Medical record homepage main diagnosis recommendation method and system based on rule-model collaboration

Abstract

The invention provides a medical record homepage main diagnosis recommendation method and system based on rule-model collaboration, and belongs to the technical field of medical information. The method comprises the following steps: extracting, analyzing and standardizing structured and unstructured data in a home page of a medical record to generate structured data for feature calculation; based on the structured data, constructing a multi-dimensional feature vector including clinical priority, ICD coding attributes, resource consumption and patient basic information; enhancing the multi-dimensional feature vector through a rule-model cooperation mechanism to form an enhanced feature vector; and inputting the enhanced feature vector into a trained machine learning sorting model to obtain a recommendation probability of each candidate main diagnosis, and outputting a main diagnosis recommendation list based on the recommendation probability. According to the invention, the accuracy and efficiency of main diagnosis selection are effectively improved, and the risk of medical insurance payment refusing and the workload of doctors are reduced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of medical information technology, and particularly relates to a rule-model collaborative-based main diagnosis recommendation method and system for a medical record home page. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute prior art.

[0003] The main diagnosis of the medical record home page is the core basis for medical data statistics, medical insurance settlement and clinical pathway management, and its selection needs to follow the clinical standard of "the greatest harm to the patient's health, the most consumption of medical resources and the longest hospitalization time".

[0004] However, in the current process of determining the main diagnosis of the medical record home page, the personal experience and subjective judgment of the clinician are mainly relied on, and there is a lack of objective and unified decision support. This easily leads to the problems of misjudging complications or comorbidities as the main diagnosis, and ignoring the direct cause of admission. According to statistics, the proportion of medical insurance refusal due to the wrong selection of the main diagnosis is high, which seriously affects the accuracy and reliability of the medical record data.

[0005] In addition, the current hospital generally uses a medical record management system or an electronic medical record system, the function of which is mostly limited to the query and input of ICD coding, and it is essentially a "coding dictionary", and lacks an active and intelligent recommendation mechanism for the selection of the main diagnosis. Doctors need to manually review and compare a large amount of patient data, which is tedious, time-consuming and easy to miss key information, and the work efficiency is low. At the same time, due to the periodic update of the ICD coding standard, it is difficult for clinicians to real-time and comprehensively master all the coding changes and the accurate correspondence of disease names, and the situation of "mismatch between diagnosis name and ICD code" often occurs, which affects the standardization of data reporting.

[0006] The medical record home page data contains structured information and unstructured information. The existing technology lacks effective means to deeply integrate and semantically understand these two types of data, and cannot build a comprehensive and multi-dimensional analysis model, resulting in single and one-sided basis for main diagnosis recommendation. SUMMARY

[0007] To overcome the above shortcomings of the prior art, the present application provides a rule-model collaborative-based main diagnosis recommendation method and system for a medical record home page, which can effectively integrate multi-source heterogeneous medical record data, realize automatic, high-precision and clinically standardized main diagnosis recommendation through deep collaboration of rules and models, thereby reducing the workload of doctors and improving the data quality of the medical record home page.

[0008] To achieve the above purpose, one or more embodiments of the present application provide the following technical solutions: The first aspect of the present application provides a rule-model collaborative-based main diagnosis recommendation method for a medical record home page; The rule-model collaborative-based main diagnosis recommendation method for a medical record home page comprises: structured and unstructured data in the medical record home page are extracted, analyzed and standardized to generate structured data for feature calculation; Based on the structured data, a multi-dimensional feature vector containing clinical priority, ICD coding attribute, resource consumption and patient basic information is constructed; The multi-dimensional feature vector is enhanced through a rule-model collaborative mechanism to form an enhanced feature vector; The enhanced feature vector is input into a trained machine learning ranking model to obtain the recommendation probability of each candidate main diagnosis, and a main diagnosis recommendation list is output based on the recommendation probability.

[0009] As a further technical solution, the structured and unstructured data in the medical record home page are extracted, analyzed and standardized to generate structured data for feature calculation, comprising: A main diagnosis related data dictionary library is constructed by defining core data fields and establishing field and recommendation dimension mapping relationships; A double-layer fusion architecture of regular expression pre-extraction and medical NLP deep analysis is used to extract unstructured data; ICD coding standardization, resource consumption quantification calculation and disease severity quantification calculation are used to realize data standardization verification.

[0010] As a further technical solution, a multi-dimensional feature vector containing clinical priority, ICD coding attribute, resource consumption and patient basic information is constructed, comprising: A multi-dimensional feature vector containing clinical priority, ICD coding attribute, resource consumption and patient basic information is constructed; Normalization is used to standardize each feature item in the feature vector, and the feature value is mapped to a specific interval; Weakly correlated features are removed through variance analysis calculation for feature importance pre-screening.

[0011] As a further technical solution, the multi-dimensional feature vector is enhanced through a rule-model collaborative mechanism to form an enhanced feature vector, comprising: Based on the pre-defined clinical rule library, the candidate main diagnosis set is selected from the original disease diagnosis list based on the candidate pool transmission mechanism; The matching degree of the candidate main diagnosis and the clinical rule library is quantified as a rule matching feature, and is supplemented to the multi-dimensional feature vector to form an enhanced feature vector.

[0012] As a further technical solution, the machine learning ranking model is an XGBoost model, whose objective function includes a loss function and a regularization term. The objective function is minimized by iteratively adding decision trees, thereby outputting the recommendation probability of the candidate main diagnosis.

[0013] As a further technical solution, a feedback optimization step is also included, in which the final master diagnosis result confirmed by the user and its corresponding enhanced feature vector are used as new samples and fed back to the model training set for iterative optimization of the machine learning ranking model.

[0014] The second aspect of this invention provides a rule-model collaborative primary diagnosis recommendation system for medical record homepages.

[0015] A rule-based model-driven system for recommending the primary diagnosis of a patient's medical record, including: The data processing module is configured to extract, parse, and standardize the structured and unstructured data in the medical record front page to generate structured data for feature calculation. The feature vector construction module is configured to: construct a multi-dimensional feature vector containing clinical priority, ICD coding attributes, resource consumption, and patient basic information based on the structured data; The rule-model collaborative recommendation module is configured to enhance the multi-dimensional feature vector through a rule-model collaborative mechanism to form an enhanced feature vector; The result output module is configured to: input the enhanced feature vector into a trained machine learning ranking model to obtain the recommendation probability of each candidate main diagnosis, and output a main diagnosis recommendation list based on the recommendation probability.

[0016] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the rule-model collaborative method for recommending the primary diagnosis of a medical record homepage as described in the first aspect of the present invention.

[0017] A fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the rule-model collaborative method for recommending the primary diagnosis of a medical record homepage as described in the first aspect of the present invention.

[0018] The above one or more technical solutions have the following beneficial effects: The rule layer and the model layer in the application are in deep coordination rather than simple superposition. The rule layer first performs preliminary screening on the diagnosis list according to authoritative clinical norms, ensures compliance of the recommended basis, effectively filters options obviously not meeting the main diagnosis standard, and eliminates common sense errors from the source. The model layer can accurately quantify the priority of each candidate diagnosis through joint learning of multi-dimensional feature vectors and rule matching features, so as to make a comprehensive judgment closer to the level of clinical experts in complex cases. Through the above mechanism, the system recommendation result has rigidity of specification requirement and flexibility of judgment in complex scenarios, so as to maintain the overall error rate of main diagnosis selection at a very low level. This not only directly improves the data quality of the medical record home page, provides a reliable cornerstone for medical statistics and clinical research, but more importantly, it can significantly reduce medical insurance rejection and reduction caused by improper selection of main diagnosis, and avoid potential economic losses and compliance risks for medical institutions.

[0019] The application solves the problem of automatic analysis of unstructured text information by fusing the preprocessing module of regular expressions and medical NLP technology, and realizes accurate extraction of key medical entities and their semantic relationships from complex clinical records. At the same time, the system automatically completes a series of work such as ICD coding standardization, resource consumption quantification, and disease severity score calculation, which originally needs to be manually queried or calculated by doctors. Doctors no longer need to repeatedly switch and check between multiple systems and multiple pages, greatly reducing the time of main diagnosis selection and reducing the burden and decision-making pressure of doctors.

[0020] The application continuously feeds back the final confirmation result of the doctor as a high-quality annotation sample to the rule base and model training set of the system. The rule base can be supplemented and fine-tuned according to the feedback, and the model can adapt to changes in clinical practice and new data patterns through regular retraining, so as to realize continuous and steady improvement of the recommendation accuracy.

[0021] The advantages of the additional aspects of the application will be partially given in the following description, partially become obvious from the following description, or be learned by the practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings, which form a part of the specification, are included to provide a further understanding of the application and are incorporated herein in conjunction with the description of the application. The schematic embodiments of the application and their descriptions are used to explain the application and do not constitute an improper limitation on the application.

[0023] Figure 1 The method flowchart of the first embodiment.

[0024] Figure 2 The system structure diagram of the second embodiment. DETAILED DESCRIPTION

[0025] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. 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 application belongs.

[0026] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application.

[0027] In the case of no conflict, the embodiments in the application and the features in the embodiments can be combined with each other.

[0028] Embodiment one The embodiment discloses a rule-model collaborative medical record home page main diagnosis recommendation method, first, the structured and unstructured data in the medical record home page are fused, processed and standardized, a multi-dimensional feature set covering clinical priority, ICD coding, resource consumption and patient basic information is constructed; then, the candidate main diagnosis is preliminarily screened out from the diagnosis list based on clinical norms through the rule layer, and the rule matching degree is converted into model features through a feature enhancement mechanism; finally, the integrated learning algorithm is used in the model layer to probabilistically sort and output the candidate diagnosis, and the ICD coding automatic checking is realized. Through the closed-loop cooperation of rules and models, the recommended results ensure clinical compliance and adaptability to complex scenarios, significantly improve the accuracy and efficiency of main diagnosis selection, and reduce the medical insurance risk.

[0029] Further, as shown in Figure 1 The rule-model collaborative medical record home page main diagnosis recommendation method specifically comprises the following steps: Step S1, the structured and unstructured data in the medical record home page are extracted, analyzed and standardized, and structured data used for feature calculation are generated. Specifically, in step S1, the following specific contents are further included: Step S11, a main diagnosis related data dictionary library is constructed by defining core data fields and establishing field and recommendation dimension mapping relationship. The dictionary library includes the unified semantics of the main diagnosis recommendation core dimension, the mapping rule of the original data field to the recommendation dimension, and the standardized expression of the storage medical term. By simultaneously processing the semantic mapping of the structured data (such as age, gender, ICD coding) containing the medical record home page, and the unstructured text (such as admission condition description, operation record), the dictionary library is constructed for subsequent data extraction and feature construction.

[0030] Specifically, the core data fields are defined, including "chief complaint (text), main diagnosis and treatment purpose (text), disease diagnosis list (diagnosis name and ICD coding), treatment measures (such as operation, medication, examination item), hospitalization days, itemized medical expenses, disease severity score.

[0031] Establish field-recommendation dimension mapping relationships, such as "treatment measures → clinical priority dimension", "medical expenses → resource consumption dimension", and "ICD code → code matching dimension", to ensure that each field serves the main diagnosis recommendation logic. Based on the relevant hospitalization case data filling specification and the core logic of main diagnosis selection, the fields are divided into four types of recommendation dimensions: Clinical priority dimension: associated with admission reason, main treatment, and disease severity, such as treatment measures (surgery / medication) directly mapping to clinical priority dimension - main treatment relevance; ICD code dimension: associated with coding hierarchy and acute / chronic properties, such as ICD code directly mapping to ICD code dimension - ICD hierarchy; Resource consumption dimension: associated with expenses, days, and treatment item proportion, such as medical expenses mapping to resource consumption dimension - expense proportion; Patient basis dimension: associated with age, gender, and underlying diseases, such as age mapping to patient basis dimension - age segment.

[0032] Step S12, using a double-layer fusion architecture of regular expression pre-extraction and medical NLP deep analysis to extract unstructured data.

[0033] For unstructured text such as admission condition description and surgery records, a double-layer fusion architecture of regular expression pre-extraction and medical NLP deep analysis is used to solve the limitations of single technology in medical text processing.

[0034] Regular expression pre-extraction (rule layer) is used to quickly locate fixed-format key information in the text, reducing the computational load of the NLP model.

[0035] Specifically, based on the writing specifications of clinical documents, multiple regular templates are constructed. For example, disease-treatment association templates, disease progression templates, and treatment outcome templates.

[0036] Example: For the text "patient admitted due to acute myocardial infarction, underwent PCI surgery, and has a history of hypertension", regular expression can directly extract: admission cause = acute myocardial infarction, main surgery = PCI surgery, and underlying disease = hypertension.

[0037] Further, medical NLP deep analysis is used to extract implicit clinical associations from non-fixed format text such as ambiguous expressions and omitted sentences. A pre-trained medical language model is used to identify entities such as "disease, symptom, and treatment method" in the text, and entity relationship extraction is performed by constructing clinical associations between entities. Subsequently, context disambiguation is used to solve the problem of polysemy.

[0038] By fusing the logic, the regular extraction result is directly used as a structured field, and the NLP parsing result is used as a supplement. If the hospital cause extracted by regular extraction conflicts with the main symptom identified by NLP, manual confirmation is triggered, and the corrected result is fed back to the model optimization template.

[0039] In step S13, data standardization verification is realized by ICD coding standardization, resource consumption quantification calculation and disease severity quantification calculation.

[0040] Firstly, ICD coding standardization is performed by connecting related standard databases to map the old codes filled by doctors into standard codes and eliminate invalid codes.

[0041] Resource consumption quantification calculation is performed by calculating the proportion of disease-related days in total hospitalization days and the proportion of disease-related expenses in total expenses, and the severity of the disease is reflected by APACHE II score.

[0042] In step S2, based on the structured data, a multi-dimensional feature vector including clinical priority, ICD coding attribute, resource consumption and patient basic information is constructed.

[0043] Based on the structured data of step S1, a feature set of four core dimensions is constructed, and the feature vector is denoted as , wherein represents the clinical priority, represents the ICD code, represents the resource consumption, represents the patient's basic information.

[0044] For the above feature vector, Min-Max normalization is used to standardize each feature item, and the feature value is mapped to a specific interval.

[0045] Feature importance pre-screening is performed by variance analysis calculation, and weakly correlated features are removed. Specifically, the core formula of variance analysis is:

[0046] , wherein is the between-group sum of squares; is the within-group sum of squares; is the number of groups; is the total sample size. By retaining the top 80% of features in terms of F value, weakly correlated features are removed.

[0047] In step S3, the multi-dimensional feature vector is enhanced by a rule-model collaborative mechanism to form an enhanced feature vector.

[0048] From the disease diagnosis list, candidate main diagnoses are selected and a clinical rule library is constructed, and the core rules include: (1) Diseases with "direct cause of admission + primary treatment target" should be given priority for inclusion; (2) Diseases with a resource consumption ratio of ≥60% are included; (3) Acute diseases take precedence over chronic diseases; (4) Exclude cured diseases and symptomatic diagnoses.

[0049] A candidate pooling mechanism is employed to select 3-5 candidate primary diagnoses that meet clinical criteria from the original disease diagnosis list using the four rule bases mentioned above. These diagnoses are then directly used as input to the model layer. Simultaneously, if a diagnosis triggers the fourth rule base, such as "postoperative incision infection cured" or "fever of unknown origin," it is directly discarded and not included in the model layer calculation, thus avoiding unnecessary computational waste.

[0050] Finally, the filtering results of the rule layer are transformed into rule matching features and added to the multi-dimensional feature vector to improve the model's attention to clinical norms.

[0051] Step S4: Input the enhanced feature vector into the trained machine learning ranking model to obtain the recommendation probability of each candidate main diagnosis, and output the main diagnosis recommendation list based on the recommendation probability.

[0052] Among them, the machine learning ranking model adopts the XGBoost model. The core function of the XGBoost model is to perform probability quantification and ranking of the candidate diagnoses selected by the rule layer. The specific data flow is divided into four stages: input preparation, model training, prediction ranking, and output feedback.

[0053] Specifically, the XGBoost model training process includes: integrating multi-dimensional feature vectors with added rule-matching features into a sample matrix, with each row corresponding to one candidate diagnosis and each column corresponding to one feature.

[0054] Collect qualified medical records that have passed the medical record quality review within the past 3 years, label each candidate diagnosis with a real label, and divide them into training set, validation set and test set in a 6:2:2 ratio; Construct the objective function:

[0055] in, This is a real label; This is the predicted value for the (t-1)th iteration; For the t-th decision tree, pair of samples The predicted contribution; This is the tree complexity penalty coefficient; This represents the total number of decision trees; The leaf node weight decay coefficient; Let be the predicted weight for the j-th node; Cross-entropy loss is used to measure the model's predicted probability. With real labels The deviation. ; This is a regularization term used to prevent the model from overfitting.

[0056] The decision tree iterative learning process is performed. For the first tree: samples are partitioned based on resource consumption characteristics, and samples with a cost ratio exceeding a set threshold are classified as high-priority nodes and assigned higher initial weights. For the second tree, high-priority nodes are further divided into extremely high-priority nodes based on clinical priority characteristics. For the t-th tree, all features are combined to correct the prediction bias of the first t-1 trees, and the final output is the primary diagnosis probability for each candidate diagnosis.

[0057] A 5-fold cross-validation optimization method is used to maximize the value of the validation set, and the optimal value is determined through iteration.

[0058] The candidate diagnostic feature matrix of new cases is input into the trained XGBoost model, which outputs the principal diagnosis probability of each candidate diagnosis, sorted in descending order of probability P, and outputs a recommendation list. At this point, if a high-probability conflict is detected, core discrepancy features are displayed to assist the doctor's decision-making.

[0059] Show the model's recommended results to doctors, for example: Recommendation 1: Acute ST-segment elevation myocardial infarction (ICD code I21.001), rationale: R1 match (direct cause of hospitalization + PCI treatment), cost accounts for 72%, model prediction probability 0.93; Recommendation 2: Stage 3 hypertension (ICD code I10.x05), reason for recommendation: the correlation score between the underlying disease and this disease is 0.6, and the model prediction probability is 0.38.

[0060] Once the doctor confirms the final primary diagnosis, the candidate diagnostic features and the doctor's confirmation result are further used as new samples to supplement the model training set for the next model iteration.

[0061] Furthermore, during the result output process, ICD encoding verification is performed, and a standard encoding list is popped up by calculating the encoding matching degree.

[0062] Example 2 This embodiment discloses a medical record homepage main diagnosis recommendation system based on rule-model collaboration; like Figure 2 As shown, the rule-model collaborative medical record homepage main diagnosis recommendation system includes: The data processing module is configured to extract, analyze and standardize structured and unstructured data in the medical record front page, and generate structured data for feature calculation. The feature vector construction module is configured to construct a multi-dimensional feature vector containing clinical priority, ICD coding attribute, resource consumption and patient basic information based on the structured data. The rule-model collaborative recommendation module is configured to enhance the multi-dimensional feature vector through a rule-model collaborative mechanism to form an enhanced feature vector. The result output module is configured to input the enhanced feature vector into the trained machine learning ranking model to obtain a recommendation probability of each candidate primary diagnosis, and output a primary diagnosis recommendation list based on the recommendation probability.

[0063] Further, the rule-model collaborative medical record front page primary diagnosis recommendation system further includes a feedback optimization module, which feeds the user-confirmed final primary diagnosis result and its corresponding enhanced feature vector as new samples to the model training set for iterative optimization of the machine learning ranking model.

[0064] Embodiment three The purpose of this embodiment is to provide a computer-readable storage medium.

[0065] A computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the steps in the rule-model collaborative medical record front page primary diagnosis recommendation method of embodiment 1.

[0066] Embodiment four The purpose of this embodiment is to provide an electronic device.

[0067] An electronic device includes a memory, a processor, and a program stored on the memory and executable on the processor, and the processor executes the program to implement the steps in the rule-model collaborative medical record front page primary diagnosis recommendation method of embodiment 1.

[0068] The steps and methods involved in the above embodiments two, three and four correspond to embodiment one, and the specific implementation can refer to the relevant description part of embodiment one. The term "computer-readable storage medium" should be understood to include a single medium or multiple media of one or more instruction sets; it should also be understood to include any medium that can store, encode or carry instruction sets for execution by a processor and make the processor execute any method in the present application.

[0069] Those skilled in the art should understand that the modules or steps of the present application described above can be realized by a general computer device, or alternatively, they can be realized by program codes executable by a computing device, so that they can be stored in a storage device and executed by a computing device, or they can be respectively manufactured into individual integrated circuit modules, or a plurality of modules or steps among them can be manufactured into a single integrated circuit module. The present application is not limited to any specific combination of hardware and software.

[0070] The specific embodiments of the present application described above in conjunction with the accompanying drawings are not intended to limit the protection scope of the present application. Those skilled in the art should understand that various modifications or changes made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the protection scope of the present application.

Claims

1. A rule-model collaborative-based method for recommending a primary diagnosis on a medical record home page, characterized in that, The method comprises the following steps: extracting, analyzing and standardizing the structured and unstructured data in the front page of the medical record to generate structured data for feature calculation; based on the structured data, a multi-dimensional feature vector containing clinical priority, ICD coding attribute, resource consumption and patient basic information is constructed; the multi-dimensional feature vector is enhanced through a rule-model collaborative mechanism to form an enhanced feature vector; the enhanced feature vector is input into a trained machine learning ranking model to obtain the recommendation probability of each candidate primary diagnosis, and a primary diagnosis recommendation list is output based on the recommendation probability. 2.The rule-model collaboration based method for recommending a primary diagnosis on a medical record form according to claim 1, wherein, The structured and unstructured data in the front page of the medical record are extracted, analyzed and standardized to generate structured data for feature calculation, including: a primary diagnosis related data dictionary library is constructed by defining core data fields and establishing field and recommended dimension mapping relationships; a double-layer fusion architecture of regular expression pre-extraction and medical NLP deep analysis is used to extract unstructured data; data standardization verification is realized through ICD coding standardization, resource consumption quantification calculation and disease severity quantification calculation. 3.The rule-model collaboration based method for recommending a primary diagnosis on a case history form according to claim 1, wherein, The multi-dimensional feature vector containing clinical priority, ICD coding attribute, resource consumption and patient basic information is constructed, including: a multi-dimensional feature vector containing clinical priority, ICD coding attribute, resource consumption and patient basic information is constructed; normalization is used to standardize each feature item in the feature vector, and the feature value is mapped to a specific interval; feature importance pre-screening is performed through variance analysis calculation to eliminate weakly correlated features. 4.The rule-model collaboration based method for recommending a primary diagnosis on a medical record form according to claim 1, wherein, The multi-dimensional feature vector is enhanced through a rule-model collaborative mechanism to form an enhanced feature vector, including: based on the pre-defined clinical rule library, the candidate primary diagnosis set is selected from the original disease diagnosis list based on the candidate pool transmission mechanism; the matching degree of the candidate primary diagnosis and the clinical rule library is quantified as a rule matching feature, and is supplemented to the multi-dimensional feature vector to form an enhanced feature vector. 5.The rule-model collaboration based method for recommending a primary diagnosis on a case history form according to claim 1, wherein, The machine learning ranking model is an XGBoost model, and its objective function contains a loss function and a regularization term. Decision trees are added iteratively to minimize the objective function, thereby outputting the recommendation probability of the candidate primary diagnosis. 6.The rule-model collaboration based method for recommending a main diagnosis on a case history form according to claim 1, wherein, It also includes a feedback optimization step, in which the final primary diagnosis result confirmed by the user and its corresponding enhanced feature vector are fed back to the model training set for iterative optimization of the machine learning ranking model.

7. A rule-model collaborative-based system for recommending a main diagnosis on a medical record first page, characterized in that, The method comprises the following steps: a data processing module configured to extract, analyze and standardize the structured and unstructured data in the front page of the medical record to generate structured data for feature calculation; a feature vector construction module configured to construct a multi-dimensional feature vector containing clinical priority, ICD coding attribute, resource consumption and patient basic information based on the structured data; a rule-model collaborative recommendation module configured to enhance the multi-dimensional feature vector through a rule-model collaborative mechanism to form an enhanced feature vector; The result output module is configured to input the enhanced feature vector into a trained machine learning ranking model to obtain a recommended probability of each candidate primary diagnosis, and output a primary diagnosis recommendation list based on the recommended probability. 8.The rule-model collaboration based system for recommending a main diagnosis on a case history form according to claim 7, wherein, A feedback optimization module is further included, which feeds a user-confirmed final primary diagnosis result and its corresponding enhanced feature vector as a new sample to a model training set for iterative optimization of the machine learning ranking model.

9. A computer-readable storage medium having stored thereon a program, characterized in that, The program, when executed by a processor, implements the steps in the rule-model collaborative-based primary diagnosis recommendation method for a medical record header as claimed in any one of claims 1-6.

10. An electronic device comprising a memory, a processor, and a program stored on the memory and executable on the processor, characterized in that, The processor, when executing the program, implements the steps in the rule-model collaborative-based primary diagnosis recommendation method for a medical record header as claimed in any one of claims 1-6.

Images