter the training data to identify areas for expansion based on the similarity between the patient questionnaires corresponding to each triage situation and the training data, as well as the frequency of inappropriate triage within a short period of time.

[0026] This invention addresses the scenario where a triage model acquired through deep learning can automatically learn the complex nonlinear relationships between massive combinations of options and departmental diseases, overcoming the rigidity of traditional rule-based systems. It efficiently handles atypical or overlapping symptoms, achieving rapid and accurate initial screening and triage, significantly improving efficiency and consistency. Furthermore, it continuously evolves with data accumulation, optimizing the allocation of medical resources. This invention eliminates incorrectly selected options based on the differences between patient questionnaire choices and the options in the most similar training data. It assesses the current model's triage capability based on the differences in the magnitude of the correlation between questionnaire choices and different triage options, as well as the similarity between them. Finally, it filters data in the training data that should be expanded based on the similarity between patient questionnaires corresponding to different triage situations and the training data within a short period, as well as the frequency of inappropriate triage.

[0027] The following description, in conjunction with the accompanying drawings, details a specific solution for a deep learning-based intelligent generation system for triage decisions in community healthcare terminals provided by this invention.

[0028] Please see Figure 1 The diagram shows a structural schematic of the first embodiment of the community medical terminal triage decision intelligent generation system based on deep learning provided by the present invention.

[0029] like Figure 1 As shown, the community medical terminal triage decision intelligent generation system 10 based on deep learning includes: a training module 100, which trains a model using training data from different triage departments to obtain a triage model; an optimization module 200, which filters questionnaire data of target objects using preset training data; judges the triage model using the degree of conformity and similarity of preset triage departments and the questionnaire data to obtain triage capability; determines the degree of triage defect of diseases to be supplemented by combining the number of times they cannot be triaged and the triage capability; determines the degree of necessity for feature supplementation of diseases to be supplemented by combining the similarity between the questionnaire data of diseases to be supplemented and the questionnaire data of different triage departments; and supplements the training data of the preset training model based on the degree of necessity for feature supplementation.

[0030] The training module involves collecting data such as symptom characteristics and medical history through structured questionnaires. Different combinations of options and their corresponding departments are used as training samples, and a deep neural network is employed for model training. This module establishes a mapping relationship between symptom combinations and departmental classifications, providing a benchmark model for subsequent optimization. The optimization module's screening module identifies potential selection biases by comparing the options in the current questionnaire with historical data. The triage capability assessment determines the reliability of the model's decisions by calculating the dispersion of departmental compliance. The feature supplementation necessity analysis identifies training data categories requiring enhancement based on triage defect records within a time window. The presentation module transforms the multidimensional assessment results into visual charts to assist medical staff in understanding the system's decision-making basis.

[0031] Specifically, the system first processes patient questionnaires using a deep learning model to generate preliminary triage suggestions. The optimization module simultaneously initiates three analyses: comparing the symptom differences between the current options and historical data, eliminating abnormal options that significantly deviate from typical values; calculating the standard deviation of the departmental compliance, triggering manual review when the distribution is too uniform; and statistically analyzing recent triage failure cases to identify frequently occurring deficient symptoms. The triage results, after multi-dimensional optimization, along with departmental matching curves, abnormal option markers, and other information, are simultaneously displayed to medical staff through a visual interface.

[0032] Compared to existing technologies, traditional triage systems only perform one-way questionnaire-department matching, lacking a mechanism to verify the reliability of input data. This solution innovatively introduces a three-level optimization mechanism: eliminating option bias during data processing, evaluating classification reliability during model decision-making, and dynamically supplementing defective data during system maintenance. This closed-loop optimization system effectively solves traditional pain points such as misjudgment of cross-symptoms, interference from abnormal options, and delayed data updates.

[0033] Through the above technical solutions, this application effectively reduces the interference of incorrectly selected options on triage results and improves the ability to handle atypical cases. The system can automatically identify low-reliability triage cases requiring manual intervention and continuously optimize the composition of training data. Visual displays help medical staff quickly understand the multi-departmental correlation characteristics of complex cases, shortening triage decision-making time.

[0034] In some embodiments, the training process of the training module is as follows: Design a questionnaire that includes at least the following: symptom characteristics, medical history information, and basic information.

[0035] All symptoms were given questionnaire options, and the options were assigned to the corresponding departments. The labeled data of suspected symptom types were also obtained.

[0036] Labeled data is used as training data to train the model in order to obtain a triage model.

[0037] The questionnaire design involves constructing a multi-dimensional set of questions that includes symptom characteristics, medical history information, and basic information. This can be achieved by medical experts designing structured questionnaires based on clinical experience, ensuring that the questions cover common symptoms and disease-related factors. Option combination refers to arranging and combining different symptom options to form multiple possible questionnaire answers. This can be done by algorithms generating all possible symptom combinations and mapping them to corresponding departments to simulate real patient choices. Labeled data refers to associating each option combination with its corresponding department allocation and suspected symptom type. This can be achieved through manual review or automated rule matching, providing supervisory signals for model training.

[0038] Specifically, during the questionnaire design phase, medical experts design structured questions based on disease characteristics, covering symptom presentation, past medical history, and basic personal information. Each question has multiple options, such as symptom severity grading or medical history type selection. An algorithm then generates all possible combinations of symptom options, and each combination is assigned to a corresponding department and labeled with a suspected disease type based on medical knowledge. For example, a combination of headache and blurred vision might be assigned to neurology and labeled as migraine or suspected glaucoma. The labeled data is then fed into a deep learning model for training. The model learns the complex relationship between option combinations and department assignments to establish triage decision rules. Cross-validation is used during training to evaluate model performance, ensuring it can accurately identify the mapping relationship between symptom combinations and departments.

[0039] For example, questionnaire design requires setting a clear range of options for each question, which can be set according to the actual situation, including dimensions such as symptom characteristics, medical history information, and basic information. Medical experts combine questionnaire options based on existing diseases and assign different questionnaire option combinations to different departments, marking suspected disease types. The labeled data is used as training data to train the model using the TensorFlow deep learning framework, resulting in the corresponding questionnaire triage model.

[0040] Through the above technical solutions, this application achieves systematic modeling of the relationship between symptom combinations and departments, improving the model's ability to handle atypical symptom combinations. The structured questionnaire design ensures that training data covers more potential symptom interaction scenarios, and the option combination generation mechanism enhances the model's robustness in identifying complex cases. Departmental assignment labeling based on medical knowledge effectively reduces triage confusion for diseases with similar characteristics, enabling the trained model to more accurately distinguish the target department corresponding to symptom combinations.

[0041] In some embodiments, the optimization module may include the following: The filtering module uses preset training data to filter the questionnaire data of the target objects.

[0042] The triage capability assessment module uses the degree of conformity and similarity of preset triage departments and questionnaire data to assess the triage model and obtain its triage capability.

[0043] The triage defect judgment module combines the number of times a patient cannot be triaged with the ability to triage to determine the degree of triage defect for the disease to be supplemented.

[0044] The data supplementation judgment module determines the necessity of supplementing the features of the disease to be supplemented based on the similarity between the questionnaire data of the disease to be supplemented and the questionnaire data of different triage departments, combined with the degree of triage defects.

[0045] The determination module supplements the training data of the preset training model based on the degree of necessity of the feature supplementation.

[0046] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of the second embodiment of the community medical terminal triage decision intelligent generation system based on deep learning in this application.

[0047] like Figure 2 As shown, the deep learning-based intelligent triage decision generation system 10 for community healthcare terminals includes: a training module 100 and an optimization module 200. The optimization module 200 includes: a screening module 210, which uses preset training data to screen questionnaire data of target subjects; a triage capability judgment module 220, which uses the matching and similarity of preset triage departments and questionnaire data to judge the triage model and obtain triage capability; a triage defect judgment module 230, which combines the number of times a disease cannot be triaged with the triage capability to determine the degree of triage defect of the disease to be supplemented; a data supplementation judgment module 240, which determines the necessity of supplementing features of the disease to be supplemented based on the similarity between the questionnaire data of the disease to be supplemented and the questionnaire data of different triage departments, combined with the degree of triage defect; and a determination module 250, which supplements the training data of the preset training model based on the degree of feature supplementation necessity.

[0048] The triage module comprises several modules: The screening module compares the number and severity of shared symptom options in the target questionnaire with similar combinations in the training data to identify potentially misselected options. This can be achieved by calculating the ratio of the maximum difference in severity among shared symptom options to the selection similarity, thus eliminating interference from abnormal data. The triage capability assessment module analyzes the distribution of conformity between questionnaire option combinations and different triage departments to evaluate the model's triage capability for the current questionnaire. This can be achieved by calculating the minimum, mean, and similarity values ​​of conformity, thus identifying the model's limitations in triaging ambiguous symptoms. The triage defect assessment module analyzes the number of times a disease to be supplemented cannot be triaged and its triage capability to determine the degree of triage defect for that disease. The data supplementation assessment module determines the priority of supplementing training data by statistically analyzing the correlation between the degree of triage defect and disease types within a preset time period. This can be achieved by calculating the ratio of the degree of triage defect to the number of disease types and triage departments, thus dynamically optimizing the coverage of the model's training data. The determination module refers to supplementing the training data of the preset training model based on the necessary degree of feature supplementation.

[0049] Specifically, the screening module first obtains the number of common symptom options and the severity differences between the target questionnaire option combinations and each training data combination. It then calculates the similarity to identify and remove options with a high probability of misselection. The triage capability assessment module calculates the overall degree of bias and similarity index based on the triage conformity distribution output by the model, determining whether the model's triage capability for the current questionnaire meets the threshold. The data supplementation assessment module quantifies the necessity of feature supplementation based on the correlation between the degree of triage defects and disease types and triage departments, screening for disease types that require priority supplementation of training data. The determination module supplements the training data of the preset training model based on the aforementioned feature supplementation necessity to construct the final training model. This allows the input of patient questionnaire data into the final training model to generate the final conformity score, providing data support for visualization.

[0050] Through the above technical solutions, this application can accurately identify and eliminate misselected data during the questionnaire filling process, evaluate the model's triage capability boundary for complex symptoms in real time, and supplement missing training data in a targeted manner, thereby improving the triage model's ability to handle cross-symptoms and atypical symptoms, reducing the workload of manual review, and ensuring the accuracy and reliability of triage decisions.

[0051] In some embodiments, the filtering module may include the following.

[0052] Obtain the number of common symptom options between the first questionnaire option combination filled in by the target subjects and the second questionnaire option combination of each training data.

[0053] Obtain the magnitude of the severity difference for each option between the first questionnaire option combination and the second questionnaire option combination.

[0054] The similarity of first choice between the first questionnaire option combination and the second questionnaire option combination is determined by using the number of shared symptom options and the magnitude of the difference in severity.

[0055] The necessity of eliminating the first questionnaire option combination is determined by using the severity difference and the similarity of the first choice.

[0056] In response to the necessity of elimination exceeding a second preset threshold, the target option is removed from the first questionnaire option combination.

[0057] The number of shared symptom options refers to the number of identical symptom options in the patient questionnaire and the training data questionnaire. This can be achieved using text matching algorithms or hash table comparison, and is used to measure the degree of overlap in symptom descriptions. The severity difference refers to the difference in severity scores for the same symptom option between the patient questionnaire and the training data. This can be calculated using numerical subtraction; for example, if a patient selects a symptom score of 5 while the corresponding score in the training data is 7, the difference is 2 points, used to quantify the degree of deviation in option selection. First-choice similarity is a similarity index calculated by combining the number of shared symptoms and the severity difference. This can be achieved using weighted summation or multiplication, for example, the ratio of the number of shared symptom options to the mean of the severity differences of all shared symptoms, used to assess the overall matching degree between the patient questionnaire and the training data.

[0058] Specifically, the screening module calculates the similarity between patient questionnaire options and training data combinations by comparing the number of shared symptoms and the severity scores of each shared symptom. For example, if a patient selects a headache symptom score of 6, while the corresponding symptom score in the training data is 8, the difference is 2. A higher similarity is indicated if there are many shared symptoms and generally small differences. The similarity score is used to determine if there are incorrectly selected options. For instance, if the similarity exceeds a preset threshold, options in the patient questionnaire are considered to deviate significantly from the training data and need to be removed to improve triage accuracy.

[0059] Through the above technical solution, this application can accurately identify possible incorrectly selected options in patient questionnaires and improve the reliability of input data for the triage model by automatically removing abnormal data. For example, when a patient mistakenly selects 8 points for the degree of lower back pain when it should actually be 3 points, the system can automatically correct the abnormal value by comparing the score distribution of similar cases in the training data, avoiding triage errors caused by a single option error, thereby reducing the workload of manual review by medical staff.

[0060] In some embodiments, determining the necessity of eliminating the first questionnaire option combination by utilizing the severity difference magnitude and the first selection similarity may further include the following operations.

[0061] Get the maximum value of the severity difference among all options.

[0062] By using the maximum value of the severity difference and the similarity of the first choice, the necessity of removing the target option from the first questionnaire option combination is determined.

[0063] If the necessity for elimination exceeds the second preset threshold, the target option is removed from the first questionnaire option combination.

[0064] The severity difference refers to the numerical difference in symptom severity between the patient's questionnaire options and the corresponding options in the training data. This can be achieved by quantifying the severity of symptom options using a preset scoring standard and calculating the difference. For example, eye pain can be scored from 1 to 10; a smaller difference indicates a closer similarity in severity. The first preset threshold is a critical value used to filter symptom options with small severity differences. This can be determined using empirical values ​​or statistical analysis methods, such as setting it to 1 to filter options with insignificant differences. The necessity of removal is an indicator measuring the degree to which the target option interferes with the triage results. This can be calculated using the ratio of the maximum severity difference to the selection similarity. For example, normalization can be used to map the results to the 0-1 range to determine whether the removal criteria are met. The second preset threshold is the critical value that triggers the removal operation. This can be dynamically adjusted based on the impact of incorrectly selected options on triage accuracy in historical data, such as setting it to 0.8. When the necessity of removal exceeds this value, the removal operation is performed.

[0065] Specifically, after a patient completes the questionnaire, the system first identifies the common symptoms shared by their chosen combinations and similar combinations in the training data, and quantifies the severity differences between each option. For symptom options with differences less than a first preset threshold, the system counts their number and records the maximum difference value among all options. Subsequently, the system calculates the necessity of removal by combining selection similarity. For example, it uses a formula to correlate the number of common symptoms, the maximum difference value, and the similarity, and after normalization, determines whether it exceeds a second preset threshold. If it exceeds the threshold, the system automatically removes the target option and corrects the questionnaire data. For example, it adjusts the patient's "severe headache" to "moderate headache" corresponding to a similar combination in the training data, thereby avoiding interference from abnormal options on the triage model.

[0066] For example, the necessity of removing an option can be determined based on the differences between the questionnaire option combination and the individual options in the closest training data. Then, the first questionnaire option combination j completed by the patient (target subject) is compared with each option i in the second questionnaire option combination m of each training data set. The number of shared symptom options in the first questionnaire option combination j and the second questionnaire option combination m is obtained. .

[0067] The severity of specific symptoms under different options was numerically assigned (e.g., for the symptom option of eye swelling and pain, mild eye swelling and pain was scored 1-3 points, severe eye swelling and pain was scored 4-6 points, and severe eye swelling and pain was scored 7-10 points, etc.). The difference in numerical values ​​was used to assess the difference in severity between option combination j of the first questionnaire and option combination m of the second questionnaire for each option i. To obtain.

[0068] When the number of shared symptom options The larger the value, and the greater the difference in severity among all shared symptom options. The sum The smaller the value, the more similar the first questionnaire option combination j is to the second questionnaire option combination m. Therefore, the first selection similarity between the first questionnaire option combination j and the second questionnaire option combination m can be obtained. .

[0069] Get the severity difference size of all options. The maximum value in When the first questionnaire option combination j and the second questionnaire option combination m have similar first choices... The larger the value, the greater the difference in severity of target option i. With the maximum value The difference The smaller the number of hours, the more the patient's condition matches the situation in option combination m of the second questionnaire, and the more likely the target option i is to be a misselection, and the more it should be eliminated.

[0070] Therefore, the necessity of eliminating target option i in option combination j of the first questionnaire can be obtained. : ; Using the min-max normalization method to determine the necessity of elimination After normalization, we get Its range is [0,1]. That is... Corresponding to the degree of necessity for elimination .

[0071] When the necessity for removal exceeds the second preset threshold, that is, when When the original selection of target option i is removed from the first questionnaire option combination, the data for this option is modified to match the selection of the reference questionnaire option combination m. The modified patient questionnaire is then used for triage through a deep learning model to avoid the influence of abnormal data on triage decisions.

[0072] Through the above technical solution, this application solves the problem of triage result deviation caused by patients' misfilling of questionnaire options. By quantifying the differences in options and using a dynamic correction mechanism, the interference of noisy data on the model is reduced, and the accuracy and stability of triage decisions are improved.

[0073] In some embodiments, the judgment of the triage capability assessment module may include the following:

[0074] Obtain the degree of conformity of the first questionnaire option combination filled in by the target subject to different triage categories, and obtain the minimum value of all conformity degrees.

[0075] Obtain the mean value of the conformity of all triage categories.

[0076] The triage capability of the triage model is determined by using the degree of conformity, the minimum value of the degree of conformity, and the mean value of the degree of conformity.

[0077] The degree of conformity refers to the matching degree between the first questionnaire option combination and each triage category. This can be achieved using probability values ​​or similarity scores output by a deep learning model, quantifying the correlation between the questionnaire data and the disease characteristics of each department. The minimum degree of conformity is the lowest matching value among all triage categories. This can be achieved by iterating through the conformity degrees of all triage categories and selecting the minimum value, reflecting the model's weakest triage ability for the current questionnaire data. The mean degree of conformity is the arithmetic mean of the conformity degrees of all triage categories, achieved by summing them and dividing by the total number of triage categories, measuring the model's overall triage tendency for the current questionnaire data.

[0078] Specifically, the triage model's triage capability is determined through the following steps: First, based on the matching degree of each triage category according to the first questionnaire option combination, the overall underestimation is calculated. This indicator is determined by the sum of the deviations of the total number of triage categories from the minimum value of each matching degree, reflecting whether the model's triage results for the current questionnaire data have a generally low matching problem. Second, the similarity of the matching degree of all triage categories is calculated. This indicator is determined by the sum of the absolute deviations of each matching degree from the mean, used to determine whether the model's triage results have a risk of multi-departmental confusion. Finally, the similarity is multiplied by the overall underestimation to obtain a low triage capability performance index. When this index exceeds a preset threshold, the model is deemed to have insufficient triage capability for the current questionnaire data, triggering a preset triage process.

[0079] Through the above technical solution, this application effectively solves the problem of insufficient model triage capability for atypical symptom combinations. By simultaneously evaluating the discrete distribution characteristics of the degree of conformity and the overall deviation level, the uncertainty of the model in triage decision-making can be accurately identified, avoiding the risk of misdiagnosis caused by the model's over-reliance on a single feature. When the model's triage capability is insufficient, the system automatically triggers a preset triage mechanism to ensure that patients receive timely human intervention, thereby improving the reliability and security of the triage system.

[0080] In some embodiments, the triage capability of the triage model is determined by using the degree of conformity, the minimum value of the degree of conformity, and the mean value of the degree of conformity, and may also include the following operations.

[0081] Obtain the total number of triage categories, and use the total number of triage categories, the degree of conformity, and the minimum value of the degree of conformity to determine the overall degree of conformity of the first questionnaire option combination with respect to all triage categories.

[0082] By using the mean of the degree of conformity, the total number of triage categories, and the degree of conformity, the similarity of the degree of conformity of the first questionnaire option combinations to all triage categories is determined.

[0083] By using similarity and overall smallness, the low triage capability performance of the first questionnaire option combination was determined.

[0084] If the low triage capability exceeds the third preset threshold, it is determined that the triage model has low triage capability, and the combination of the first questionnaire options is used for preset triage.

[0085] The total number of triage categories refers to the number of pre-set departmental categories in the current system. This can be achieved by using the number of labels corresponding to different departments in the statistical model training data, quantifying the diversity of triage categories. Overall underestimation refers to the sum of differences between the minimum and the degree of conformity of the questionnaire option combinations for each triage category. This can be calculated by summing the differences between the degree of conformity for each triage category and the minimum value, reflecting the overall insufficient matching degree of the questionnaire option combinations within each triage category. Similarity refers to the dispersion of the distribution of the degree of conformity of the questionnaire option combinations for each triage category. This can be calculated by summing the absolute differences between the degree of conformity and the mean, measuring whether the difference in matching degree between different triage categories is significant. Low triage capability performance is a comprehensive indicator of insufficient reliability of the model's triage for the current questionnaire option combinations. This can be obtained by multiplying the overall underestimation by the similarity, used to determine whether the pre-set triage process needs to be activated.

[0086] Specifically, after a patient completes the questionnaire, the system first calculates the degree of conformity of the questionnaire to each triage category and extracts the minimum value. It should be noted that the degree of conformity for each triage category is calculated by inputting the completed questionnaire into the system's initial training model, which then outputs the degree of conformity for each triage category. The overall matching deviation of the questionnaire option combination across all triage categories is quantified by summing the differences between the total number of triage categories and the degree of conformity for each category. Simultaneously, the mean of the degree of conformity is calculated, and the uniformity of the matching distribution among triage categories is measured by summing the absolute differences between each degree of conformity and the mean. If the overall deviation is large and the similarity is high, it indicates that the model's matching degree for each triage category is generally low and the differences are small. In this case, the system determines that the model's triage capability for the current questionnaire is insufficient and triggers a preset triage mechanism. For example, when the normalized value of the low triage capability exceeds 0.8, the system automatically transfers the questionnaire to manual processing to avoid triage errors caused by ambiguity in model judgment.

[0087] Through the above technical solution, this application solves the problem of ambiguous judgment in complex symptom scenarios of traditional triage models. By dynamically evaluating the matching deviation and distribution characteristics between triage categories, it effectively identifies low-confidence triage results and switches to the preset triage process when the threshold is triggered, which significantly improves the accuracy of triage decision-making and the robustness of the system.

[0088] In some embodiments, the following operations may also be included.

[0089] Obtain the percentage of the number of times the target symptom cannot be triaged by the triage model out of the total number of times the target symptom is identified.

[0090] In response to the target symptom being untriageable, the system obtains the number of different triage departments corresponding to the highest degree of conformity and the preset triage departments, as well as the magnitude of the low triage ability performance corresponding to each target symptom and the target subject questionnaire option combination.

[0091] By utilizing the magnitude, frequency, and proportion of low triage capability performance, the degree of triage defect of the triage model for the target symptoms can be determined.

[0092] By utilizing the degree of triage defects, we determine the first set of training data that needs to be added to optimize the triage model.

[0093] The percentage of times the model cannot be effectively triaged refers to the proportion of times the model fails to effectively triage the target symptom out of the total number of occurrences of that symptom. This can be achieved by comparing the number of untriageable records to the total number of cases within the statistical period, quantifying the model's failure rate in triaging a specific symptom. The number of times the triage department differs refers to the number of times the model's recommended department differs from the manual triage department. This can be achieved by comparing the model's output with the results of manual review, reflecting the reliability defects of the model's triage results. The magnitude of low triage ability performance is a quantitative indicator of the difference in the model's conformity to questionnaire option combinations across different triage categories. This can be achieved by calculating the product of the conformity dispersion and the overall deviation, assessing the model's insufficient ability to distinguish complex symptoms. The degree of triage defect is a composite indicator formed by combining the frequency of untriageable cases, the triage error rate, and the model's insufficient distinguishing ability. This can be calculated by combining the above factors in a product form, used to identify symptom categories that require priority supplementation of training data.

[0094] Specifically, the system periodically analyzes triage records for the target symptom. When an untriageable case is found, it simultaneously acquires the difference between the manual review result and the model output. By calculating the proportion of untriageable cases to the total number of cases, combined with the number of differences between the model-recommended departments and the manual triage departments, and the low triage capability data of each untriageable case, a multi-dimensional triage defect assessment model is constructed. This assessment model couples the untriage frequency, the number of triage errors, and the degree of model discrimination insufficiency in a product form to generate a triage defect level index. When this index exceeds a preset threshold, the system automatically triggers a training data supplementation mechanism, selecting case data from other triage departments with high similarity to the symptom as new training samples. Incremental training optimizes the model's feature extraction capability and classification boundary clarity for the symptom.

[0095] For example, the model's triage capability is assessed based on the degree of consistency and similarity between the patient's questionnaire answers and different triage departments. Diseases requiring additional training data are identified based on the model's triage capability for different diseases within a short period and the level of triage confusion. When the model analysis of the patient's questionnaire results in low and similar degrees of consistency across different triage departments, it indicates that the current model's description of the disease characteristics corresponding to the patient's questionnaire symptoms is excessively similar to certain other diseases, lacking specific features corresponding to this disease. Therefore, the model's triage capability for the patient's questionnaire is low.

[0096] Then we have: The degree of conformity of the patient's first questionnaire option combination j to different triage categories k is obtained through the triage model. The minimum degree of conformity among all triage categories is obtained through comparison. The degree to which the patient's questionnaire option combination j conforms to all triage categories k can be obtained. Overall small . This represents the total number of triage categories.

[0097] Calculate the degree of conformity of the patient's first questionnaire option combination j with all triage categories k. mean size The degree to which the patient's questionnaire option combination j matches all triage categories k can be obtained. similarity .

[0098] When similarity The larger, and the smaller overall. The larger the value, the more similar the degree of consistency obtained by the model in analyzing the first questionnaire option combination j of the patient in each triage category, and the smaller the value, which proves that the current model training data has more defects in the inclusion of the patient's current condition, and the lower the model's ability to triage the first questionnaire option combination of the patient.

[0099] This reveals the model's low triage capability in response to the patient's first questionnaire option combination j. : ; Using the max-min normalization method to assess low triage capacity After normalization, we get Its range is [0,1]. Performance in response to low triage capacity .

[0100] When the low triage capacity performance exceeds the third preset threshold, the third preset threshold is set to 0.8, that is, when... If the current model is found to have a low triage capability for the first questionnaire option combination j answered by the patient, the questionnaire is pushed to medical staff for pre-defined triage, such as manual triage, in order to ensure that the patient is correctly triaged and avoid delays in the patient's condition.

[0101] Furthermore, the data used to train the model may not adequately describe certain diseases, leading to triage failures for some questionnaire option combinations. To improve the model's accuracy in triaging different patients' symptoms, its ability to diagnose different diseases can be analyzed at fixed time intervals based on the triage results of the current questionnaire option combinations. This analysis will serve as a reference for adding relevant training data later.

[0102] Therefore, the current version of the model is upgraded every six months to obtain the records of questionnaire option combinations that the model cannot triage within the past six months, as well as the manual triage department corresponding to each individual questionnaire option combination record, the patient's disease, and the disease corresponding to the questionnaire option combination that cannot be triaged as the disease to be supplemented, and the questionnaire option combination that cannot be triaged as the questionnaire data for the corresponding disease.

[0103] When calculating the number of times a patient's illness is considered as a target symptom, for example, if the target symptom is disease p, the triage model calculates the number of times the patient cannot be triaged. The total number of times the patient had the target symptom percentage .

[0104] When the model cannot triage disease p (target symptom), obtain the triage department corresponding to the highest degree of conformity. Count the number of differences between the triage department corresponding to the highest degree of conformity and the preset triage (manual triage) department. .

[0105] When triage is not possible, the magnitude of low triage ability performance corresponds to the combination of questionnaire options for each patient (target subject) for each disease p (each target symptom) in the case of non-triage. .

[0106] When the proportion The larger the number of different triage departments, the more frequent the visits. The greater the number of patients with no triage capacity, and the greater the number of patients with no triage capacity, the smaller the number of patients with no triage capacity corresponding to the patient questionnaire option combinations for each (j) disease p. The sum The larger the value, the worse the current model's ability to triage disease p (target symptom) becomes, and the more likely it is to fail to triage or make triage errors.

[0107] This allows us to determine the degree of deficiency in the triage model regarding disease p (target symptom). : ; Therefore, as one of the references for judging the necessity of supplementing the features corresponding to disease p in the model, the larger the value, the more problematic the current model is in recognizing disease p, the less accurate the triage of disease p, the greater the defect, and the more necessary it is to add the corresponding first training data; then the first training data can be used to optimize and train the triage model.

[0108] Through the above technical solution, this application realizes the automatic identification and targeted optimization of the symptoms of defects in the triage model, which significantly improves the accuracy of the model in triaging cross symptoms and rare cases, reduces the risk of misdiagnosis caused by imbalanced training data, and reduces the workload of manual screening of defective cases.

[0109] In some embodiments, the judgment of the data supplementation judgment module may include the following:

[0110] Obtain the number of first disease types and the number of triage departments for each triage department, and obtain the number of second disease types corresponding to all triage departments.

[0111] The difference between the degree of triage defect of the target symptom and the maximum value of the triage defect is obtained.

[0112] The necessity of feature supplementation for the triage model is determined by using the difference, the number of second disease categories, the number of triage departments, and the number of first disease categories.

[0113] The first number of disease categories refers to the number of disease types currently covered by the triage department for the target symptom. This can be achieved by statistically analyzing the department's historical case database and is used to measure the breadth of disease coverage for that department. The second number of disease categories refers to the total number of disease types corresponding to all triage departments within the system. This can be achieved by summarizing the disease classification lists of each department and is used to assess the overall disease identification range of the model. The triage defect severity difference refers to the difference between the current triage defect assessment value of the target symptom and the maximum defect value within the system. This can be achieved by calculating the arithmetic difference between the maximum defect severity value and the current value, reflecting the relative severity of the triage defect for that symptom within the system.

[0114] Specifically, when a target symptom shows a high degree of similarity to questionnaire options from other triage departments, the system will extract disease type data from the associated departments. By calculating a composite index of disease type coverage ratio and triage deficiency degree, a quantitative assessment value of the necessity for feature supplementation is formed. For example, when a department has a low disease coverage ratio and triage deficiency is close to the system's maximum value, the training data corresponding to that symptom will be prioritized for supplementation. The system continuously optimizes the model's ability to distinguish between overlapping symptoms by dynamically adjusting the distribution ratio of data from different departments.

[0115] In some specific implementations, when a respiratory symptom is detected to have a similar combination of options to a cardiovascular case, the system will compare the differences in the number of disease types in the two departments, and combine this with the ratio of the number of triage errors occurring for this symptom to the maximum number of error cases within six months to generate a feature supplementation priority score. If the score exceeds a set threshold, the system will automatically trigger the collection of case data corresponding to that symptom, and the newly added data will be included in the next round of model training.

[0116] Through the above technical solution, this application can accurately identify the weak links in the model caused by overlapping disease features or insufficient training samples, and supplement key case data in a targeted manner, thereby improving the accuracy of the triage system in identifying complex symptoms and reducing the risk of misdiagnosis of cross-departmental diseases.

[0117] In some embodiments, the necessity of feature supplementation for the triage model is determined by using the difference, the number of second disease types, the number of triage departments, and the number of first disease types, which may include the following.

[0118] Obtain the second choice similarity between the third questionnaire option combination corresponding to the target symptom and the fourth questionnaire option combination of other triage departments.

[0119] In response to the second choice similarity being greater than the fourth preset threshold, the number of first disease types and the corresponding number of triage departments corresponding to the fourth questionnaire option combinations of other triage departments are obtained.

[0120] The first value is determined by using the number of the first disease type and the number of the second disease type, and the second value is determined by using the number of triage departments and the difference.

[0121] Using the first and second values, determine the degree of necessity for supplementing the characteristics of the target symptom.

[0122] In response to the fact that the necessity of feature supplementation is greater than a fifth preset threshold, the corresponding feature training data for the target symptom is supplemented, and the triage model is optimized and trained using the disease to be supplemented.

[0123] The second similarity refers to the degree of similarity between the questionnaire option combination for the target symptom and other triage department questionnaire option combinations. Specifically, it can be calculated as the ratio of the number of shared symptom options to the difference in severity, used to identify cases with similar symptom characteristics across different triage departments. The first number of disease types refers to the number of disease types covered by the corresponding questionnaire option combinations for other triage departments. This can be achieved by statistically analyzing historical case data from different departments, used to measure the difference in disease coverage between different departments. The triage defect severity difference refers to the difference between the current symptom's triage defect severity and the maximum triage defect severity. This can be calculated using normalized numerical differences, used to reflect the current symptom's identification priority in the model.

[0124] Specifically, when the questionnaire options for a target symptom show high similarity to options from other triage departments, the system extracts data on the types of diseases and the number of departments in those departments. By comparing the differences in disease coverage and the distribution of the number of triage departments, and considering the difference between the current symptom's triage deficiency and the maximum deficiency value, the system comprehensively calculates the necessity of feature supplementation. For example, if a symptom is highly similar to case data from multiple departments, and its triage deficiency is close to the system's maximum value, the system will determine that specific training data corresponding to that symptom needs to be supplemented, thereby optimizing the model's ability to distinguish similar cases.

[0125] For example, some diseases are relatively rare, and triage data from the past six months may not adequately reflect the classification capabilities for certain diseases. Therefore, the differences in training data for different diseases can be used as another reference to determine the necessity of adding training data for different diseases.

[0126] Then, we obtain the second choice similarity between the third questionnaire option combination corresponding to the target symptom and the fourth questionnaire option combination of other triage departments, that is, we calculate the second choice similarity between the training data j of each questionnaire option combination corresponding to disease p (target symptom) and the training data m of the questionnaire option combination corresponding to other triage departments z. Size, where the second choice similarity Similarity to the first choice The calculation method is the same; please refer to the first choice similarity. The calculation process will not be elaborated here. And the maximum-minimum normalization pair is used... After normalization, we get Its range is [0,1]. Corresponding to the second choice similarity .

[0127] If the similarity of the second selection is greater than the fourth preset threshold (set to 0.8), then the selection is successful. The number of the first disease type corresponding to the fourth questionnaire option combination training data m for other triage departments >0.8 and the corresponding number of triage departments Get the number of second disease categories currently being treated by all triage departments. .

[0128] When the number of the first disease type The number of second disease types diagnosed and treated ratio The larger the size, the more triage departments are required. The more [symptoms], the greater the degree of triage defects in disease p (target symptom). Compared to the maximum level of triage defect The difference The smaller the number of diseases, the more diseases in the triage model are similar to the training data of disease p (target symptom), the more triage departments are involved, the greater the triage defects, and the more necessary it is to supplement the training data with features. It should be noted that the first number of disease types is calculated by determining the second choice similarity between the training data j of each questionnaire option combination corresponding to disease p (target symptom) and the training data m of the questionnaire option combination corresponding to other triage departments z. The number of training questionnaire option combinations with a second choice similarity greater than 0.8 is recorded as the first number of disease types. Therefore, the number of disease categories can be determined by analyzing the similarity between the questionnaire option combinations for the diseases to be added and the corresponding questionnaire option combinations for other triage departments. It should be noted that the target symptoms here refer to the diseases to be added.

[0129] This allows us to determine the degree of feature supplementation necessity for the triage model, which is also the degree of feature supplementation necessity for disease p (target symptom). : ; in, The first value, This is the second value.

[0130] Using min-max normalization to assess the necessity of feature supplementation After normalization, we get Its range is [0,1]. Corresponding feature supplementation necessity .

[0131] In response to a situation where the necessity of feature supplementation exceeds a fifth preset threshold, the fifth preset threshold is set to 0.8. Therefore, when... At this time, it is required to supplement the training data of the corresponding features of disease p (target symptom), and to optimize the triage model with the diseases to be supplemented, so as to ensure that the model can correctly triage disease p to the correct department.

[0132] In some embodiments, the questionnaires completed by different patients are triaged through an intelligent triage system using the above method. The combinations of questionnaire options for different patients and the corresponding triage results from the system are transmitted and stored in a database. The triage results of different patient questionnaire option combinations are displayed in tabular form on the staff's computer screen for reference during triage.

[0133] When the need for supplementing disease features is significant and training data for the corresponding features is required, questionnaire option combinations with a similarity greater than 0.8, along with their corresponding diseases and triage departments, are displayed to staff. This allows staff to query and add disease-distinguishing features based on other disease features with high similarity, thereby improving the model's ability to identify the disease and perform accurate triage operations.

[0134] Through the above technical solution, this application can accurately identify the problem of insufficient model triage ability caused by overlapping features of training data, and supplement key data in a targeted manner. For example, in cases where there are similar chest pain symptoms in respiratory and cardiology departments, the system can automatically screen out symptoms with high priority for feature supplementation, and strengthen the model's distinguishing ability by adding specific indicator data, thereby reducing the occurrence of cross-departmental triage errors.

[0135] The intelligent triage decision generation system for community medical terminals based on deep learning proposed in this application constructs a triage model through a training module, dynamically optimizes the identification of incorrect selections, assesses capabilities, and supplements features through an optimization module, and provides visualized results through a display module. It has the advantages of dynamically identifying incorrect questionnaire options, evaluating the model's triage capabilities in real time, and automatically supplementing training data defects, which can effectively improve triage accuracy and reduce the workload of medical staff.

[0136] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0137] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A community healthcare terminal triage decision-making intelligent generation system based on deep learning, characterized in that, include: The training module uses training data from different triage departments to train the model in order to obtain a triage model. The optimization module uses preset training data to filter the questionnaire data of the target audience; The triage model is judged based on the degree of conformity and similarity of the preset triage departments and the questionnaire data to obtain the triage capability; The degree of triage deficiency for diseases to be supplemented is determined by combining the number of times a disease cannot be triaged with the triage capacity. Based on the similarity between the questionnaire data of the disease to be supplemented and the questionnaire data of different triage departments, the necessity of supplementing the features of the disease to be supplemented is determined in combination with the degree of triage defects; based on the necessity of supplementing the features, the training data of the preset training model is supplemented.

2. The intelligent triage decision generation system for community medical terminals based on deep learning according to claim 1, characterized in that, The training process of the training module is as follows: Design a questionnaire that includes at least: symptom characteristics, medical history information, and basic information; All symptoms were given a questionnaire with options, and the options were assigned to the corresponding departments based on the combinations. The labeled data of suspected symptom types were also obtained. The labeled data is used as the training data to train the model, thereby obtaining the triage model.

3. The intelligent triage decision generation system for community medical terminals based on deep learning according to claim 1, characterized in that, The optimization module includes: A filtering module that uses preset training data to filter questionnaire data of the target object; The triage capability assessment module uses the degree of conformity and similarity of preset triage departments and the questionnaire data to assess the triage model and obtain its triage capability. The triage defect judgment module combines the number of times a disease cannot be triaged with the ability to triage to determine the degree of triage defect for the disease to be supplemented. The data supplementation judgment module determines the necessity of supplementing the features of the disease to be supplemented based on the similarity between the questionnaire data of the disease to be supplemented and the questionnaire data of different triage departments, combined with the degree of triage defects. The determining module supplements the training data of the preset training model based on the degree of necessity of the feature supplementation.

4. The intelligent triage decision generation system for community medical terminals based on deep learning according to claim 3, characterized in that, The filtering module includes: Obtain the number of common symptom options between the first questionnaire option combination filled out by the target subjects and the second questionnaire option combination of each training data; Obtain the magnitude of the severity difference for each option between the first questionnaire option combination and the second questionnaire option combination; The first selection similarity between the first questionnaire option combination and the second questionnaire option combination is determined by using the number of shared symptom options and the magnitude of the difference in severity. The necessity of eliminating the first questionnaire option combination is determined by using the severity difference magnitude and the first selection similarity. In response to the necessity of elimination exceeding a second preset threshold, the target option is removed from the first questionnaire option combination.

5. The intelligent generation system for triage decisions in community medical terminals based on deep learning according to claim 4, characterized in that, The step of determining the necessity of eliminating the first questionnaire option combination by utilizing the severity difference and the first selection similarity further includes: Get the maximum value of the severity difference among all options; By using the maximum value of the severity difference and the first selection similarity, the necessity of removing the target option from the first questionnaire option combination is determined.

6. The intelligent generation system for triage decisions in community medical terminals based on deep learning according to claim 3, characterized in that, The triage capability assessment module makes the following judgments: Obtain the degree of conformity of the first questionnaire option combination filled in by the target object to different triage categories, and obtain the minimum value of all the conformity degrees; Obtain the mean value of the conformity of all triage categories; The triage capability of the triage model is determined by using the degree of conformity, the minimum value of the degree of conformity, and the mean value of the degree of conformity.

7. The intelligent triage decision generation system for community medical terminals based on deep learning according to claim 6, characterized in that, The step of determining the triage capability of the triage model by utilizing the degree of conformity, the minimum value of the degree of conformity, and the mean value of the degree of conformity further includes: Obtain the total number of triage categories, and use the total number of triage categories, the degree of conformity, and the minimum value of the degree of conformity to determine the overall degree of conformity of the first questionnaire option combination with respect to all triage categories. Using the mean value of the degree of conformity, the total number of triage categories, and the degree of conformity, the similarity of the degree of conformity of the first questionnaire option combination to all triage categories is determined; Using the similarity and the overall smallness, the low triage capability performance of the first questionnaire option combination is determined; In response to the low triage capability exceeding a third preset threshold, the triage model is determined to have low triage capability, and the first questionnaire option combination is used for preset triage.

8. The intelligent triage decision generation system for community medical terminals based on deep learning according to claim 7, characterized in that, The triage defect judgment module includes: Obtain the percentage of the number of times the target symptom could not be triaged by the triage model out of the total number of times the target symptom was diagnosed; In response to the target symptom being untriageable, the number of different triage departments between the triage department corresponding to the maximum degree of conformity and the preset triage department is obtained, as well as the magnitude of the low triage ability performance corresponding to the target subject questionnaire option combination for each target symptom; The degree of triage deficiency of the triage model for the target symptoms is determined by using the magnitude of the low triage capability performance, the different frequency, and the proportion.

9. The intelligent generation system for triage decisions in community medical terminals based on deep learning according to claim 3, characterized in that, The data supplementation judgment module makes the following judgments: Obtain the number of first disease types and the number of triage departments for each triage department, and obtain the number of second disease types corresponding to all triage departments; Obtain the difference between the degree of triage deficiency of the target symptom and the maximum value of the degree of triage deficiency; The necessity of feature supplementation for the triage model is determined by using the difference, the number of the second disease types, the number of triage departments, and the number of the first disease types.

10. The intelligent generation system for triage decisions in community medical terminals based on deep learning according to claim 9, characterized in that, The process of supplementing the training data of the preset training model based on the necessity of the feature supplementation includes: Obtain the second choice similarity between the third questionnaire option combination corresponding to the target symptom and the fourth questionnaire option combination of other triage departments; In response to the second selection similarity being greater than a fourth preset threshold, the number of the first disease types and the corresponding number of the triage departments corresponding to the fourth questionnaire option combinations of other triage departments are obtained; A first value is determined using the number of the first disease types and the number of the second disease types, and a second value is determined using the number of triage departments and the difference. Using the first and second values, determine the degree of necessity for supplementing the features of the target symptom; In response to the fact that the necessity of feature supplementation is greater than a fifth preset threshold, the corresponding feature training data of the target symptom is supplemented, and the triage model is optimized and trained using the disease to be supplemented.

Images