Gynecological clinical pathway prediction method based on neuron-level feature selection

Abstract

The invention provides an obstetrics and gynecology clinical path prediction method based on neuron-level feature selection, and belongs to the technical field of path prediction, and the method comprises the steps: dividing the obstetrics and gynecology electronic medical record data of a patient into time sequence features and baseline features, inputting the time sequence features into a trained GRU network, and carrying out the feature coding, obtaining dynamic change rules of various clinical indexes along with time, and converting the baseline features into feature vectors through an embedded layer; the feature importance is calculated based on the last gynecological and obstetrical doctor-seeing record of a patient and a doctor-seeing scene, and the dynamic change rule of various clinical indexes along with time is projected based on the feature importance, and the feature vector is mapped to a new representation space to obtain a projection matrix to generate a gating vector. And the features of the main task layer and the auxiliary task layer are screened and fused to obtain a final feature representation to obtain a multi-dimensional prediction result. And more personalized and accurate diagnosis and treatment suggestions are provided for doctors.

Description

Technical Field

[0001] The present invention relates to the technical field of path prediction, and particularly to an obstetrics and gynecology clinical path prediction method based on neuron-level feature selection. Background Art

[0002] With the development of medical informatization, obstetrics and gynecology hospitals have accumulated a large amount of electronic medical record data including patients' basic information, inspection and examination reports, medication and treatment records, etc. However, these data are often stored in an original and discrete form, lacking a suitable organizational form and analysis method, and it is difficult to provide effective support for doctors' clinical decisions. Although deep learning models have shown excellent performance in clinical prediction tasks, they face challenges such as limited data volume and complex models in practical applications, and are prone to overfitting phenomena, affecting the reliability of predictions.

[0003] In clinical practice, experienced doctors often refer to previous similar cases to assist in diagnosis and treatment decisions. However, this kind of experience requires long-term professional accumulation, and medical institutions generally face the problem of shortage of expert resources. At the same time, even if different patients have similar diseases, there will be significant differences in many aspects such as age and physical constitution, and doctors lack a quantitative method to evaluate the impact of such differences on treatment effects. In addition, the health status of patients is dynamically changing, and simply classifying patients into fixed categories is difficult to accurately reflect the development process of the disease, and there is currently a lack of effective methods to capture and predict such dynamic changes.

[0004] Therefore, the present invention proposes an obstetrics and gynecology clinical path prediction method based on neuron-level feature selection. Summary of the Invention

[0005] The present invention provides an obstetrics and gynecology clinical path prediction method based on neuron-level feature selection, which is used to predict future clinical visits by establishing a separate auxiliary network and extract health status representations from a long-term perspective. At the same time, through the neuron-level filtering gate mechanism, feature units helpful for target prediction are adaptively selected, so as to provide more personalized and accurate diagnosis and treatment suggestions for doctors. This method makes full use of the existing clinical data in the hospital without additional annotation work, and can provide more scientific and reliable decision-making support for doctors.

[0006] The present invention provides an obstetrics and gynecology clinical path prediction method based on neuron-level feature selection, including: Step 1: Divide the obstetrics and gynecology electronic medical record data of patients into time-series features and baseline features, where the time-series features are test data, vital signs, and medical treatment records, and the baseline features are demographic features and medical history information; Step 2: Input the temporal features into the trained GRU network for feature encoding to obtain the dynamic change rules of various clinical indicators over time. At the same time, convert the baseline features into feature vectors through the embedding layer; Step 3: Calculate the feature importance based on the patient's most recent obstetrics and gynecology visit record and the visit scenario. Project the dynamic change rules of various clinical indicators over time and the feature vectors onto a new representation space based on the feature importance to generate a gating vector, and respectively screen and fuse the features of the main task layer and the auxiliary task layer to obtain the final feature representation, where the feature importance calculation and feature screening constitute the neuron-level feature selection mechanism; Step 4: Obtain the multi-dimensional prediction results of the clinical pathway according to the final feature representation.

[0007] The present invention provides a method for predicting the obstetrics and gynecology clinical pathway based on neuron-level feature selection, which divides the obstetrics and gynecology electronic medical record data of patients into temporal features and baseline features, including: Collect the multi-modal data in the obstetrics and gynecology electronic medical record, where the multi-modal data includes: structured data and text data; Use natural language processing technology to perform semantic analysis on the text data, extract key semantics, and combine the meanings of clinical indicators in the structured data to establish the semantic association relationship between data items; Divide the multi-modal data into temporal features and baseline features based on the semantic association relationship.

[0008] The present invention provides a method for predicting the obstetrics and gynecology clinical pathway based on neuron-level feature selection. In the process of obtaining the multi-dimensional prediction results of the clinical pathway according to the final feature representation, it further includes: Extract the disease type of the patient, the treatment stage for each disease type, and the severity of the condition from the medical history information, construct the disease feature vector for each disease type, and perform clustering analysis on all disease feature vectors; Determine the influence relationship between each cluster center and the current feature vector of the patient under the most recent obstetrics and gynecology visit record based on the disease interaction relationship table, and reverse infer the enlightenment effect on each feature based on the influence relationship, where the enlightenment effect includes: positive enlightenment, negative enlightenment, and no enlightenment, where positive enlightenment is assigned a value of 1, no enlightenment is assigned a value of 0, and negative enlightenment is assigned a value of -1; Sort the values of all enlightenment effects under each feature in order from largest to smallest according to the relationship of the influence relationship, and at the same time, sort the values of all enlightenment effects under each feature in order from largest to smallest according to the relationship of the assigned enlightenment values; Perform linear fitting on the curves obtained from the first sorting and the curves obtained from the second sorting respectively to obtain the first fitting value and the second fitting value; Sort the first fitting values of all features from large to small to obtain the first serial number of each feature. At the same time, sort the second fitting values of all features from large to small to obtain the second serial number of each feature. According to the first serial number and the second serial number, obtain the first rough attention coefficient; Determine the absolute value of the difference between the first fitting value of the linear fitting result of the curve obtained from the first sorting and the original value of the corresponding point, determine the locked horizontal interval around the corresponding point, and obtain the original average value of the horizontal interval And the linear average value , determine the consistency ; Correct the first rough attention coefficient according to the consistency to obtain the second rough attention coefficient, and use it as the important attention degree of the corresponding feature in the process of clinical pathway prediction; The present invention provides a method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection, obtaining the second rough attention coefficient, including: Wherein, represents the second rough attention coefficient; represents the first rough attention coefficient; represents the adjustment parameter; represents the corresponding absolute value of the difference; b1 is a set threshold.

[0009] The present invention provides a method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection, and each part of the time series features corresponds to a trained GRU network; Each part of the baseline features corresponds to a learnable embedding layer.

[0010] The present invention provides a method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection, and obtaining the dynamic change law of various clinical indicators over time, including: Obtain the feature information of the set indicators of each part in the time series features under the corresponding trained GRU network, and extract the information structure under the same set indicator from all the feature information respectively to obtain the first dynamic change law vector corresponding to the set indicator, wherein the first dynamic change law vector includes the local change value of the ith set indicator at time t; The multi-scale feature extraction network uses convolutional kernels of different sizes to perform multi-scale convolutional operations on the input temporal features, extract feature information of different granularities, and capture the global dynamic change law vector of the temporal features. Among them, the global dynamic change law vector includes the global change value at time t. Based on the global dynamic change law vector, determine the temporal influence on each first dynamic change law vector: Among them, is the global dynamic change law vector The temporal influence on the i-th first dynamic change law vector ; and respectively represent parameter vectors; Based on the wisdom layer, merge the first dynamic change law vectors and the global dynamic change law vectors of all set indicators, and combine the temporal influences under different indicators to obtain the dynamic change laws of various indicators over time. Among them, the output of the GRU network and the output of the multi-scale feature extraction network are used as the input of the wisdom layer.

[0011] The present invention provides an obstetrics and gynecology clinical pathway prediction method based on neuron-level feature selection. Based on the dynamic change laws of various clinical indicators over time projected by feature importance and the feature vectors mapped to a new representation space, a projection matrix is generated to generate a gating vector, including: Construct a graph structure with the mapping elements mapped to the new representation space and the associated clinical indicator features, and deeply analyze the graph structure based on a graph neural network to obtain mapping logic; Convert the mapping logic into a structured standard representation and input it into a preset representation filling table to obtain a representation vector and a representation weight. Among them, the preset representation filling table is a multi-layer perceptron model, which takes the semantic and structural information of the standard representation as input and outputs the representation vector and the representation weight; Based on all the representation vectors and representation weights, construct a projection matrix , and use a regularization method to constrain the projection matrix. Among them, , Bn respectively represent the representation vectors corresponding to the first mapping element and the n-th mapping element; and respectively represent the representation weights corresponding to the first mapping element and the n-th mapping element; Use an asymmetric quantization method to quantize the projection matrix, map the continuous numerical values in the matrix to a finite discrete numerical value set, and combine a generative adversarial network to enhance the representation ability of the projection matrix; Multiply the enhanced projection matrix by the original eigenvector to obtain the projection result of each feature. Based on the numerical values and distribution of the projection results, the importance of the features is evaluated in combination with weights. Based on the feature importance scores, use the gating mechanism of the gated recurrent unit to generate gating vectors.

[0012] The present invention provides an obstetrics and gynecology clinical pathway prediction method based on neuron-level feature selection, which screens the features of the main task layer and the auxiliary task layer respectively, including: Screen the features of the main task layer according to the value division mechanism and using the gating vectors. Based on the auxiliary task layer, mine the potential supplementary information of each gating vector, and verify the features screened by the main task layer according to the potential supplementary information. Feature supplementation is performed on the screened features in the main task layer according to the verification association relationship.

[0013] Compared with the prior art, the beneficial effects of the present application are as follows: Through an innovative neuron-level feature selection mechanism and multi-channel feature processing method, in-depth mining and efficient utilization of obstetrics and gynecology electronic medical record data are realized, providing strong technical support for obstetric clinical decision-making.

[0014] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structures specifically pointed out in the written specification and the drawings.

[0015] The technical solution of the present invention will be further described in detail below through the drawings and embodiments. Description of the Drawings

[0016] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings: Figure 1 It is a flowchart of an obstetrics and gynecology clinical pathway prediction method based on neuron-level feature selection in an embodiment of the present invention; Figure 2 It is a main hierarchical structure diagram in an embodiment of the present invention; Figure 3 It is a feature extraction flowchart in an embodiment of the present invention; Figure 4 It is a feature selection flowchart in an embodiment of the present invention; Figure 5 It is a schematic diagram of an application scenario in an embodiment of the present invention. Detailed Embodiments

[0017] The preferred embodiments of the present invention will be described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not intended to limit the present invention.

[0018] The present invention provides an obstetrics and gynecology clinical pathway prediction method based on neuron-level feature selection, as Figure 1 shown, including: Step 1: Divide the obstetrics and gynecology electronic medical record data of the patient into time-series features and baseline features, where the time-series features are test data, vital signs, and medical records, and the baseline features are demographic features and medical history information; Step 2: Input the time-series features into the trained GRU network for feature encoding to obtain the dynamic change rules of various clinical indicators over time. At the same time, convert the baseline features into feature vectors through the embedding layer; Step 3: Calculate the feature importance based on the patient's most recent obstetrics and gynecology visit record and visit scenario. Project the dynamic change rules of various clinical indicators over time and the feature vectors onto a new representation space based on the feature importance to generate a gating vector, and respectively screen and fuse the features of the main task layer and the auxiliary task layer to obtain the final feature representation, where the feature importance calculation and feature screening constitute the neuron-level feature selection mechanism; Step 4: Obtain the multi-dimensional prediction results of the clinical pathway according to the final feature representation.

[0019] In this embodiment, for the time-series feature part, including test data (such as blood routine, biochemical test results, etc.), vital signs (such as blood pressure, heart rate monitoring data, etc.), and medical records (such as medication, surgery and other treatment information), they are respectively input into three trained GRU networks for feature encoding, so as to capture the dynamic change rules of various clinical indicators over time. For the baseline feature part, including demographic features (such as age, parity, etc.) and medical history information (such as past history, family history, etc.), they are converted into dense feature vectors through two learnable embedding layers.

[0020] The features after the above processing are input into the feature fusion module, which integrates features from different sources and different types to generate a unified feature representation. The key information is further refined through the feature selection step, and finally the fused features are output for subsequent risk prediction and benefit evaluation. This multi-channel feature processing scheme can make full use of various information in the patient's electronic medical record and provide comprehensive data support for clinical decision-making. For example, the system can analyze multiple dimensions such as the change trend of the patient's test results, treatment response characteristics, and basic conditions, so as to provide a reliable basis for doctors to formulate personalized treatment plans, as Figure 3 shown.

[0021] In the feature importance calculation stage, the system first calculates the feature importance based on the patient's most recent visit record, and then maps the features to a new representation space through importance projection. To improve the expressive ability of features, the system designs an innovative gating mechanism. By calculating the correlation of neuron activation values, a decorrelation loss function is constructed to prompt different neurons to learn complementary feature representations, as Figure 4 shown.

[0022] In the feature selection stage, a separated gating strategy is adopted to adaptively assign features to the main task and the auxiliary task. Specifically, the system generates a gating vector through a learnable projection matrix to screen the features of the main task layer and the auxiliary task layer respectively. For the main task layer, the system focuses on retaining the features directly related to target prediction; for the auxiliary task layer, the system pays attention to the features that can provide supplementary information. Finally, the system fuses the screened features to obtain the final feature representation. This feature selection mechanism can not only effectively identify and retain key information, but also reduce feature redundancy and improve the prediction performance of the model.

[0023] Through the above feature selection mechanism, the key features of different patients are adaptively identified, providing more reliable feature support for subsequent risk prediction and treatment plan formulation. The importance degree of features can be dynamically adjusted according to specific medical scenarios and patient conditions, so as to achieve personalized feature selection and prediction.

[0024] Temporal features are dynamic medical data that change over time. Among them, test data: blood routine (such as white blood cell count, hemoglobin), biochemical indicators (such as blood glucose, creatinine); vital signs: heart rate, blood pressure, body temperature, blood oxygen saturation; medical records: medication dosage, surgical operation records, nursing records.

[0025] Baseline features are relatively stable patient basic information. Among them, demographic features: age, gestational week, BMI, ethnicity; medical history information: chronic disease history (such as diabetes, hypertension), allergy history, previous pregnancy complications. It should be noted that structured data is extracted from the hospital HIS system, and unstructured information is extracted from electronic medical record texts through NLP.

[0026] Among them, feature encoding is to convert the original data into a low-dimensional semantic vector, and the embedding layer is a neural network layer that maps discrete features to a continuous vector space Feature importance is calculated as follows: , where is the feature vector of the j-th feature, , , , are trainable parameters.

[0027] 1) Input layer: responsible for receiving and preprocessing electronic medical record data 1. Temporal feature processing: processing temporal data such as various test and examination results 2. Baseline feature embedding: processing static information such as demographic features 2) Processing layer: implementing the core algorithm process 1. Multi-channel feature extraction: encoding different types of features 2. Main task network: completing the target prediction task 3. Auxiliary prediction network: predicting future clinical indicators 4. Neuron-level feature selection: screening important features 3) Output layer: generating multi-dimensional prediction results 1. Risk prediction: evaluating the risk level of patients 2. Path analysis: analyzing the disease development trend 3. Similar cases: matching similar patient cases In this embodiment, through the innovative neuron-level feature selection mechanism and multi-channel feature processing method, the in-depth mining and efficient utilization of obstetrics and gynecology electronic medical record data are realized, providing strong technical support for obstetric clinical decision-making.

[0028] Such as Figure 5 shown, the present invention can be applied to three core scenarios: pregnancy management, childbirth management, and postpartum follow-up. In the pregnancy management stage, the system starts risk assessment from the first antenatal examination, formulates personalized intervention plans and conducts dynamic monitoring. In the childbirth management stage, the system conducts childbirth risk assessment on the admitted parturients, provides suggestions on childbirth methods and implements hierarchical care. In the postpartum follow-up stage, the system predicts the risk of postpartum complications, formulates rehabilitation plans and conducts long-term health management. Through the state transfer mechanism, the system realizes the whole-process and continuous health management of pregnant and lying-in women.

[0029] First of all, the technical solution of the present invention shows significant advantages in key obstetric issues such as pregnancy complication prediction. By designing an auxiliary prediction network to predict the change trend of clinical indicators during pregnancy, richer feature representations can be obtained without additional labeled data. At the same time, the innovative neuron-level filtering gate mechanism can adaptively select features related to pregnancy outcome prediction, such as key indicators like blood pressure and coagulation function, effectively avoiding the impact of feature redundancy on prediction performance. Experimental results show that this solution has achieved excellent performance in tasks such as pregnancy complication warning and postpartum complication prediction.

[0030] Secondly, the technical solution of the present invention provides a scientific basis for the hierarchical management of key populations in obstetrics. Through multi-channel feature extraction, the system can comprehensively analyze various indicators of pregnant women, including pregnancy examination results, vital sign changes, and past medical history. This design enables the model to capture long-term dependencies in the clinical pathway during pregnancy, timely detect high-risk pregnant women, and provide decision-making support for obstetricians to develop personalized management plans.

[0031] From the perspective of clinical application, the technical solution of the present invention significantly improves the accuracy and interpretability of obstetric prediction. By explicitly modeling the importance of features, the system can help doctors understand the key risk factors leading to pregnancy complications. Especially for complications that endanger the safety of mothers and infants, such as preeclampsia, this solution can achieve early warning and provide an important basis for doctors' timely intervention. This not only optimizes the allocation efficiency of obstetric medical resources but also provides a strong guarantee for reducing maternal and infant risks.

[0032] Therefore, the technical solution of the present invention has important practical value in improving the performance of obstetric prediction models, promoting risk management during pregnancy and childbirth, and driving the development of intelligent healthcare.

[0033] The present invention provides an obstetrics and gynecology clinical pathway prediction method based on neuron-level feature selection, which divides the obstetrics and gynecology electronic medical record data of patients into temporal features and baseline features, including: Collect multi-modal data in the obstetrics and gynecology electronic medical records, where the multi-modal data includes: structured data and text data; Use natural language processing technology to perform semantic analysis on the text data, extract key semantics, and establish semantic association relationships between data items in combination with the meanings of clinical indicators in the structured data; Divide the multi-modal data into temporal features and baseline features based on the semantic association relationships.

[0034] In this embodiment, the multi-modal data is comprehensive information containing multiple data types.

[0035] The structured data is standardized data stored in tabular form, such as test values (white blood cell count 10.2×10 9 / L), vital signs (systolic blood pressure 120 mmHg), demographic information (age 28 years, gestational week 32 weeks), and directly read the database tables (such as test result tables, patient basic information tables).

[0036] The text data is unstructured free text, such as the chief complaint ("recurrent dizziness during pregnancy accompanied by lower limb edema"), medical records ("doctor's advice: monitor blood pressure daily, take nifedipine controlled-release tablets 30 mg orally"), and extract the original text from the medical record text fields (such as the chief complaint, course of disease record).

[0037] In this embodiment, the core information with medical significance in the text, such as the disease name ("gestational diabetes"), symptoms ("abdominal pain"), treatment measures ("cesarean section"), etc., can all be regarded as key semantics.

[0038] In this embodiment, the logical connection between structured data and text data, such as "elevated blood glucose" in the text corresponding to "fasting blood glucose value of 8.2 mmol / L" in the structured test data, is regarded as a semantic association relationship.

[0039] In this embodiment, for example, "blood pressure at the first antenatal examination: 110 / 70 mmHg" and "blood pressure at the second antenatal examination: 130 / 85 mmHg" are classified as time-series features; data that is independent of time and describes the patient's inherent attributes (such as "age 30 years" and "previous cesarean section history") are classified as baseline features. Filter noise by combining the semantic association strength. For example, "occasional headache mentioned by the patient" is not used as a baseline feature if it is not associated with the structured examination results. Dynamically divide based on the disease type. For patients with gestational hypertension, data related to blood pressure fluctuations are forcibly included in the time-series features.

[0040] The beneficial effects of the above technical solution are: converting unstructured text into computable semantic information to enhance data relevance, dividing time series and baseline based on semantic association relationships to ensure the accuracy of data division, and achieving a leap from "data stacking" to "semantic-driven" through the deep integration of NLP and structured data, providing high-quality feature input for subsequent tasks such as clinical pathway prediction and disease diagnosis.

[0041] The present invention provides an obstetrics and gynecology clinical pathway prediction method based on neuron-level feature selection. In the process of obtaining multi-dimensional prediction results of the clinical pathway according to the final feature representation, it further includes: Extracting the patient's disease type, treatment stage for each disease type, and disease severity from the medical history information, constructing a disease feature vector for each disease type, and performing cluster analysis on all disease feature vectors; Determining the influence relationship between each cluster center and the current feature vector of the patient under the most recent obstetrics and gynecology visit record based on the disease interaction relationship table, and inversely inferring the enlightenment effect on each feature based on the influence relationship. Among them, the enlightenment effect includes: positive enlightenment, negative enlightenment, and no enlightenment. Among them, positive enlightenment is assigned a value of 1, no enlightenment is assigned a value of 0, and negative enlightenment is assigned a value of -1; Sorting the values of all enlightenment effects under each feature in order of the magnitude of the influence relationship from large to small at one time, and at the same time, sorting the values of all enlightenment effects under each feature in order of the magnitude of the assigned enlightenment values from large to small at a second time; Perform linear fitting on the curve obtained by the first sorting and the curve obtained by the second sorting respectively to obtain the first fitting value and the second fitting value; Sort the first fitting values of all features from largest to smallest to obtain the first serial number of each feature. At the same time, sort the second fitting values of all features from largest to smallest to obtain the second serial number of each feature. According to the first serial number and the second serial number, obtain the first rough attention coefficient; Determine the absolute value of the difference between the first fitting value of the linear fitting result of the curve obtained by the first sorting and the original value of the corresponding point, determine the locked horizontal interval around the corresponding point, and obtain the original average value of the horizontal interval and the linear average value , determine the consistency ; Correct the first rough attention coefficient according to the consistency to obtain the second rough attention coefficient, and use it as the important attention degree of the corresponding feature in the clinical path prediction process; The present invention provides an obstetrics and gynecology clinical path prediction method based on neuron-level feature selection, obtaining the second rough attention coefficient, including: wherein, represents the second rough attention coefficient; represents the first rough attention coefficient; represents the adjustment parameter; represents the corresponding absolute value of the difference; b1 is a set threshold.

[0042] In this embodiment, natural language processing technology is used to extract keywords such as disease type, treatment stage, and disease severity from the medical history information text, and then convert them into numerical values through a preset coding rule to construct a disease feature vector. For example, from the text "The patient has gestational diabetes, is currently in the insulin treatment stage, and the condition is moderate", the information is extracted and encoded as [gestational diabetes coding, 2, 2]. Then, a clustering algorithm (such as K-means clustering) is used to cluster all disease feature vectors, and patients with similar disease features are grouped into the same class.

[0043] Establish a disease interaction relationship table, which can be preset based on medical knowledge and clinical experience to record the interaction relationships between different diseases. Then, calculate the similarity between each cluster center and the patient's current feature vector to determine the influence relationship. Based on the influence relationship table and the calculation results, judge the enlightenment effect of each feature and assign corresponding values. For example, if the cluster center of gestational diabetes has a high correlation with the patient's current feature vector of poor blood glucose control, and the disease interaction relationship table shows that gestational diabetes will increase the risk of premature birth, then the enlightenment effect of this feature is positive and a value of 1 is assigned.

[0044] In this embodiment, the enlightenment effect values under each feature are sorted once according to the size of the influence relationship to reflect the priority of the influence degree; then sorted a second time according to the size of the enlightenment value to highlight the differences in the influence direction. Linear fittings are respectively performed on the curves obtained from these two sorts to obtain the first fitting value and the second fitting value. For example, for multiple features related to the risk of premature birth, first sort them according to the strength of their influence relationship with the risk of premature birth, then sort them according to the positive or negative and size of the enlightenment effect, and then obtain the numerical values reflecting the change trend of the features through fitting.

[0045] The enlightenment effect represents the influence direction of the disease cluster center on the patient's current features. A positive enlightenment indicates that the disease factor has a promoting or associated effect on the current disease development (such as gestational hypertension may increase the risk of premature birth, and a value of 1 is assigned); a negative enlightenment indicates an inhibitory or reverse association (such as good prenatal nutritional intervention may reduce the risk of pregnancy complications, and a value of -1 is assigned); no enlightenment indicates no obvious association (such as the patient's previous irrelevant medical history, and a value of 0 is assigned).

[0046] The first fitting value and the second fitting value are the fitting numerical values at the central position points after linear fittings are respectively performed on the curves obtained from the first sort and the second sort.

[0047] Based on the disease interaction relationship table and clinical experience, considering the actual associations between diseases, it conforms to medical common sense and clinical practice. For example, the interaction relationships between common diseases such as gestational diabetes and pregnancy-induced hypertension have been clearly studied in medicine. The solution determines the influence relationships and enlightenment effects based on this, with solid medical evidence. By quantifying the medical history information and performing data analysis, complex text information is transformed into computable and comparable numerical values, avoiding the limitations of subjective judgment. Using methods such as cluster analysis and sorting fitting, valuable information is mined from a large amount of data to provide objective and accurate support for clinical pathway prediction. Considering the dynamic changes in the patient's condition, by continuously updating the medical history information and the current feature vector, the importance degree of features is re-evaluated, enabling the solution to adapt to the needs of different patients and different stages of the condition. For example, as the gestational week increases, the patient's disease status and treatment stage may change, and the solution can timely adjust the feature attention coefficient to provide real-time and effective reference for clinical decision-making.

[0048] In this embodiment, the determination of the horizontal interval is as follows: Assume that the absolute value of the difference is d, the abscissa of the corresponding point is x0, and the expansion multiple is set as k. At this time, the horizontal interval is: [x0 - kd, x0 + kd]. It should be noted that x0 is the central position point.

[0049] The expansion multiple k is obtained by matching from the difference absolute value - multiple comparison table. This table contains the expansion multiples involved under different absolute values of differences and can be directly used. Generally, the larger the absolute value of the difference, the worse the consistency, and the larger the corresponding expansion multiple. By examining the data within this interval, the overall situation of the data near this point can be analyzed more comprehensively, avoiding the excessive influence of the error or abnormality of a single point on the result, and thus providing a data basis for subsequent calculation of the original average value and the linear average value.

[0050] If there are n0 points in the horizontal interval, at this time, calculate the average value of the n0 points corresponding to the n0 points on the fitting line under the linear fitting situation and the average value of the n0 points on the corresponding curve before linear fitting .

[0051] In this embodiment, the consistency calculation formula is used to measure the consistency between the linear fitting result of a ranking curve and the original value. In clinical pathway prediction, by evaluating the consistency between the fitting value and the original value, it can be judged whether the trend of feature importance obtained based on ranking fitting is reliable, providing an effective reference for subsequent coefficient correction.

[0052] In this embodiment, in clinical pathway prediction, different degrees of adjustment are required for different fitting consistency situations. According to the actual fitting differences, reasonably determine the adjustment parameters so that when correcting the rough attention coefficients, the importance degree of features can be more accurately reflected. In clinical pathway prediction, through the formula, optimize the initially determined feature attention coefficients according to the fitting effect, so that the finally obtained second rough attention coefficients can more accurately reflect the importance degree of features in clinical pathway prediction, thereby assisting in more accurate clinical decision-making.

[0053] The beneficial effects of the above technical solution are as follows: Extract key disease features from medical history information, analyze the relationships between them and their impacts on the current condition, and quantify the importance degree of each feature in clinical pathway prediction. Thereby helping doctors more accurately grasp the patient's condition, formulate personalized clinical pathways, improve the quality and efficiency of medical treatment, and reduce the risk of adverse pregnancy outcomes.

[0054] In obstetrics and gynecology clinical practice, doctors need to comprehensively consider the local changes of patients' individual indicators and the development trend of the overall condition. For example, in pregnancy management, both the specific changes of various indicators of pregnant women and fetuses during each prenatal examination need to be concerned, and the physiological development laws of the entire pregnancy need to be grasped. This solution meets the actual clinical needs by integrating local and global change laws, and helps to improve the accuracy and scientific nature of medical decision-making. Therefore, the following technical solution is proposed. The present invention provides an obstetrics and gynecology clinical pathway prediction method based on neuron-level feature selection, and each part of the time series features corresponds to a trained GRU network; Each part of the baseline features corresponds to a learnable embedding layer.

[0055] The present invention provides an obstetrics and gynecology clinical pathway prediction method based on neuron-level feature selection, and obtains the dynamic change laws of various clinical indicators over time, including: Obtain the feature information of the set indicators of each part in the time series features under the corresponding trained GRU network, and respectively extract the information structure under the same set indicator from all the feature information to obtain the first dynamic change law vector corresponding to the set indicator, where the first dynamic change law vector includes the local change value of the i1th set indicator at time t; Based on the multi-scale feature extraction network, use convolution kernels of different sizes to perform multi-scale convolution operations on the input time series features, extract feature information of different granularities, and capture the global dynamic change law vector of the time series features, where the global dynamic change law vector includes the global change value at time t; Determine the temporal influence on each first dynamic change law vector based on the global dynamic change law vector: Wherein, is the global dynamic change law vector The time - series influence on the i - th first dynamic change law vector ; , respectively represent parameter vectors; Based on the wisdom layer, the first dynamic change law vectors of all set indicators and the global dynamic change law vector are merged, and the dynamic change laws of various indicators over time are obtained by combining the time - series influences under different indicators. Among them, the outputs of the GRU network and the multi - scale feature extraction network are used as the inputs of the wisdom layer.

[0056] In this embodiment, the set indicators are key medical indicators pre - determined for measurement and analysis in the time - series feature analysis of electronic medical records in obstetrics and gynecology. For example, in obstetrics, it may include uterine contraction frequency, fetal heart rate, maternal blood pressure, blood glucose value, etc.; in gynecology, it may have hormone levels (such as estrogen, progesterone), relevant parameters of pelvic ultrasound examination (such as uterine size, endometrial thickness), etc.

[0057] The feature information is the information obtained by processing the time - series features through the GRU network or the multi - scale feature extraction network, which reflects the internal laws and characteristics of the set indicators. For example, after the GRU network processes the uterine contraction frequency data, the information about the change trend and fluctuation range of the uterine contraction frequency over time is obtained.

[0058] The information structure is the organizational and presentation form of the feature information in the time dimension under the same set indicator. For example, the measured values of a certain maternal's blood pressure over a period of time arranged in chronological order, as well as the change relationship between these measured values, etc., constitute the information structure of the set indicator of blood pressure.

[0059] The first dynamic change law vector is a vector obtained by the GRU network for a certain set indicator, which reflects the change of the indicator within a local time range. For example, for the set indicator of fetal heart rate, at this time, the first dynamic change law vector contains the change value of the fetal heart rate within every 5 minutes. For example, at time t = 30 minutes, the fetal heart rate increases by 10 beats per minute compared with the previous time point, and this change value is an element in the vector.

[0060] The global dynamic change law vector contains the comprehensive change trends of the vital signs (such as blood pressure, heart rate, etc.) and test indicators (such as various indicators in blood routine) of the entire obstetrics and gynecology patient group at a certain time point. The 3×1 small convolution kernel captures the local feature changes during each uterine contraction, and the 7×1 large convolution kernel captures the overall trend of these features during the entire labor process. Finally, a vector reflecting the global changes of the entire delivery process is obtained.

[0061] In this embodiment, the intelligent layer integrates these vectors by combining the temporal impacts calculated previously. By means of weighted fusion, feature splicing, etc., the local change rules and global change rules are organically combined to obtain the dynamic change rules of various indicators over time. For example, for the delivery process of a parturient, the intelligent layer fuses the local change rule of the uterine contraction intensity and the global change rule of the entire labor process to obtain the final rule comprehensively reflecting how the uterine contraction intensity changes under the influence of the overall labor process during the entire labor process.

[0062] In this embodiment, is a value between 0 and 1, and this value is used to measure the influence degree of the global change on the local change of a specific indicator at time t. refers to the vector formed by splicing, which is to comprehensively consider the information of local change and global change in a unified vector space. The parameter vector 、 can both be learned and optimized through the training process of the model. During the training process, by continuously adjusting these parameters, the model can find the most suitable correlation relationship between the global change and the local change according to the actual data, so that the calculated temporal impact is more in line with the actual situation.

[0063] The beneficial effects of the above technical solution are as follows: Based on the GRU network, it can accurately capture the dynamic change details of each set indicator within the local time range, providing a basis for subsequent analysis. Determining the global change rule to grasp the overall change rule of the temporal characteristics from a macroscopic perspective, avoiding only focusing on local details while ignoring the overall trend. Based on the temporal impact, the time correlation and interaction between the global change and the local change are clarified, providing key parameters for subsequent information merging. Based on the fusion of vectors by the intelligent layer, more comprehensive and accurate dynamic change rules of various indicators can be obtained, providing strong support for the prediction of obstetrics and gynecology clinical pathways, disease diagnosis, and treatment decisions.

[0064] There are numerous and complex relationships among obstetrics and gynecology clinical indicators. This solution aims to, through a series of processes, explore the internal relationships among clinical indicators, quantify the importance of features, and screen out key features. Using these key features to more accurately analyze the dynamic change rules of various clinical indicators over time, providing strong support for clinical pathway prediction, disease diagnosis, and treatment. Projecting the dynamic change rules of various clinical indicators over time and the feature vectors onto a new representation space based on the feature importance to generate a gating vector, including: Constructing a graph structure by mapping the mapping elements mapped to the new representation space and the associated clinical indicator features, and deeply analyzing the graph structure based on a graph neural network to obtain the mapping logic; Convert the mapping logic into a structured standard representation and input it into a preset representation filling table to obtain a representation vector and representation weights, where the preset representation filling table is a multi-layer perceptron model that takes the semantics and structural information of the standard representation as input and outputs the representation vector and representation weights; Construct a projection matrix based on all the representation vectors and representation weights and use a regularization method to constrain the projection matrix, where and respectively represent the representation vectors corresponding to the 1st mapping element and the nth mapping element; and respectively represent the representation weights corresponding to the 1st mapping element and the nth mapping element; Quantize the projection matrix using an asymmetric quantization method, map the continuous numerical values in the matrix to a finite discrete numerical value set, and combine a generative adversarial network to enhance the representation ability of the projection matrix; Perform a multiplication operation on the enhanced projection matrix and the original feature vector to obtain the projection result of each feature, and evaluate the importance of the feature based on the numerical size and distribution of the projection result in combination with the representation weights; Based on the feature importance score, use the gating mechanism of the gated recurrent unit to generate a gating vector.

[0065] In this embodiment, a mapping element refers to an element obtained after a clinical index feature is mapped to a new representation space. For example, in an obstetrics and gynecology medical record, for the clinical index feature of "the blood glucose value of a pregnant woman", after a certain transformation (such as a linear transformation or a non-linear transformation) is used to map it to a new space, the corresponding element obtained is the mapping element.

[0066] In this embodiment, a graph structure is a data structure composed of nodes and edges. In this scenario, the nodes can be mapping elements and clinical index features, and the edges represent the association relationships between them. For example, the mapping elements corresponding to the two clinical index features of "the blood glucose value of a pregnant woman" and "the insulin dosage used" can be used as nodes, and the possible causal relationship or correlation between them can be represented by an edge.

[0067] In this embodiment, a graph neural network learns the relationships and feature propagation between the nodes in the graph, and the mapping logic refers to the internal connection and action mechanism between the mapping elements and the clinical index features. For example, the logical relationship such as "the change in the diet structure of a pregnant woman will cause a change in the blood glucose value, which in turn affects the adjustment of the insulin dosage" obtained after analysis by the graph neural network.

[0068] In this embodiment, the structured standard representation is a representation format in which the mapping logic is organized according to certain specifications and formats. For example, the above mapping logic is organized into a structured form of "[reason (change in diet structure of pregnant women), result 1 (change in blood sugar level), result 2 (adjustment of insulin dosage)]".

[0069] In this embodiment, a filling table (multi-layer perceptron model) is preset, for example, structured information such as "the relationship between the age of pregnant women and the risk of premature birth" is input, and a vector representing the relationship and the important weight of the relationship in the overall feature is output.

[0070] In this embodiment, the representation vector is a vector form used to represent the mapping logic. For example, the mapping logic of "exercise during pregnancy and weight management of pregnant women" is represented by a vector (such as [0.3, 0.5, 0.2]), and the dimension and value of the vector are obtained according to the specific model setting and learning.

[0071] In this embodiment, the asymmetric quantization method is a method of converting continuous values ​​in a matrix into a finite number of discrete values, and the processing of positive and negative values ​​during the conversion process may be different. For example, the continuous weight values ​​[0.1, 0.3, -0.2] in the projection matrix are converted into discrete values ​​[1, 2, -1].

[0072] The gating vector is a binary or weighted vector, such as [1,0,1], which indicates that some features are open (value 1) and some features are closed (value 0).

[0073] The enhanced projection matrix is ​​multiplied by the original eigenvector to obtain the projection result of each feature. For example, the original eigenvector represents various indicators of pregnant women (age, gestational age, medical history, etc.). After the projection matrix transformation, a new vector is obtained. According to the numerical value and distribution of the projection result, combined with the representation weight, the importance of each feature is evaluated. For example, a feature with a large value and a high corresponding representation weight indicates that it is more important in the overall feature.

[0074] The beneficial effects of the above technical scheme are: graph structure and graph neural network can well handle the complex situation of obstetrics and gynecology clinical data with multiple indicators and relationships between indicators, and mine hidden information; multi-layer perceptron can quantify complex logic; operations such as projection matrix can effectively evaluate and screen features. The entire scheme is adapted to the characteristics of clinical data. By generating gated vectors to screen features, it can help doctors focus on the key points from a large amount of clinical data and improve decision-making efficiency and accuracy. The scheme integrates multiple advanced technologies such as graph neural networks, multi-layer perceptrons, and generative adversarial networks. Each technology plays a unique role in the corresponding link, and processes and analyzes clinical data from different angles. The technical integration is scientific and reasonable, and can effectively achieve the design goals.

[0075] In tasks such as the prediction of clinical pathways and disease diagnosis in obstetrics and gynecology, the features of the main task layer are the core concerns, but there may be incomplete information. Although the features of the auxiliary task layer do not directly participate in the main task prediction, they may contain valuable supplementary information for the main task. This solution aims to screen the features of the main task layer through gating vectors to improve processing efficiency. At the same time, it uses the auxiliary task layer to mine potential supplementary information, verify and supplement the features of the main task layer, so as to analyze the condition more comprehensively and accurately and provide a more reliable basis for clinical decision-making. Therefore, it is proposed to screen the features of the main task layer and the auxiliary task layer respectively, including: Screen the features of the main task layer according to the value division mechanism and using the gating vector; Mine the potential supplementary information of each gating vector based on the auxiliary task layer, verify the features of the main task layer after screening according to the potential supplementary information, and supplement the screened features in the main task layer according to the verification correlation relationship.

[0076] In this embodiment, the feature corresponding to the element with a value of 1 in the gating vector is retained, and the feature corresponding to the element with a value of 0 is removed, which is the value division mechanism.

[0077] In this embodiment, the main task layer is a set of features directly related to the main clinical prediction or diagnosis task. For example, in the task of predicting whether a pregnant woman will have a premature birth, the features of the main task layer may include indicators directly related to premature birth such as the gestational age of the pregnant woman, the frequency of uterine contractions, and the cervical length. The auxiliary task layer is a set of features that provide supplementary information. These features do not directly participate in the prediction of the main task, but are helpful for understanding the overall condition or assisting in the judgment of the main task. For example, in premature birth prediction, the features of the auxiliary task layer may include the psychological state of the pregnant woman, the detailed situation of nutritional intake during pregnancy, etc. Although these information do not directly determine whether there is a premature birth, they may have an indirect impact on the risk of premature birth.

[0078] In this embodiment, the potential supplementary information is information hidden in the features of the auxiliary task layer that can supplement or verify the features of the main task layer after screening. For example, when analyzing the risk of premature birth in a pregnant woman, the feature of "whether the pregnant woman has had a mental stress event recently" in the auxiliary task layer may contain potential supplementary information that affects the body's stress response of the pregnant woman and thus indirectly affects the risk of premature birth.

[0079] In this embodiment, the verification correlation relationship is the logical connection and mutual verification relationship between the potential supplementary information of the auxiliary task layer and the features of the main task layer after screening. For example, if "abnormal uterine contraction frequency of the pregnant woman" is selected as a key feature in the main task layer, the potential supplementary information of "the pregnant woman has been under great mental stress recently" in the auxiliary task layer may verify and supplement the features of the main task layer through an association relationship such as "great mental stress can lead to abnormal uterine contraction frequency".

[0080] In this embodiment, the verification association relationship is the logical connection and mutual verification relationship between the potential supplementary information in the auxiliary task layer and the features selected in the main task layer. For example, if the main task layer selects "abnormal uterine contraction frequency of pregnant women" as a key feature, the potential supplementary information "high mental stress of pregnant women recently" in the auxiliary task layer may verify and supplement the features in the main task layer through an association relationship such as "high mental stress can lead to abnormal uterine contraction frequency".

[0081] The beneficial effects of the above technical solution are as follows: Through the screening of the gating vector, features with relatively low relevance to the current main clinical task in the main task layer can be removed, redundant information can be reduced, the computational efficiency and prediction accuracy of the model can be improved, the potential supplementary information in the auxiliary task layer can be mined and used to verify and supplement the features in the main task layer, the feature dimension can be enriched, the model can consider more factors that may affect clinical outcomes, and the comprehensiveness and reliability of the model can be improved.

[0082] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection, characterized in that, Including: Step 1: Divide the electronic medical record data of the patient's obstetrics and gynecology department into time-series features and baseline features, where the time-series features are test data, vital signs, and medical records, and the baseline features are demographic features and medical history information; Step 2: Input the time-series features into the trained GRU network for feature encoding to obtain the dynamic change rules of various clinical indicators over time. At the same time, convert the baseline features into feature vectors through the embedding layer; Step 3: Calculate the feature importance based on the patient's most recent obstetrics and gynecology visit record and the visit scenario. Project the dynamic change rules of various clinical indicators over time and the feature vectors onto a new representation space based on the feature importance to generate a gating vector, and respectively screen and fuse the features of the main task layer and the auxiliary task layer to obtain the final feature representation, where the feature importance calculation and feature screening constitute a neuron-level feature selection mechanism; Step 4: Obtain the multi-dimensional prediction results of the clinical pathway according to the final feature representation.

2. The method for predicting the obstetrics and gynecology clinical pathway based on neuron-level feature selection according to claim 1, wherein Dividing the electronic medical record data of the patient's obstetrics and gynecology department into time-series features and baseline features includes: Collect multi-modal data in the electronic medical record of obstetrics and gynecology, where the multi-modal data includes: structured data and text data; Use natural language processing technology to perform semantic analysis on the text data, extract key semantics, and establish semantic association relationships between data items in combination with the meanings of clinical indicators in the structured data; Divide the multi-modal data into time-series features and baseline features based on the semantic association relationships.

3. The method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection according to claim 1, wherein, In the process of obtaining the multi-dimensional prediction results of the clinical pathway according to the final feature representation, it also includes: Extract the disease types of the patient, the treatment stages for each disease type, and the severity of the condition from the medical history information, construct disease feature vectors for each disease type, and perform clustering analysis on all disease feature vectors; Determine the influence relationship between each cluster center and the current feature vector of the patient under the most recent obstetrics and gynecology visit record based on the disease interaction relationship table, and reverse infer the enlightenment effect on each feature based on the influence relationship, where the enlightenment effect includes: positive enlightenment, negative enlightenment, and no enlightenment, where positive enlightenment is assigned a value of 1, no enlightenment is assigned a value of 0, and negative enlightenment is assigned a value of -1; Sort the values of all enlightenment effects under each feature in order from largest to smallest according to the influence relationship, and at the same time, sort the values of all enlightenment effects under each feature in order from largest to smallest according to the assigned enlightenment values for the second time; Perform linear fitting on the curve obtained from the first sorting and the curve obtained from the second sorting to obtain the first fitting value and the second fitting value; Sort all the first fitting values of the features in order from largest to smallest to obtain the first serial number of each feature. At the same time, sort all the second fitting values of the features in order from largest to smallest to obtain the second serial number of each feature. According to the first serial number and the second serial number, obtain the first rough attention coefficient; Determine the absolute value of the difference between the first fitting value of the linear fitting result of the curve obtained by the first sorting and the original value of the corresponding point, determine the locked horizontal interval around the corresponding point, and obtain the original average value of the horizontal interval and the linear average value , determine the consistency ; The first rough attention coefficient is corrected according to the consistency to obtain a second rough attention coefficient, which is used as the important attention degree of the corresponding feature in the clinical pathway prediction process.

4. The method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection according to claim 1, wherein, Obtaining the second rough attention coefficient includes: ； ； Among them, represents the second rough attention coefficient; represents the first rough attention coefficient; represents the adjustment parameter; represents the corresponding absolute value of the difference; b1 is the set threshold.

5. The method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection according to claim 1, characterized in that Each part in the temporal features corresponds to a trained GRU network; Each part in the baseline features corresponds to a learnable embedding layer.

6. The method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection according to claim 5, wherein Obtaining the dynamic change rules of various clinical indicators over time includes: Obtaining the feature information of the set indicators of each part in the temporal features under the corresponding trained GRU network, and respectively extracting the information structure under the same set indicator from all the feature information to obtain the first dynamic change rule vector corresponding to the set indicator, where the first dynamic change rule vector includes the local change value of the i1-th set indicator at time t; Based on the multi-scale feature extraction network using convolution kernels of different sizes, performing multi-scale convolution operations on the input temporal features to extract feature information of different granularities, and capturing the global dynamic change rule vector of the temporal features, where the global dynamic change rule vector includes the global change value at time t; determining the temporal influence on each first dynamic change rule vector based on the global dynamic change rule vector: ； Among them, is the global dynamic change law vector the timing influence on the i-th first dynamic change law vector ; , respectively represent parameter vectors; Based on the intelligent layer, the first dynamic change rule vectors of all set indicators and the global dynamic change rule vector are combined, and the dynamic change rules of various indicators over time are obtained by combining the temporal influence under different indicators, where the output of the GRU network and the output of the multi-scale feature extraction network are used as the input of the intelligent layer.

7. The method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection according to claim 1, wherein, Projecting the dynamic change rules of various clinical indicators over time and the feature vectors based on the feature importance into a new representation space to generate a gating vector, including: Constructing a graph structure with the mapped elements mapped to the new representation space and the associated clinical indicator features, and deeply analyzing the graph structure based on the graph neural network to obtain the mapping logic; Converting the mapping logic into a structured standard representation and inputting it into a preset representation filling table to obtain a representation vector and a representation weight, where the preset representation filling table is a multi-layer perceptron model, taking the semantic and structural information of the standard representation as the input and the representation vector and the representation weight as the output; Construct a projection matrix based on all the representation vectors and representation weights , and use a regularization method to constrain the projection matrix, where , Bn respectively represent the representation vectors corresponding to the first mapping element and the nth mapping element; , respectively represent the representation weights corresponding to the first mapping element and the nth mapping element; Quantifying the projection matrix using an asymmetric quantization method, mapping the continuous numerical values in the matrix to a finite discrete numerical set, and enhancing the representation ability of the projection matrix by combining with a generative adversarial network; Performing a multiplication operation on the enhanced projection matrix and the original feature vector to obtain the projection result of each feature, and evaluating the importance of the feature by combining the representation weight according to the numerical size and distribution of the projection result; Based on the feature importance score, using the gating mechanism of the gated recurrent unit to generate a gating vector.

8. The method for predicting the clinical pathway of obstetrics and gynecology based on neuron-level feature selection according to claim 1, wherein Screening the features of the main task layer and the auxiliary task layer respectively, including: Screening the features of the main task layer according to the value division mechanism and using the gating vector; Mine the potential supplementary information of each gating vector based on the auxiliary task layer, verify the features filtered by the main task layer according to the potential supplementary information, and supplement the features of the filtered features in the main task layer according to the verification correlation relationship.

Images