Inspection Item Recommendation Method and Auxiliary Decision-making System Based on Bayesian Network

Abstract

The present application provides a method for recommending examination items based on a Bayesian network and an auxiliary decision-making system. The method includes: obtaining a first conditional probability distribution of the examined items for the target disease, a second conditional probability distribution of the examined items for the to-be-examined items, and a third conditional probability distribution of the examined items for the to-be-examined items based on the target disease according to the disease occurrence probability grading corresponding to the examined items and the disease occurrence probability grading corresponding to the to-be-examined items; calculating a fourth conditional probability distribution of the to-be-examined items for the target disease based on the examined items according to the first conditional probability distribution, the second conditional probability distribution, and the third conditional probability distribution; and selecting the first priority recommended examination items from the to-be-examined items according to the fourth conditional probability distribution and the confidence level corresponding to the to-be-examined items. By adopting the above method, the present application solves the problem that the existing disease screening methods fail to fully utilize the known patient diagnosis data, resulting in inaccurate recommendation results.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly relates to a method for recommending examination items based on a Bayesian network and an auxiliary decision-making system. Background Art

[0002] In traditional clinical disease diagnosis and screening scenarios, doctors often give a possible diagnosis by asking about the existing symptoms of patients and combining their own domain knowledge and experience. At the same time, some necessary examination items are recommended to improve the physical symptoms of patients. Then, doctors make medical diagnoses and suggestions based on clinical experience and the report results of the current examination items of patients. Since the diagnostic experience and preferences of each doctor are different, some doctors may ask patients to undergo all relevant examinations. This screening plan not only brings great economic pressure to patients, but also causes harm to patients' bodies.

[0003] With the development of artificial intelligence technology, the application of artificial intelligence technology in the medical field has developed rapidly. Disease screening based on medical electronic record data and using AI technology has become a hot research direction in the field of medical applications. However, in related technologies, these disease screening methods have not fully utilized the known patient diagnosis data to recommend detection items, which also makes the recommended results inaccurate.

[0004] Currently, there is an urgent need for a method for recommending examination items based on a Bayesian network and an auxiliary decision-making system to solve the problems existing in related technologies. Summary of the Invention

[0005] This application provides a method for recommending examination items based on a Bayesian network and an auxiliary decision-making system, which is used to solve the problem that in related technologies, disease screening methods have not fully utilized the known patient diagnosis data to recommend detection items, which also makes the recommended results inaccurate.

[0006] The first aspect of the present application provides a method for recommending examination items based on a Bayesian network. The method includes: obtaining real-time examination item data of a target patient, where the real-time examination item data includes one or more examined item data, and the examined item data includes one or more types of multimodal data; inputting the examined item data into a disease occurrence probability prediction model corresponding to the examined item to obtain the disease occurrence probability grading and confidence level corresponding to the examined item; for an item to be examined, setting the disease occurrence probability grading corresponding to the item to be examined and the confidence level corresponding to the item to be examined; inputting the disease occurrence probability grading corresponding to the examined item and the disease occurrence probability grading corresponding to the item to be examined into a preset Bayesian network model to obtain a first conditional probability distribution of the examined item on the target disease, a second conditional probability distribution of the examined item on the item to be examined, and a third conditional probability distribution of the examined item on the item to be examined based on the target disease. The preset Bayesian network model is trained according to the disease occurrence probability grading and disease labels corresponding to a variety of historical examination item data; calculating a fourth conditional probability distribution of the item to be examined on the target disease based on the examined item according to the first conditional probability distribution, the second conditional probability distribution, and the third conditional probability distribution; selecting a first priority recommended examination item from the items to be examined according to the fourth conditional probability distribution and the confidence level corresponding to the item to be examined.

[0007] Optionally, the step of selecting a first priority recommended examination item from the items to be examined according to the fourth conditional probability distribution and the confidence level corresponding to the item to be examined specifically includes:

[0008] obtaining the KL divergence of the fourth conditional probability distribution of the item to be examined on the target disease;

[0009] performing weighted calculation on the KL divergence according to the confidence level corresponding to the item to be examined to obtain calculation results of multiple items to be examined, sorting the calculation results from high to low to obtain a first priority recommended item;

[0010] determining the first priority recommended item as the item to be preferentially recommended for examination.

[0011] Optionally, the multimodal data includes one or more of structured data, text data, and image data.

[0012] Optionally, input the data of the items to be examined into the disease occurrence probability prediction model corresponding to the items to be examined, and obtain the disease occurrence probability grading and confidence level corresponding to the items to be examined, which specifically includes: according to the data types of the multimodal data included in the data of the items to be examined, configure multiple encoders for the multimodal data, one encoder for one type of multimodal data, and the multiple encoders are used to process the multimodal data in the data of the items to be examined; obtain the initial disease occurrence probability according to the output results of the multiple encoders; discretize the initial disease occurrence probability according to multiple preset disease occurrence probability thresholds to obtain the disease occurrence probability grading corresponding to the items to be examined; normalize the reciprocal of the average cross-entropy loss of the disease occurrence probability prediction model corresponding to the items to be examined to obtain the confidence level corresponding to the items to be examined.

[0013] Optionally, according to the multimodal data types of the data of the items to be examined, configure multiple encoders for the data of the items to be examined, which specifically includes: perform feature extraction on the data of the items to be examined to obtain various feature data of the data of the items to be examined; among them, when the feature data is structured feature data, use the first encoder to process the structured feature data, and the first encoder is encoded according to a multi-layer perceptron; when the feature data is text feature data, use the second encoder to process the text feature data, and the second encoder is encoded according to a multi-view convolutional neural network or a CNN neural network; when the feature data is image feature data, use the third encoder to process the image feature data, and the third encoder is encoded according to a residual neural network.

[0014] Optionally, for the items to be examined, set the disease occurrence probability grading corresponding to the items to be examined and the confidence level corresponding to the items to be examined, which specifically includes: allocate a disease occurrence probability prediction model corresponding to each item to be examined to obtain multiple disease occurrence probability prediction models; set a preset value as the disease occurrence probability grading corresponding to any one of the items to be examined; obtain the average cross-entropy loss of any one of the disease occurrence probability prediction models; normalize the reciprocal of any one of the average cross-entropy losses to obtain the confidence level corresponding to any one of the items to be examined.

[0015] Optionally, the method further includes: obtaining the new inspection item data corresponding to the first preferentially recommended inspection item; according to the new inspection item data, updating the first conditional probability distribution of the inspected items for the target disease, the second conditional probability distribution of the inspected items for the items to be inspected, and the third conditional probability distribution of the inspected items for the items to be inspected based on the target disease; calculating the fourth conditional probability distribution of the remaining items to be inspected for the target disease based on the inspected items according to the updated first conditional probability distribution, second conditional probability distribution, and third conditional probability distribution as the updated fourth conditional probability distribution, where the remaining items to be inspected do not include the first preferentially recommended inspection item; selecting a second preferentially recommended inspection item from the remaining items to be inspected according to the updated fourth conditional probability distribution and the confidence levels corresponding to the remaining items to be inspected; obtaining the preferentially recommended inspection items multiple times to obtain the recommended inspection item path for the target disease.

[0016] The second aspect of the present application provides an auxiliary decision-making system, which includes a collection unit, a first model application unit, a second model application unit, and a sorting and recommendation unit.

[0017] The collection unit is used to obtain the real-time inspection item data of the target patient, and the real-time inspection item data includes one or more inspected item data, and the inspected item data includes one or more types of multimodal data;

[0018] The first model application unit is used to input the inspected item data into the disease occurrence probability prediction model corresponding to the inspected item to obtain the disease occurrence probability and confidence level corresponding to the inspected item; for the items to be inspected, set the disease occurrence probability grading corresponding to the items to be inspected and the confidence level corresponding to the items to be inspected;

[0019] The second model application unit is used to input the disease occurrence probability grading corresponding to the inspected item and the disease occurrence probability grading corresponding to the item to be inspected into a preset Bayesian network model to obtain the first conditional probability distribution of the inspected item for the target disease, the second conditional probability distribution of the inspected item for the item to be inspected, and the third conditional probability distribution of the inspected item for the item to be inspected based on the target disease, and the preset Bayesian network model is trained according to the disease occurrence probability grading and disease labels corresponding to a variety of historical inspection item data;

[0020] The sorting and recommendation unit is used to calculate the fourth conditional probability distribution of the item to be inspected for the target disease based on the inspected item according to the first conditional probability distribution, the second conditional probability distribution, and the third conditional probability distribution; and is also used to select a first preferentially recommended inspection item from the items to be inspected according to the fourth conditional probability distribution and the confidence level corresponding to the item to be inspected.

[0021] Optionally, the auxiliary decision-making system further includes a path planning unit.

[0022] The path planning unit is configured to obtain the new inspection item data corresponding to the first preferentially recommended inspection item; update the first conditional probability distribution of the inspected items for the target disease, the second conditional probability distribution of the inspected items for the to-be-inspected items, and the third conditional probability distribution of the inspected items for the to-be-inspected items based on the target disease according to the new inspection item data, calculate the fourth conditional probability distribution of the remaining to-be-inspected items for the target disease based on the inspected items as the updated fourth conditional probability distribution according to the updated first conditional probability distribution, second conditional probability distribution, and third conditional probability distribution, where the remaining to-be-inspected items do not include the first preferentially recommended inspection item; select the second preferentially recommended inspection item again from the remaining to-be-inspected items according to the updated fourth conditional probability distribution and the confidence corresponding to the remaining to-be-inspected items; obtain the preferentially recommended inspection items multiple times to obtain the recommended inspection item path for the target disease.

[0023] A third aspect of the present application provides an electronic device, which includes a processor, a user interface, a network interface, and a memory. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method in any one of the above.

[0024] A fourth aspect of the present application provides a computer-readable storage medium, which stores instructions that, when executed, execute the method in any one of the above.

[0025] Compared with the related art, the beneficial effects of the present application include: 1. The present application constructs a unified two-stage model framework through a deep neural network and a Bayesian network, which can utilize multi-modal data in the scenario of missing modalities, make full use of the known patient diagnosis data to recommend inspection items, and improve the accuracy of the recommendation results.

[0026] 2. The present application defines a quantifiable index using the KL divergence to characterize the addition of the disease screening effect, and combines the Bayesian network model to solve the problem of personalized inspection item path recommendation in disease screening. It solves the problem that the existing research solution of directly using multi-modal data to construct a diagnostic model can only perform single-point inspection item recommendation and cannot achieve personalized inspection path recommendation.

[0027] 3. The difference in the contribution degrees of different examination items reflected in the recommendation model of the present application can more reasonably give the recommendation result. In the recommendation ranking stage of the present application, the reciprocal of the average cross-entropy loss of different examination items in the disease prediction stage is added as the confidence level of disease diagnosis and participates in the recommendation ranking process as a weight, which can improve the accuracy of the recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a schematic flowchart of a method for recommending examination items based on a Bayesian network provided by an embodiment of the present application;

[0029] Figure 2 is a schematic diagram of the principle of a method for recommending examination items based on a Bayesian network provided by an embodiment of the present application;

[0030] Figure 3 is a schematic diagram of the structure of an auxiliary decision-making system provided by an embodiment of the present application;

[0031] Figure 4 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application.

[0032] Reference numerals: 31, acquisition unit; 32, first model application unit; 33, second model application unit; 34, ranking and recommendation unit; 35, path planning unit; 400, electronic device; 401, processor; 402, communication bus; 403, user interface; 404, network interface; 405, memory. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments.

[0034] In the description of the embodiments of the present application, words such as "exemplary", "for example" or "for illustration" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary", "for example" or "for illustration" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of words such as "exemplary", "for example" or "for illustration" is intended to present relevant concepts in a specific manner.

[0035] In the description of the embodiments of the present application, the term "and / or" merely describes the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, B exists alone, and A and B exist simultaneously. Additionally, unless otherwise specified, the meaning of the term "plural" refers to two or more. For example, a plurality of systems means two or more systems, and a plurality of screen terminals means two or more screen terminals. Furthermore, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. The terms "include", "comprise", "have" and their variants all mean "including but not limited to", unless otherwise particularly emphasized in other ways.

[0036] The application scenarios of the method in the embodiments of the present application mainly include medical examination item recommendation, etc. Different diseases in the medical field have heterogeneity, so the screening schemes corresponding to disease diagnosis are different. For diseases with early symptoms not obvious, not easily detectable, high incidence rate and poor prognosis, early and effective diagnosis can intervene in the treatment of patients in advance, slow down the development of the disease, save medical resources, reduce the economic pressure of patients, and improve the quality of life of patients.

[0037] The embodiments of the present application provide a method for recommending examination items and an auxiliary decision-making system based on a Bayesian network, which are used to solve the problem that in the related technologies, the disease screening methods fail to fully utilize the known patient diagnosis data to recommend detection items, resulting in inaccurate recommendation results.

[0038] As Figure 1 shown, it is a schematic flowchart of a method for recommending examination items based on a Bayesian network provided by the embodiments of the present application, and the method includes steps S101 - S106.

[0039] S101, obtain the real-time examination item data of the target patient. The real-time examination item data includes one or more examined item data, and the examined item data includes one or more types of multimodal data.

[0040] In the embodiments of the present application, obtaining the real-time examination item data of the target patient specifically includes: collecting various examined item data of the target patient, including demographic data, laboratory data, chest radiograph data, electrocardiogram data, ultrasound data, and other examination data. Each examination item contains one or more of multimodal data (structured data, text data, and image data). Determine the disease label and patient characteristics of the target patient according to the clinical criteria corresponding to the disease, and construct a basic data set according to the patient characteristics and disease label of the target patient.

[0041] Preprocess the obtained basic data set of the target patient.

[0042] Specifically, there are characteristics of multiple examined items in the basic data set, which are respectively represented as , where n is the serial number of the examined item; each examined item contains one or more of the three modal characteristics: structured characteristics , medical text characteristics and medical image characteristics , among which the null value processing method for structured characteristics is filling with model fitting results and filling with the mode, and the null value processing method for medical text characteristics is removing text noise (including garbled characters and irregular symbols), and the preprocessing method for medical image characteristics is standardization. The preprocessed standard data set can be expressed as: or or or or or or ;

[0043] Finally, output the standard data set after null value filling, text noise removal, and image standardization. This standard data set serves as the real-time examination item data of the target patient.

[0044] In the above embodiments of the present application, only an exemplary preprocessing method for the basic data set of a target patient is given. In actual applications, other types of data processing methods can also be adopted. This embodiment is only for exemplary illustration.

[0045] S102. Input the examined item data into the disease occurrence probability prediction model corresponding to the examined item to obtain the disease occurrence probability grading and confidence level corresponding to the examined item.

[0046] In the embodiments of the present application, each examined item corresponds to a disease occurrence probability prediction model. The examined item data includes one or more types of multi-modal data. The disease occurrence probability prediction model corresponding to the examined item includes one or more encoders. The encoders correspond to the multi-modal data one by one. One encoder is configured for one type of multi-modal data, and each encoder is used to process one type of multi-modal data.

[0047] The specific method of using the disease occurrence probability prediction model to obtain the disease occurrence probability grading and confidence level corresponding to the examined item will be described in detail in the subsequent embodiments, so no more details will be given here.

[0048] S103. For the item to be examined, set the disease occurrence probability grading corresponding to the item to be examined and the confidence level corresponding to the item to be examined.

[0049] For the item to be examined, set the disease occurrence probability grading of the item to be examined to a preset value. For example, set the disease occurrence probability grading of the item to be examined to -1. Each item to be examined corresponds to a disease occurrence probability prediction model. Normalize the reciprocal of the average cross-entropy loss of the disease occurrence probability prediction model corresponding to the item to be examined, and set the confidence level corresponding to the item to be examined.

[0050] The specific method for obtaining the disease occurrence probability grading and confidence level corresponding to the item to be examined will be described in detail in the subsequent embodiments, so it will not be elaborated here.

[0051] S104. Input the disease occurrence probability grading corresponding to the examined item and the disease occurrence probability grading corresponding to the item to be examined into a preset Bayesian network model to obtain the first conditional probability distribution of the examined item on the target disease, the second conditional probability distribution of the examined item on the item to be examined, and the third conditional probability distribution of the examined item on the item to be examined based on the target disease. The preset Bayesian network model is trained according to the disease occurrence probability grading and disease labels corresponding to various historical examination item data.

[0052] In the embodiment of the present application, the examined items may further include real-time data collected during the patient's hospitalization, including demographic data, laboratory data, chest X-ray data, electrocardiogram data, ultrasound data, and other examination data. Each examination item contains multi-modal data (structured data, text data, and image data). The processing of this part of the data can refer to the processing process of the above basic data set, and the embodiment of the present application will not elaborate here.

[0053] In the embodiment of the present application, since each patient has different examined items, the values of the multi-modal data in the examined items are also different, and the disease occurrence probability grading corresponding to each examined item is also different. Therefore, the disease occurrence probability grading contains the patient's personalized information. Input the disease occurrence probability grading corresponding to the examined item and the disease occurrence probability grading corresponding to the item to be examined into the preset Bayesian network, and the output result is the first conditional probability distribution of each examined item on the target disease, the second conditional probability distribution of the examined item on the item to be examined, and the third conditional probability distribution of the examined item on the item to be examined based on the target disease. The output result contains the patient's personalized information, realizing the recommendation of subsequent examination items by introducing the patient's personalized information.

[0054] The examined items for each patient are different, and the values of multi-modal data also vary, which reflects the unique health status and disease risks of the patient. Inputting these personalized disease occurrence probability grades into a preset Bayesian network can more accurately reflect the patient's current disease state and risks, thus more precisely matching the patient's examination needs. The Bayesian network can calculate the conditional probability distribution of each examined item for the target disease based on historical data and the results of the examined items. This probability-based reasoning method can comprehensively consider various factors of the patient's examined items, such as medical history, symptoms, signs, etc., so as to more comprehensively evaluate the patient's disease risks. By introducing the patient's personalized information, the Bayesian network can learn the relationship between each examination item and the target disease, and thus determine the examination items to be examined that are highly relevant to the target disease from the known patient information. This means that the recommended subsequent examination items are more likely to reveal the patient's potential diseases or confirm the existing diagnoses, thereby improving the accuracy of the recommendation. Since the patient's disease state may change over time, the recommendation method containing personalized information can dynamically adjust the recommendation strategy. For example, as new examination items are carried out and results are obtained, the system can re-evaluate the patient's disease risks and adjust the subsequent examination item recommendations accordingly. The personalized examination item recommendation can avoid unnecessary examinations and reduce the patient's economic burden and time cost.

[0055] In the embodiment of the present application, the disease occurrence probability grades corresponding to multiple historical examination item data and disease labels are used as inputs to train an initial Bayesian network model. The historical examination item data is the examination item data of historical patients, including demographic data, laboratory data, chest X-ray data, electrocardiogram data, ultrasound data, and other examination data. Each examination item contains multi-modal data (structured data, text data, and image data). The disease label is determined according to the clinical criteria corresponding to the disease. The disease label , where 0 represents not having the disease and 1 represents having the disease; the initial Bayesian network model is a directed acyclic graph, which contains nodes and edges in the graph. The historical examination item corresponds to a node in the Bayesian network; in the embodiment of the present application, the target disease is also a node in the Bayesian network. The relationship between the historical examination item and the target disease corresponds to the edge in the initial Bayesian network. The disease occurrence probability grades corresponding to multiple historical examination item data and disease labels are input into the initial Bayesian network model to train the initial Bayesian network model.

[0056] The training process is as follows: Initial parameter values are assigned to each node and edge, and these parameters usually represent conditional probabilities or prior probabilities. The disease occurrence probability grading corresponding to the historical examination item data (including multimodal data) and the corresponding disease labels are used as inputs and input into the initial Bayesian network model. According to the definition and formula of the Bayesian network, the conditional probability distribution of each node (i.e., historical examination items and target diseases) is calculated. This usually involves complex probability calculations and inference processes. According to the calculated conditional probability distribution and the true disease labels, appropriate algorithms (such as maximum likelihood estimation, Bayesian estimation, etc.) are used to update the parameter values of the model. Repeat the above steps and continuously iterate to optimize the parameter values of the model until the performance of the model on the validation set reaches the preset standard or the maximum number of iterations is reached. An independent test set is used to evaluate the performance of the trained preset Bayesian network model. Commonly used evaluation metrics include accuracy, recall, F1-score, etc. If the model performance does not meet the requirements, the model structure can be adjusted, the algorithm can be optimized, or more training data can be added to improve the model. Finally, the trained preset Bayesian network model is output. According to the definition and formula of the Bayesian network, the method and formula for calculating the conditional probability distribution of each node (i.e., historical examination items and target diseases) are prior arts.

[0057] S105, calculate the fourth conditional probability distribution of the item to be examined on the target disease based on the examined items according to the first conditional probability distribution, the second conditional probability distribution, and the third conditional probability distribution.

[0058] Specifically, the first conditional probability distribution of the examined items on the target disease, the second conditional probability distribution of the examined items on the item to be examined, and the third conditional probability distribution of the examined items on the item to be examined based on the target disease are input into the preset Bayesian network model to obtain the fourth conditional probability distribution of the item to be examined on the target disease based on the examined items. In a Bayesian network, calculating the fourth conditional probability distribution \(P(D|X_k)\) of the item to be examined (denoted as \((X_k)\)) on the target disease (denoted as \((D)\)) usually involves multiple nodes and edges in the network, as well as the conditional probability relationships between these nodes. Inference algorithms of the Bayesian network (such as variable elimination method, clique tree propagation method, etc.) can be used to approximately calculate this conditional probability. These algorithms usually rely on the conditional probability tables (CPTs) learned in the network, and these tables store the conditional probabilities of child nodes given the states of parent nodes. Conceptually, the process of calculating \(P(D|X_k)\) can be described as:

[0059] Determine the parent nodes of \((X_k)\): In the Bayesian network, find all the nodes that directly affect \((X_k)\) (i.e., the parent nodes of \((X_k)\)), and \((Pa(X_k))\) represents the set of all nodes that directly affect \((X_k)\) (i.e., the parent nodes of \((X_k)\));

[0060] Collect relevant conditional probabilities: Locate the conditional probability tables (CPTs) associated with (X_k) and its parent nodes, which will be used to calculate (P(X_k|Pa(X_k))) and other relevant conditional probabilities;

[0061] Apply Bayesian inference: Use the inference algorithm of the Bayesian network, combined with the disease occurrence probability grading of (X_k) (i.e., the current state of (X_k)) and the relevant conditional probability tables, to calculate (P(D|X_k)). This typically involves traversing the relevant nodes and edges in the network and applying the multiplication formula and chain rule of conditional probabilities. The inference algorithm of the Bayesian network, conditional probability tables, penalty formula, and chain rule of conditional probabilities are all prior arts.

[0062] S106. Select the first priority recommended inspection item from the items to be inspected according to the fourth conditional probability distribution and the confidence level corresponding to the item to be inspected.

[0063] In the embodiment of the present application, according to the fourth conditional probability distribution of the item to be inspected for the target disease, calculate the KL divergence value of the fourth conditional probability distribution for the target disease on the premise that the item to be inspected of the target patient is assumed to have been performed. Then, based on the confidence level corresponding to the item to be inspected and the KL divergence value of the fourth conditional probability distribution of the item to be inspected for the target disease, select the first priority recommended inspection item from the items to be inspected. The specific steps are as follows:

[0064] In a possible implementation manner, in step S106, selecting the first priority recommended inspection item from the items to be inspected according to the fourth conditional probability distribution of the item to be inspected for the target disease and the confidence level corresponding to the item to be inspected specifically includes steps S1061 - S1063.

[0065] S1061. Obtain the KL divergence of the fourth conditional probability distribution of the item to be inspected for the target disease.

[0066] S1062. Perform weighted calculation on the KL divergence according to the confidence level corresponding to the item to be inspected, obtain the calculation results of multiple items to be inspected, sort the calculation results from high to low, and obtain the first priority recommended items.

[0067] S1063. Determine the first priority recommended items as the items to be preferentially recommended for inspection.

[0068] For steps S1061 - S1063, specifically, obtain the fourth conditional probability distribution of the items to be examined for the target disease, then calculate the KL divergence value of the fourth conditional probability distribution of the items to be examined for the target disease, and multiply it by the corresponding confidence level for weighting. There can be multiple items to be examined. Therefore, the calculation results of multiple items to be examined can be obtained, sorted from largest to smallest, and the item with the largest weighted KL divergence is taken as the final recommended item for examination and output. In this embodiment, a quantifiable index is defined through the KL divergence to represent the addition of the disease screening effect, and the Bayesian network model is combined to solve the problem of personalized examination item path recommendation in disease screening. It solves the problem that the existing research scheme of directly using multimodal data to construct a diagnostic model can only recommend single-point examination items and cannot achieve personalized examination path recommendation.

[0069] In the embodiments of the present application, the calculation method of the KL divergence value can refer to the related technology, and the embodiments of the present application will not elaborate too much here. Specific application embodiments of steps S1061 - S1063 can be referred to the subsequent embodiments.

[0070] In a possible implementation manner, the multimodal data includes one or more of structured data, text data, and image data.

[0071] Among them, the structured data is data information such as test results or measurement indicators, such as demographic data, test data, etc.; the medical text is text information such as the diagnosis results issued by doctors; the medical image is image information such as ultrasound images, CT images, etc.

[0072] In a possible implementation manner, in step S102, input the data of the examined items into the disease occurrence probability prediction model corresponding to the examined items to obtain the disease occurrence probability grading and confidence level corresponding to the examined items, specifically including steps S1021 - S1024.

[0073] S1021, according to the data types of the multimodal data included in the data of the examined items, configure multiple encoders for the multimodal data, and the multiple encoders are used to process the multimodal data in the examination item data.

[0074] S1022, obtain the initial disease occurrence probability according to the output results of the multiple encoders.

[0075] S1023, discretize the initial disease occurrence probability according to multiple preset disease occurrence probability thresholds to obtain the disease occurrence probability grading corresponding to the examined items;

[0076] S1024, normalize the reciprocal of the average cross-entropy loss of the disease occurrence probability prediction model corresponding to the examined items to obtain the confidence level corresponding to the examined items.

[0077] The embodiments of the present application provide a method for obtaining the disease occurrence probability grading and confidence corresponding to the inspected items, which is specifically as follows:

[0078] As Figure 2 shown, in the embodiments of the present application, steps S1021 - S1024 specifically include: obtaining the final prediction result through a classifier according to the output results of multiple encoders , is the initial disease occurrence probability corresponding to the th inspected item. Then, based on a predetermined threshold, is discretized into risk probability grades 0 / 1 / 2, and the disease occurrence probability grade corresponding to each inspected item is output. The reciprocal of the average cross-entropy loss of the model corresponding to each inspected item is normalized, and the confidence of each inspected item on this disease is output .

[0079] In a possible implementation manner, in step S1021, according to the multi-modal data types of the inspected item data, multiple encoders are configured for the inspected item data, which specifically includes S1021A - S1021D.

[0080] S1021A, perform feature extraction on the inspected item data to obtain various feature data of the inspected item data; where

[0081] S1021B, when the feature data is structured feature data, use the first encoder to process the structured feature data, and the first encoder encodes according to a multi-layer perceptron;

[0082] S1021C, when the feature data is text feature data, use the second encoder to process the text feature data, and the second encoder encodes according to a multi-view convolutional neural network or a CNN neural network;

[0083] S1021D, when the feature data is image feature data, use the third encoder to process the image feature data, and the third encoder encodes according to a residual neural network.

[0084] In the embodiments of the present application, based on the preprocessed standard data set, each inspected item corresponds to a model. Each model inputs the multi-modal data of a certain inspected item, where each modality has a specific encoder. For the th inspected item, map the corresponding multi-modal data to a high-dimensional feature space. Among them, the structured features are encoded using a multi-layer perceptron (MLP); considering the modular expression characteristics of medical texts, the embodiments of the present application select a CNN-based model to encode medical texts, and specifically select a multi-view convolutional neural network (MultiResCNN); medical images are encoded using a residual neural network (ResNet).

[0085] In a possible implementation, the method further includes step S107.

[0086] S107, obtaining the confidence level of the item to be inspected, specifically including steps S1071 - S1073.

[0087] S1071, allocating a corresponding disease occurrence probability prediction model to any item to be inspected, obtaining multiple disease occurrence probability prediction models.

[0088] Set a preset value as the disease occurrence probability grading corresponding to any of the items to be inspected; S1072, obtaining the average cross - entropy loss of any disease occurrence probability prediction model.

[0089] S1073, performing normalization processing on the reciprocal of any average cross - entropy loss to obtain the confidence level of any item to be inspected.

[0090] In the embodiments of the present application, for the specific obtaining methods of the disease occurrence probability grading and confidence level corresponding to the items to be inspected, reference can be made to the content of the disease occurrence probability grading and confidence level corresponding to the items that have been detected above. In the first example of the present application, for patient A, patient data is collected, and the input features include:

[0091] Demographic examination items: patient age 50, gender male, weight 70 kg, chief complaint text, current medical history text.

[0092] ECG examination item: not done.

[0093] Echocardiogram examination item: not done.

[0094] Use the pre - processing module to process the patient feature data and output the pre - processed demographic examination item data.

[0095] Input the processed demographic examination item data above into the trained model corresponding to the demographic examination item in the multi - modal disease occurrence probability prediction model module, output the initial disease occurrence probability corresponding to the demographic examination item, and discretize it into disease occurrence probability grading, output the disease occurrence probability grading as 1; for the items to be inspected, set the disease occurrence probability grading and confidence level corresponding to the items to be inspected, where the disease occurrence probability grading of the undone ECG examination item is given a default value of - 1; the disease occurrence probability grading of the undone echocardiogram examination item is given a default value of - 1. And output the confidence level of each previously deployed examination item in pulmonary hypertension disease (abbreviation: PH) (among the items that have been inspected, the confidence level of the demographic examination item , among the items to be inspected, set the confidence level of the ECG examination item , the confidence level of the echocardiogram examination item ).

[0096] Taking the disease occurrence probability grading of each inspected item and the disease occurrence probability grading corresponding to the item to be inspected as inputs, and inputting them into the pre-trained preset Bayesian network model in the second stage, outputting the first conditional probability distribution of each inspected item on the target disease PH, the second conditional probability distribution of each inspected item on the item to be inspected, and the third conditional probability distribution of the inspected item on the item to be inspected based on the target disease; Representing the demographic inspection item as , representing the electrocardiogram inspection item as , and representing the cardiac ultrasound inspection item as u.

[0097] Suppose the first conditional probability distribution of the demographic inspection item on PH is as follows: .

[0098] Among them, : That is, under the condition that the demographic inspection item has been done, the probability of PH occurrence is 0.4; That is, under the condition that the demographic inspection item has been done, the probability of PH not occurring is 0.6; That is, under the condition that the demographic inspection item has not been done, the probability of PH occurrence is 0.3; . That is, under the condition that the demographic inspection item has not been done, the probability of PH not occurring is 0.7.

[0099] Suppose the second conditional probability distribution of the demographic inspection item on the electrocardiogram inspection item is as follows:

[0100]

[0101] The second conditional probability distribution of the demographic inspection item on the cardiac ultrasound inspection item is as follows:

[0102]

[0103] The third conditional probability distribution of the demographic inspection item on the electrocardiogram inspection item based on PH is as follows:

[0104]

[0105] ,

[0106] 。

[0107] The third conditional probability distribution of the demographic inspection item on the cardiac ultrasound inspection item based on PH is as follows:

[0108]

[0109] ,

[0110] 。

[0111] (1) Given the first conditional probability distribution of demographic examination items on PH, the second conditional probability distribution of demographic examination items on other items to be examined, and the third conditional probability distribution of demographic examination on electrocardiogram examination items based on the target disease PH, the calculation of the above conditional probability distributions is based on the well-known Bayes' formula:

[0112]

[0113] Bayes' formula has been extended to multiple variables. Calculate the fourth conditional probability distribution of electrocardiogram examination items corresponding to PH based on demographic examination items, and further calculate the KL divergence of the fourth conditional probability distribution of the patient on PH , and multiply it by the confidence level , and the result is 0.04.

[0114] (2) Given the first conditional probability distribution of demographic examination items on PH, the second conditional probability distribution of demographic examination items on other items to be examined, and the third conditional probability distribution of demographic examination on electrocardiogram examination items based on the target disease PH, calculate the disease occurrence probability grading corresponding to the echocardiogram examination items based on demographic examination items, that is, the fourth conditional probability distribution, and further calculate the KL divergence of the fourth conditional probability distribution of the patient on PH , and multiply it by the confidence level , and the result is 0.06.

[0115] Sort the above two values from largest to smallest and select the examination item corresponding to the maximum weighted KL divergence. Therefore, the next recommended examination item is echocardiogram examination.

[0116] Finally, the next echocardiogram examination item is recommended for patient A and is prompted to the clinician. The clinician makes a final diagnosis and treatment decision for the patient with reference to the output of the system. Subsequently, update the echocardiogram examination item information of patient A, regard the echocardiogram examination item as an examined item, and continue to recommend the next examination item according to the logic of the above example.

[0117] In a possible implementation manner, the method further includes: obtaining new examination item data corresponding to the first priority recommended examination item; according to the new examination item data, updating the first conditional probability distribution of the examined items for the target disease, the second conditional probability distribution of the examined items for the to-be-examined items, and the third conditional probability distribution of the examined items for the to-be-examined items based on the target disease, and calculating the fourth conditional probability distribution of the remaining to-be-examined items for the target disease based on the examined items according to the updated first conditional probability distribution, second conditional probability distribution, and third conditional probability distribution as the updated fourth conditional probability distribution, where the first priority recommended examination item is not included in the remaining to-be-examined items; selecting a second priority recommended examination item again from the remaining to-be-examined items according to the updated fourth conditional probability distribution and the confidence levels corresponding to the remaining to-be-examined items; obtaining the priority recommended examination items multiple times to obtain the recommended examination item path for the target disease.

[0118] Exemplarily, referring to the two-round recommendation results of the above patient A, the following gives a specific example of the recommended examination items for the screening of pulmonary hypertension disease (abbreviated as PH). The examination items that the patient can do include demographic examination, laboratory examination, electrocardiogram examination, chest X-ray examination, and echocardiogram examination. The modalities include three types of modality data: structured, text, and image. Among them, the structured features include demographic data and laboratory information, the text includes the chief complaint, current medical history, chest X-ray conclusion, and echocardiogram conclusion, and the image features include electrocardiogram signal maps, chest X-ray maps, and echocardiogram maps.

[0119] Assume that the initial information of patient A is as follows, including: the items that have been done are the demographic examination item and the laboratory examination item, and the items that have not been done are the electrocardiogram examination item, the chest X-ray examination item, and the echocardiogram examination item.

[0120] Based on the following application example, in the first round (the above example), it is recommended that patient A do the echocardiogram examination item. Based on the recommendation of the first round, patient A finishes the echocardiogram examination item and updates the examination item information. Based on the updated patient data, in the second round, it is recommended that patient A do the chest X-ray examination item. If further examinations are still needed to confirm whether it is PH disease, the remaining electrocardiogram examination item is used as the final recommended item.

[0121] Therefore, the examination path recommended for patient A based on BN is: echocardiogram examination item -> chest X-ray examination item -> electrocardiogram examination item.

[0122] By adopting the above method embodiments, the present application can achieve the following beneficial effects:

[0123] 1. This application constructs a unified two-stage model framework through a deep neural network and a Bayesian network, which can utilize multi-modal data in scenarios with missing modalities to achieve personalized inspection path recommendations.

[0124] 2. This application defines a quantifiable metric using KL divergence to characterize the enhancement of disease screening effects, and combines a Bayesian network model to solve the problem of personalized inspection item path recommendation in disease screening. It solves the problem that the existing research solution of directly using multi-modal data to construct a diagnostic model can only perform single-point inspection item recommendations and cannot achieve personalized inspection path recommendations.

[0125] 3. In the process of disease diagnosis, the contribution degrees of different inspection items to the decision-making information provided for clinical judgment are different. Therefore, reflecting the differences in the contribution degrees of different inspection items in the recommendation model can more reasonably give recommendation results. This application adds the reciprocal of the average cross-entropy loss of different inspection items in the disease prediction stage as the confidence level of disease diagnosis in the recommendation ranking stage, and participates in the recommendation ranking process as a weight, which can improve the accuracy of the recommendation.

[0126] As Figure 3 shown, an embodiment of this application provides an auxiliary decision-making system, which includes a collection unit 31, a first model application unit 32, a second model application unit 33, and a ranking and recommendation unit 34.

[0127] The collection unit 31 is used to obtain real-time inspection item data of a target patient. The real-time inspection item data includes one or more inspected item data, and the inspected item data includes one or more multi-modal data.

[0128] The first model application unit 32 is used to input the inspected item data into the disease occurrence probability prediction model corresponding to the inspected item to obtain the disease occurrence probability and confidence level corresponding to the inspected item; for the to-be-inspected item, set the disease occurrence probability level corresponding to the to-be-inspected item and the confidence level of the to-be-inspected item.

[0129] The second model application unit 33 is used to input the disease occurrence probability levels corresponding to the inspected items and the disease occurrence probability levels corresponding to the to-be-inspected items into a preset Bayesian network model to obtain a first conditional probability distribution of the inspected items on the target disease, a second conditional probability distribution of the inspected items on the to-be-inspected items, and a third conditional probability distribution of the inspected items on the to-be-inspected items based on the target disease. The preset Bayesian network model is trained according to the disease occurrence probability levels and disease labels corresponding to various historical inspection item data.

[0130] A sorting and recommendation unit 34, configured to calculate, according to a first conditional probability distribution, a second conditional probability distribution, and a third conditional probability distribution, a fourth conditional probability distribution of items to be examined on a target disease based on the examined items; and further configured to select a first preferentially recommended examination item from the items to be examined according to the fourth conditional probability distribution and the confidence level corresponding to the items to be examined.

[0131] A path planning unit 35, configured to obtain new examination item data corresponding to the first preferentially recommended examination item; update, according to the new examination item data, the first conditional probability distribution of the examined items on the target disease, the second conditional probability distribution of the examined items on the items to be examined, and the third conditional probability distribution of the examined items on the items to be examined based on the target disease, calculate a fourth conditional probability distribution of the remaining items to be examined on the target disease based on the updated first conditional probability distribution, second conditional probability distribution, and third conditional probability distribution as the updated fourth conditional probability distribution, where the remaining items to be examined do not include the first preferentially recommended examination item; select a second preferentially recommended examination item from the remaining items to be examined according to the updated fourth conditional probability distribution and the confidence level corresponding to the remaining items to be examined; obtain preferentially recommended examination items multiple times to obtain a recommended examination item path for the target disease.

[0132] In a possible implementation manner, the sorting and recommendation unit 34 is specifically configured to obtain the KL divergence of the fourth conditional probability distribution of the items to be examined on the target disease; perform weighted calculation on the KL divergence according to the confidence level corresponding to the items to be examined to obtain calculation results of multiple items to be examined, sort the calculation results from high to low to obtain a first preferentially recommended item; and determine the first preferentially recommended item as the item to be preferentially recommended for examination.

[0133] In a possible implementation manner, the first model application unit 32 is specifically configured to configure multiple encoders for multimodal data according to the data types of the multimodal data included in the examined item data, configure one encoder for one type of multimodal data, and the multiple encoders are used to process the multimodal data in the examination item data; obtain an initial disease occurrence probability according to the output results of the multiple encoders; perform discretization processing on the initial disease occurrence probability according to multiple preset disease occurrence probability thresholds to obtain a disease occurrence probability classification corresponding to the examined item data; and perform normalization processing on the reciprocal of the average cross-entropy loss of the disease occurrence probability prediction model corresponding to the examined item to obtain the confidence level corresponding to the examined item.

[0134] In a possible implementation manner, the first model application unit 32 is specifically configured to perform feature extraction on the inspected item data to obtain multiple types of feature data of the inspected item data; among them, when the feature data is structured feature data, a first encoder is used to process the structured feature data, and the first encoder is encoded according to a multi-layer perceptron; when the feature data is text feature data, a second encoder is used to process the text feature data, and the second encoder is encoded according to a multi-view convolutional neural network or a CNN neural network; when the feature data is image feature data, a third encoder is used to process the image feature data, and the third encoder is encoded according to a residual neural network.

[0135] In a possible implementation manner, the auxiliary decision-making system further includes a confidence level acquisition unit.

[0136] The confidence level acquisition unit is configured to assign a corresponding disease occurrence probability prediction model to any item to be inspected, obtaining multiple disease occurrence probability prediction models; obtaining the average cross-entropy loss of any disease occurrence probability prediction model; performing normalization processing on the reciprocal of any average cross-entropy loss to obtain the confidence level of the item to be inspected.

[0137] It should be noted that: when the device provided in the above embodiment realizes its functions, only the division of the above-mentioned functional modules is used for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be seen in the method embodiment, which will not be repeated here.

[0138] Please refer to Figure 4 , which is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 4 shown, the electronic device 400 may include: at least one processor 401, at least one network interface 404, a user interface 403, a memory 405, and at least one communication bus 402.

[0139] Among them, the communication bus 402 is used to realize the connection and communication between these components.

[0140] Among them, the user interface 403 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 403 may further include a standard wired interface and a wireless interface.

[0141] Among them, the network interface 404 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0142] Among them, the processor 401 may include one or more processing cores. The processor 401 connects various parts within the entire server through various interfaces and lines, and executes various functions of the server and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 405, and by calling the data stored in the memory 405. Optionally, the processor 401 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 401 may integrate a combination of one or more of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communication. It can be understood that the above-mentioned modem may not be integrated into the processor 401 and may be implemented separately by a single chip.

[0143] Among them, the memory 405 may include random access memory (RAM) and may also include read-only memory. Optionally, the memory 405 includes a non-transitory computer-readable storage medium. The memory 405 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 405 may include a program storage area and a data storage area. Among them, the program storage area may store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store the data involved in the above-mentioned various method embodiments. Optionally, the memory 405 may also be at least one storage device located far from the aforementioned processor 401. As Figure 4 shown, the memory 405, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for inspection item recommendation.

[0144] In Figure 4In the electronic device 400 shown, the user interface 403 is mainly used to provide an interface for the user to input and obtain the data input by the user; and the processor 401 can be used to call the application program stored in the memory 405 with inspection item recommendations. When executed by one or more processors, the electronic device 400 is caused to execute one or more of the methods as described in the above embodiments.

[0145] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0146] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0147] In several embodiments provided by this application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection to each other can be through some service interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical or other form.

[0148] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0149] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0150] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned memory includes various media that can store program codes, such as USB flash drives, mobile hard disks, magnetic disks, or optical discs.

[0151] The above are only exemplary embodiments of the present disclosure, and the scope of the present disclosure cannot be limited thereby. That is, all equivalent changes and modifications made in accordance with the teachings of the present disclosure still fall within the scope covered by the present disclosure. After considering the specification and the disclosure of the practical truth, those skilled in the art will easily think of other implementation schemes of the present disclosure. This application aims to cover any variations, uses, or adaptive changes of the present disclosure, and these variations, uses, or adaptive changes follow the general principles of the present disclosure and include common general knowledge or conventional technical means in the technical field not recorded in the present disclosure.

Claims

1. A method for recommending inspection items based on a Bayesian network, characterized in that, The method includes: Obtaining real-time examination item data of a target patient, where the real-time examination item data includes one or more examined item data, and the examined item data includes one or more types of multimodal data; Inputting the examined item data into a disease occurrence probability prediction model corresponding to the examined item to obtain the disease occurrence probability grading and confidence level corresponding to the examined item; For an item to be examined, setting the disease occurrence probability grading corresponding to the item to be examined and the confidence level corresponding to the item to be examined; Inputting the disease occurrence probability grading corresponding to the examined item and the disease occurrence probability grading corresponding to the item to be examined into a preset Bayesian network model to obtain a first conditional probability distribution of the examined item on the target disease, a second conditional probability distribution of the examined item on the item to be examined, and a third conditional probability distribution of the examined item on the item to be examined based on the target disease. The preset Bayesian network model is trained according to the disease occurrence probability grading and disease labels corresponding to multiple historical examination item data; Calculating a fourth conditional probability distribution of the item to be examined on the target disease based on the examined item according to the first conditional probability distribution, the second conditional probability distribution, and the third conditional probability distribution; Selecting a first priority recommended examination item from the items to be examined according to the fourth conditional probability distribution and the confidence level corresponding to the item to be examined, The selecting a first priority recommended examination item from the items to be examined according to the fourth conditional probability distribution and the confidence level corresponding to the item to be examined specifically includes: Obtaining the KL divergence of the fourth conditional probability distribution of the item to be examined on the target disease; Performing weighted calculation on the KL divergence according to the confidence level corresponding to the item to be examined to obtain calculation results of multiple items to be examined, sorting the calculation results from high to low to obtain a first priority recommended item; Determining the first priority recommended item as the item to be preferentially recommended for examination.

2. The method according to claim 1, characterized in that The multimodal data includes one or more of structured data, text data, and image data.

3. The method according to claim 1, wherein Inputting the examined item data into the disease occurrence probability prediction model corresponding to the examined item to obtain the disease occurrence probability grading and confidence level corresponding to the examined item specifically includes: Configuring multiple encoders for the multimodal data according to the data types of the multimodal data included in the examined item data, configuring one encoder for one type of multimodal data, and the multiple encoders are used to process the multimodal data in the examination item data; Obtaining an initial disease occurrence probability according to the output results of the multiple encoders; Performing discretization processing on the initial disease occurrence probability according to multiple preset disease occurrence probability thresholds to obtain the disease occurrence probability grading corresponding to the examined item; Normalizing the reciprocal of the average cross-entropy loss of the disease occurrence probability prediction model corresponding to the examined item to obtain the confidence level corresponding to the examined item.

4. The method according to claim 3, wherein The configuring multiple encoders for the examination item data according to the multimodal data types of the examined item data specifically includes: Characterize the inspected item data to obtain various feature data of the inspected item data; among them, When the feature data is structured feature data, use a first encoder to process the structured feature data, and the first encoder encodes according to a multi-layer perceptron; When the feature data is text feature data, use a second encoder to process the text feature data, and the second encoder encodes according to a multi-view convolutional neural network or a CNN neural network; When the feature data is image feature data, use a third encoder to process the image feature data, and the third encoder encodes according to a residual neural network.

5. The method according to claim 1, characterized in that, For the item to be inspected, setting the disease occurrence probability grading corresponding to the item to be inspected and the confidence level corresponding to the item to be inspected specifically includes: Assign a disease occurrence probability prediction model corresponding to it to any one of the items to be inspected to obtain a plurality of the disease occurrence probability prediction models; Set a preset value as the disease occurrence probability grading corresponding to any one of the items to be inspected; Obtain the average cross-entropy loss of any one of the disease occurrence probability prediction models; Normalize the reciprocal of the average cross-entropy loss of any one to obtain the confidence level corresponding to any one of the items to be inspected.

6. The method according to any one of claims 1-5, characterized in that, The method further includes: Obtain the new inspection item data corresponding to the first priority recommended inspection item; According to the new inspection item data, update the first conditional probability distribution of the inspected item on the target disease, the second conditional probability distribution of the inspected item on the item to be inspected, and the third conditional probability distribution of the inspected item on the item to be inspected based on the target disease. Calculate the fourth conditional probability distribution of the remaining items to be inspected on the target disease based on the updated first conditional probability distribution, second conditional probability distribution, and third conditional probability distribution as the updated fourth conditional probability distribution. The remaining items to be inspected do not include the first priority recommended inspection item; According to the updated fourth conditional probability distribution and the confidence levels corresponding to the remaining items to be inspected, select a second priority recommended inspection item from the remaining items to be inspected again; Obtain the priority recommended inspection items multiple times to obtain the recommended inspection item path of the target disease.

7. An auxiliary decision-making system, characterized in that The system includes a collection unit, a first model application unit, a second model application unit, and a sorting and recommendation unit; The collection unit is used to obtain the real-time inspection item data of the target patient. The real-time inspection item data includes one or more inspected item data, and the inspected item data includes one or more multi-modal data; The first model application unit is used to input the inspected item data into the disease occurrence probability prediction model corresponding to the inspected item to obtain the disease occurrence probability and confidence level corresponding to the inspected item; for the item to be inspected, set the disease occurrence probability grading corresponding to the item to be inspected and the confidence level corresponding to the item to be inspected; The second model application unit is configured to input the disease occurrence probability grades corresponding to the items to be examined and the disease occurrence probability grades corresponding to the items to be examined into a preset Bayesian network model, so as to obtain a first conditional probability distribution of the examined items on the target disease, a second conditional probability distribution of the examined items on the items to be examined, and a third conditional probability distribution of the examined items on the items to be examined based on the target disease. The preset Bayesian network model is trained according to the disease occurrence probability grades and disease labels corresponding to a variety of historical examination item data; The sorting and recommendation unit is configured to calculate a fourth conditional probability distribution of the items to be examined on the target disease based on the examined items according to the first conditional probability distribution, the second conditional probability distribution, and the third conditional probability distribution; and is further configured to select a first priority recommended examination item from the items to be examined according to the fourth conditional probability distribution and the confidence level corresponding to the items to be examined, The step of selecting a first priority recommended examination item from the items to be examined according to the fourth conditional probability distribution and the confidence level corresponding to the items to be examined specifically includes: Obtaining the KL divergence of the fourth conditional probability distribution of the items to be examined on the target disease; Performing weighted calculation on the KL divergence according to the confidence level corresponding to the items to be examined, obtaining calculation results of the multiple items to be examined, sorting the calculation results from high to low, and obtaining a first priority recommended item; Determining the first priority recommended item as the item to be preferentially recommended for examination.

8. The auxiliary decision-making system according to claim 7, characterized in that The system further includes a path planning unit; The path planning unit is configured to obtain new examination item data corresponding to the first priority recommended examination item; according to the new examination item data, update the first conditional probability distribution of the examined items on the target disease, the second conditional probability distribution of the examined items on the items to be examined, and the third conditional probability distribution of the examined items on the items to be examined based on the target disease, calculate a fourth conditional probability distribution of the remaining items to be examined on the target disease based on the examined items according to the updated first conditional probability distribution, second conditional probability distribution, and third conditional probability distribution, and use it as the updated fourth conditional probability distribution. The remaining items to be examined do not include the first priority recommended examination item; according to the updated fourth conditional probability distribution and the confidence level corresponding to the remaining items to be examined, select a second priority recommended examination item from the remaining items to be examined again; obtain the priority recommended examination items multiple times to obtain the recommended examination item path of the target disease.

9. An electronic device, characterized in that, It includes a processor, a user interface, a network interface, and a memory. The memory is used to store instructions. The user interface and the network interface are used to communicate with other devices. The processor is used to execute the instructions stored in the memory, so that the electronic device executes the method according to any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1-6 is executed.

Images