An intelligent settlement scheduling method based on patient trajectory prediction

Abstract

The application discloses an intelligent settlement scheduling method based on patient trajectory prediction, and relates to the technical field of medical operation management. The method comprises the following steps: obtaining diagnosis and treatment process data of a target patient and other reference patients, and dividing the diagnosis and treatment process data into multiple diagnosis and treatment stages; based on the diagnosis and treatment similarity of the target patient and each reference patient in each diagnosis and treatment stage, the feature contribution degree of each information dimension in each diagnosis and treatment stage is calculated, and the information dimension whose feature contribution degree meets a preset condition is taken as an attention dimension; when the diagnosis and treatment process data of the target patient is detected to be updated, the updated diagnosis and treatment trajectory feature sequence is calculated based on the attention weight corresponding to the attention dimension; and based on the updated diagnosis and treatment trajectory feature sequence, the target patient is subjected to settlement resource scheduling. The application achieves the technical effect of reducing data calculation amount and quickly and accurately predicting the diagnosis and treatment trajectory of the patient.

Description

Technical Field

[0001] This invention relates to the field of medical operations management technology, specifically to an intelligent billing and scheduling method based on patient trajectory prediction. Background Technology

[0002] In traditional hospital operation models, patient treatment processes, medical resource allocation, and final billing are often independent of each other. This leads to patients experiencing long waits and cumbersome procedures due to a lack of information transparency. With the development of various AI technologies, it is possible to rely on large-scale, high-quality longitudinal medical data to build predictive models that can accurately predict patients' future treatment paths. Then, based on the patient's treatment path predicted by the predictive model and combined with the real-time status of hospital resources, the scheduling decision for the patient billing process can be realized through corresponding planning and optimization algorithms.

[0003] In related technologies, methods such as Principal Component Analysis (PCA) are used to reduce the feature dimensions of patients' relevant medical data, thereby retaining the most valuable feature subsets in the patient's medical data. However, this method requires the continuous construction of new feature spaces and iteration. When the target patient updates their examination data or medical status, the prediction model needs to be updated again based on the dimensions of the target patient's medical data. Frequent data updates mean that all dimensions need to be recalculated each time, which can easily cause a surge in the computational load of the PCA algorithm and even an explosion of feature combinations. This makes it difficult to predict the patient's medical trajectory quickly and accurately, and ultimately fails to complete the scheduling of the patient's settlement plan. Summary of the Invention

[0004] To address the technical problem in related technologies where methods such as Principal Component Analysis (PCA) require recalculation of all dimensions during data updates, leading to a surge in computational load and even feature combination explosion, making it difficult to quickly and accurately predict patients' treatment trajectories, this invention provides an intelligent settlement and scheduling method based on patient trajectory prediction.

[0005] The specific technical solution adopted is as follows:

[0006] Acquire treatment process data for the target patient and other reference patients, and divide the treatment process data into multiple treatment stages;

[0007] Based on the similarity of diagnosis and treatment between the target patient and each reference patient at each stage of diagnosis and treatment, the feature contribution of each information dimension in each stage of diagnosis and treatment is calculated. The information dimension corresponds to the data type of the examination items in the diagnosis and treatment process.

[0008] Information dimensions whose feature contribution meets preset conditions are taken as the focus dimensions;

[0009] When an update to the treatment process data of a target patient is detected, the updated treatment trajectory feature sequence is calculated based on the attention weight corresponding to the attention dimension.

[0010] Based on the updated diagnosis and treatment trajectory feature sequence, settlement resources are allocated for target patients.

[0011] In one possible implementation of this application, based on the similarity of treatment between the target patient and each reference patient at each stage of treatment, the feature contribution of each information dimension in each stage of treatment is calculated, including:

[0012] Feature extraction is performed on the examination results corresponding to the diagnosis and treatment process data at each stage of diagnosis and treatment to obtain feature vectors of different information dimensions, and the feature vectors are marked as feature descriptors;

[0013] For any stage of diagnosis and treatment, the item similarity between the target patient and the reference patient in each information dimension is calculated based on the cosine similarity of the feature descriptors of different patients in each information dimension.

[0014] Based on project similarity and a preset similarity algorithm, the similarity of the treatment trajectories of the target patient and the reference patient during the treatment process is calculated.

[0015] Based on the preset clustering algorithm and the similarity of the diagnosis and treatment trajectory, the diagnosis and treatment process data of the target patient or the reference patient are clustered to determine the diagnosis and treatment type of the target patient;

[0016] Based on the difference in item similarity between the target patient and the reference patient corresponding to the treatment type of the target patient, the feature contribution of each information dimension in each treatment stage is calculated.

[0017] In one possible implementation of this application, the item similarity between the target patient and the reference patient across each information dimension is calculated based on the cosine similarity of the feature descriptors of different patients across each information dimension, including:

[0018] Obtain the first feature descriptor corresponding to the multiple examination results of the target patient in each information dimension, and the second feature descriptor corresponding to the reference patient;

[0019] Determine the maximum cosine similarity between the first feature descriptor and the second feature descriptor;

[0020] The maximum similarity values ​​of multiple examination items corresponding to each information dimension are summed and averaged to obtain the item similarity between the target patient and the reference patient in each information dimension.

[0021] In one possible implementation of this application, the similarity of the treatment trajectories of the target patient and the reference patient during the treatment process is calculated based on project similarity and a preset similarity algorithm, including:

[0022] Determine the number of examinations performed by the target patient or reference patient at each stage of diagnosis and treatment, and calculate the first product between the similarity of the examination items and the initial attention weights corresponding to the information dimensions.

[0023] Based on the first product, the number of examination items, and the initial attention weight, the consistency of the condition between the target patient and the reference patient at the same stage of diagnosis and treatment is calculated.

[0024] The reciprocal of the consistency of disease symptoms is used as the sequence distance between the corresponding sequences of each feature descriptor;

[0025] Based on sequence distance and a preset similarity algorithm, the stage similarity between the target patient and the reference patient at each stage of diagnosis and treatment is calculated.

[0026] The maximum value of the stage similarity at each stage of diagnosis and treatment is set as the similarity of the diagnosis and treatment trajectory between the target patient and the reference patient during the diagnosis and treatment process.

[0027] In one possible implementation of this application, clustering is performed on the treatment process data of the target patient or reference patient based on a preset clustering algorithm and treatment trajectory similarity to determine the treatment type of the target patient, including:

[0028] The data space is constructed by using the diagnosis and treatment process data of the target patient or reference patient as data points and each examination item as a data dimension;

[0029] The feature distance between each data point is determined based on the reciprocal of the similarity of the diagnosis and treatment trajectory.

[0030] Based on a preset clustering algorithm and feature distance, each data point in the data space is clustered to obtain multiple clusters;

[0031] The treatment type corresponding to the cluster to which the target patient belongs is used as the treatment type for the target patient.

[0032] In one possible implementation of this application, based on the difference in item similarity between the target patient and the reference patient corresponding to the target patient's treatment type, the feature contribution of each information dimension in each treatment stage is calculated, including:

[0033] Calculate the first average of the item similarity between any two patients in the target patient and each reference patient;

[0034] Determine the first standard deviation between the first means, and perform negative correlation normalization on the first standard deviation to obtain the feature contribution of each information dimension in each stage of diagnosis and treatment.

[0035] In one possible implementation of this application, before calculating the updated treatment trajectory feature sequence based on the attention weight corresponding to the attention dimension when an update to the treatment process data of the target patient is detected, the method further includes:

[0036] Principal component analysis was used to decentralize the treatment process data for multiple patients in the target patient treatment type.

[0037] Based on the feature vectors corresponding to each dimension of interest, a new feature space is constructed after dimensionality reduction.

[0038] The feature descriptors within each dimension of interest are projected into a new feature space to obtain the first feature vector of the target patient at the current stage of diagnosis and treatment, and the first feature vector is marked as the feature of the diagnosis and treatment stage.

[0039] The treatment stage characteristics of each patient in each treatment stage within the target patient treatment type are arranged in chronological order to obtain the treatment stage characteristic sequence.

[0040] In one possible implementation of this application, when an update to the treatment process data of a target patient is detected, an updated treatment trajectory feature sequence is calculated based on the attention weight corresponding to the attention dimension, including:

[0041] When an update to the treatment process data of a target patient is detected, the feature sequence of the treatment stage is updated based on the attention weight corresponding to the attention dimension, resulting in an updated treatment trajectory feature sequence.

[0042] In one possible implementation of this application, the feature sequence of the diagnosis and treatment stage is updated based on the attention weight corresponding to the attention dimension to obtain the updated diagnosis and treatment trajectory feature sequence, including:

[0043] Based on the attention weights corresponding to the attention dimensions, update the initial attention weights of each information dimension to obtain the updated weights;

[0044] The updated treatment stage feature sequence is then updated with the updated weights to obtain the updated treatment trajectory feature sequence.

[0045] In one possible implementation of this application, settlement resource scheduling for target patients is performed based on the updated diagnosis and treatment trajectory feature sequence, including:

[0046] The updated diagnosis and treatment trajectory feature sequence is input into the LSTM model. Based on the LSTM model, the updated diagnosis and treatment trajectory feature sequence is predicted to obtain the diagnosis and treatment trajectory corresponding to the target patient in the next time period.

[0047] Based on the diagnosis and treatment trajectory, a multi-objective optimization algorithm is used to schedule settlement resources for target patients.

[0048] The present invention has, but is not limited to, the following technical effects:

[0049] By acquiring treatment process data from the target patient and other reference patients, and dividing the treatment process data into multiple treatment stages, the feature contribution of various information dimensions in each treatment stage is calculated based on the similarity of treatment between the target patient and each reference patient at each stage. During the patient's treatment process, there are various examination items, and each information dimension corresponds to the data type of the examination items in the treatment process. Therefore, the importance of each examination item in the treatment process is determined. When the feature contribution meets preset conditions, a focus dimension is selected from each information dimension. When an update to the target patient's treatment process data is detected, the updated treatment trajectory feature sequence is calculated based on the focus dimension and its corresponding focus weight. Then, the settlement resource scheduling for the target patient is performed based on the updated treatment trajectory feature sequence. When the target patient updates its examination data or treatment status, the treatment trajectory feature sequence of the target patient can be updated based on the calculated focus dimension and focus weight, thus eliminating the need to recalculate the data for all information dimensions, reducing the computational load of the treatment process data, and thereby quickly and accurately predicting the patient's treatment trajectory and completing the scheduling of the patient's settlement plan. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating the first embodiment of the intelligent billing and scheduling method based on patient trajectory prediction in this application;

[0051] Figure 2 This is a flowchart illustrating the second embodiment of the intelligent billing and scheduling method based on patient trajectory prediction in this application;

[0052] Figure 3 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application. Detailed Implementation

[0053] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0054] This application provides an intelligent billing and scheduling method based on patient trajectory prediction. In the first embodiment of this application's intelligent billing and scheduling method based on patient trajectory prediction, referring to... Figure 1 The methods include:

[0055] Step S10: Obtain the treatment process data of the target patient and other reference patients, and divide the treatment process data into multiple treatment stages;

[0056] Step S20: Based on the similarity of diagnosis and treatment between the target patient and each reference patient at each stage of diagnosis and treatment, calculate the feature contribution of each information dimension in each stage of diagnosis and treatment. The information dimension corresponds to the data type of the examination items in the diagnosis and treatment process.

[0057] Step S30: Select the information dimensions whose feature contribution meets the preset conditions as the focus dimensions;

[0058] Step S40: When an update to the treatment process data of the target patient is detected, the updated treatment trajectory feature sequence is calculated based on the attention weight corresponding to the attention dimension.

[0059] Step S50: Based on the updated diagnosis and treatment trajectory feature sequence, schedule settlement resources for the target patient.

[0060] This embodiment aims to: when the target patient updates their examination data or treatment status, the treatment trajectory feature sequence of the target patient can be updated based on the calculated attention dimensions and attention weights, thereby eliminating the need to recalculate the data for all information dimensions, reducing the computational load of the treatment process data, and thus quickly and accurately predicting the patient's treatment trajectory and completing the scheduling of the patient's settlement plan.

[0061] Step S10: Obtain the treatment process data of the target patient and other reference patients, and divide the treatment process data into multiple treatment stages.

[0062] As an example, the intelligent settlement scheduling method based on patient trajectory prediction can be applied to an intelligent settlement scheduling device based on patient trajectory prediction. The intelligent settlement scheduling device based on patient trajectory prediction belongs to an intelligent settlement scheduling system based on patient trajectory prediction, and the intelligent settlement scheduling system based on patient trajectory prediction belongs to an intelligent settlement scheduling equipment based on patient trajectory prediction.

[0063] As an example, the diagnosis and treatment process data mainly refers to the diagnosis and treatment data generated by various diagnosis and treatment activities during a patient's visit to the hospital, such as registration and payment data, examination results output by the system during medical examinations, and medical imaging data designed during CT scans and other image examinations. The diagnosis and treatment process data can be extracted from the hospital system's database, and then the patient's medical record information can be retrieved to collect the patient's corresponding medical history and various basic information, which together constitute the diagnosis and treatment process data of each patient.

[0064] As an example, the target patient is the patient whose treatment trajectory needs to be predicted, while the reference patient is a patient with the same or similar symptoms as the current target patient, and whose data is compared. Throughout the patient's diagnosis and treatment process, doctors will conduct various examinations, which will generate a series of data. Since doctors' judgment of the patient's disease depends on these examination data, and the patient's disease is certain, the entire treatment process is also roughly fixed. Therefore, examination data that does not belong to the original patient is not helpful in determining the patient's treatment trajectory. So, when predicting the patient's trajectory based on other patients with the same symptoms, it is not necessary to consider this data. Accordingly, using the examination data of the reference patient as a reference, the data dimensions of these examination data of different patients are filtered to determine the data dimensions of patients that need to be focused on at different stages, and then the treatment trajectory of the target patient is predicted based on this.

[0065] As an example, dividing the diagnosis and treatment process data into multiple stages can be done by using software to divide the data into stages according to the chronological order of the patient's medical events (such as diagnosis confirmation, surgery completion, treatment plan changes, etc.), thereby obtaining multiple diagnosis and treatment stages.

[0066] Step S20: Based on the similarity of diagnosis and treatment between the target patient and each reference patient at each stage of diagnosis and treatment, the feature contribution of each information dimension in each stage of diagnosis and treatment is calculated. The information dimension corresponds to the data type of the examination items in the diagnosis and treatment process.

[0067] As an example, if the target patient and each reference patient have similar symptoms, then different deviations will occur at each stage of diagnosis and treatment depending on the severity of the symptoms. Each stage of diagnosis and treatment involves multiple examinations, which correspond to information dimensions. For example, imaging examinations are one information dimension, and electrocardiogram (ECG) is another. Therefore, by calculating the similarity between the target patient and other reference patients at different stages of diagnosis and treatment, the importance of each information dimension, i.e., the feature contribution, is determined. The higher the feature contribution, the more attention the current information dimension needs to be paid, and the more important it is.

[0068] The intelligent billing scheduling step S20 based on patient trajectory prediction also includes steps S21-S25, including:

[0069] Step S21: Extract features from the examination results corresponding to the diagnosis and treatment process data at each stage of diagnosis and treatment to obtain feature vectors of different information dimensions, and mark the feature vectors as feature descriptors.

[0070] As an example, the diagnosis and treatment process for patients in the hospital is a long process. Each stage of diagnosis and treatment requires a series of tests, diagnoses, and procedures. These steps will generate corresponding data. However, some of this data is for observing the patient's physiological signs, and some is for observing the development of the patient's condition. Taking a stage of diagnosis and treatment as an example, there are many examination items for the patient in this stage. The results of each examination item correspond to a set of data. Therefore, we first extract features from each set of data to generate corresponding feature descriptors.

[0071] Specifically, for the text data in the inspection results, use In The bag-of-words model constructs word vectors for each word group in the text data;

[0072] For image data, first through The corner detection algorithm detects key feature points in the image, then takes a certain neighborhood (preset 16×16 pixel neighborhood) for these key feature points, and divides the neighborhood into 4×4 sub-blocks. Then, it calculates the gradient histogram in 8 directions for each sub-block, and links all the histograms together to form the corresponding high-dimensional feature vector. After these operations, or the feature vector corresponding to different information dimensions at various stages of diagnosis and treatment, the feature vector is marked as a feature descriptor.

[0073] Step S22: For any stage of diagnosis and treatment, calculate the item similarity between the target patient and the reference patient in each information dimension based on the cosine similarity of the feature descriptors of different patients in each information dimension.

[0074] As an example, for any stage of diagnosis and treatment, there will be multiple tests, which will correspond to multiple information dimensions. The cosine similarity of each information dimension in each stage of diagnosis and treatment is calculated to obtain the item similarity of each information dimension / test item. The item similarity is used to characterize the degree of similarity of the test items of two patients in any stage of diagnosis and treatment.

[0075] As an example, during a treatment phase, a patient may need to undergo several examinations for the same item. If the results of these examinations are similar, that is, the corresponding feature descriptors tend to be consistent, then it means that the examination results of the two patients on the same examination item during the treatment phase are basically consistent. Based on this, the item similarity of the data of the two patients on the same examination item (information dimension) can be constructed.

[0076] Step S22 includes:

[0077] Obtain the first feature descriptor corresponding to the multiple examination results of the target patient in each information dimension, and the second feature descriptor corresponding to the reference patient;

[0078] Determine the maximum cosine similarity between the first feature descriptor and the second feature descriptor;

[0079] As an example, the feature descriptor corresponding to each examination result of the target patient is determined as the first feature descriptor. Similarly, the second feature descriptor corresponding to the reference patient can be obtained.

[0080] As an example, the cosine similarity between the first feature descriptor and the second feature descriptor is calculated, and the maximum value of the cosine similarity among all information dimensions is determined to obtain the maximum similarity.

[0081] The maximum similarity values ​​of multiple examination items corresponding to each information dimension are summed and averaged to obtain the item similarity between the target patient and the reference patient in each information dimension.

[0082] As an example, to refer to patients The Treatment Stage and Target Patients The In the information dimension of the diagnosis and treatment stage Taking project similarity as an example, project similarity The calculation method is as follows:

[0083]

[0084] In the formula, Indicates target patients The In the diagnosis and treatment phase, the information dimension (examination items) The first The feature descriptor corresponding to the inspection result of this inspection; Indicates reference patient The In the information dimension of the diagnosis and treatment stage The first The feature descriptor corresponding to the inspection result of this inspection, where, equal This indicates the same stage of diagnosis and treatment. This indicates the calculation of the cosine similarity between two vectors. This represents a function that maximizes the value of a target patient. In the information dimension The first The patient whose test results were most similar to those of the previous test The corresponding inspection results; This represents the maximum similarity. This indicates the examinations the target patient will undergo during this stage of diagnosis and treatment. The total number of times.

[0085] Similarly, the similarity of the results of two items can be obtained by checking other information dimensions. The higher the similarity of the item, the more consistent the patients are in that clinical presentation. The target patient can be selected from various patients and does not necessarily have to be a fixed patient.

[0086] Step S23: Based on the project similarity and the preset similarity algorithm, calculate the similarity of the treatment trajectory between the target patient and the reference patient during the treatment process.

[0087] As an example, after calculating the item similarity between patients in each information dimension, it is also necessary to calculate the disease similarity between patients based on the item similarity. This is to characterize the degree of similarity between the diseases of different patients. For the same disease, the development of different patients may not be the same. There may be a situation where one person's condition is severe and another person's condition is mild. This will lead to slight differences in the treatment plans of the two patients, and therefore their treatment trajectories will also be different. Based on this, according to the item similarity and the preset similarity algorithm, the treatment trajectory similarity between the target patient and the reference patient in the treatment process is calculated. The treatment trajectory similarity is used to characterize the degree of similarity between the target patient and the reference patient in the treatment process.

[0088] Step S23 includes:

[0089] Determine the number of examinations performed by the target or reference patient at each stage of diagnosis and treatment, and calculate the first product between the similarity of the examinations and the initial attention weights corresponding to the information dimensions.

[0090] Based on the first product, the number of examination items, and the initial attention weight, the consistency of the condition between the target patient and the reference patient at the same stage of diagnosis and treatment is calculated.

[0091] As an example, when diagnosing and treating a patient's illness, a doctor will determine the specific disease and treatment plan based on the patient's various clinical manifestations. If the patients' clinical manifestations are similar, that is, if the patients' characteristics are similar in every information dimension, then it means that the two patients' conditions are consistent, and their treatment plans will also be consistent. Based on this, the consistency of the condition of two patients in the same diagnosis and treatment stage can be constructed, and the consistency of the condition of the target patient and the reference patient in the same diagnosis and treatment stage can be calculated.

[0092] As an example, to refer to patients The Treatment Stage and Target Patients The Taking the diagnosis and treatment phase as an example, the consistency of the patient's condition at this stage... The calculation method can be:

[0093]

[0094] In the formula, Indicates the patient The Treatment Stage and Target Patients The In the information dimension of the diagnosis and treatment stage Project similarity; Represents the first product. It indicates that the patients were in the first The total number of examinations (information dimension) performed during the diagnosis and treatment phase, i.e., the number of examination items. Indicating the current stage of diagnosis and treatment Information Dimensions The corresponding initial attention weight is preset to 1.

[0095] The reciprocal of the consistency of disease symptoms is used as the sequence distance between the corresponding sequences of each feature descriptor;

[0096] Based on sequence distance and a preset similarity algorithm, the stage similarity between the target patient and the reference patient at each stage of diagnosis and treatment is calculated.

[0097] As an example, when calculating the similarity of treatment trajectories, it is considered that two patients may have some similarities in their treatment trajectories. For example, they may have the same disease, but their progression may be different, one severe and the other mild. This will lead to slight differences in the treatment plans of the two patients. The patient with a more severe condition needs to undergo a series of treatments to alleviate the condition to a milder level, and then their treatment plan will be similar to that of the patient with a milder condition to some extent. Based on this, by matching the clinical manifestation data such as the patient's physiological condition and treatment plan, the treatment trajectories of each patient are aligned in chronological order. The process of aligning treatment trajectories can be as follows:

[0098] First, a comparison window is established using the total length of the treatment stages of the target patient's treatment trajectory as the window length. Then, a sliding window of corresponding length is set on the treatment trajectories of other reference patients using the established comparison window, with the step size preset to one treatment stage.

[0099] As an example, the default similarity algorithm is: (Dynamic Time Warping) algorithm, using The algorithm uses a reference patient. With target patients The reciprocal of the consistency of the condition between two corresponding treatment stages is used as the corresponding distance metric to obtain the sequence distance. Algorithm for processing target patients Throughout the diagnosis and treatment process with the patient The DTW similarity between the temporal sequences of feature descriptors in their respective information dimensions at the corresponding diagnosis and treatment stage in the sliding window h is denoted as the stage similarity. .

[0100] The maximum value of the stage similarity at each stage of diagnosis and treatment is set as the similarity of the diagnosis and treatment trajectory between the target patient and the reference patient during the diagnosis and treatment process.

[0101] As an example, the target patient was identified. The diagnosis and treatment process and reference patients In each sliding window, the similarity of clinical manifestations at each stage is calculated, and the maximum value is selected and recorded as the target patient. Treatment process and reference patients Similarity of treatment trajectories during the diagnosis and treatment process And record their corresponding relationships; similar operations can be performed to obtain target patients. The similarity between clinical manifestations and treatment data from other patients in the diagnosis and treatment process, i.e., the similarity of the corresponding diagnosis and treatment trajectory. .

[0102] Step S24: Based on the preset clustering algorithm and the similarity of the treatment trajectory, cluster the treatment process data of the target patient or the reference patient to determine the treatment type of the target patient.

[0103] As an example, if two patients have similar clinical manifestations and corresponding test results at multiple stages of their diagnosis and treatment, their treatment process will also tend to be consistent. Therefore, based on the similarity of the diagnosis and treatment process of many patients, the treatment process (diagnosis and treatment trajectory) of many patients can be divided into many different types to determine the diagnosis and treatment type of the target patient and the diagnosis and treatment type of other reference patients.

[0104] As an example, data clustering is used to cluster the diagnosis and treatment process data of each patient, resulting in multiple clusters. The cluster containing the target patient is then labeled as the target patient's diagnosis and treatment type. The pre-defined clustering algorithm can be... Density clustering algorithm.

[0105] Step S24 includes:

[0106] The data space is constructed by using the treatment process data of the target patient or reference patient as data points and each examination item as a data dimension.

[0107] The feature distance between each data point is determined based on the reciprocal of the similarity of the diagnosis and treatment trajectories.

[0108] As an example, a data space is constructed by using each patient's diagnosis and treatment process as a data point and each examination item as a data dimension. The algorithm clusters data points corresponding to numerous patients, using the reciprocal of the similarity between the treatment trajectories of two patients. It serves as the feature distance between two data points.

[0109] Based on a preset clustering algorithm and feature distance, clustering is performed on each data point in the data space to obtain multiple clusters.

[0110] The treatment type corresponding to the cluster to which the target patient belongs is used as the treatment type for the target patient.

[0111] As an example, the minimum number of points is preset. The value is set to 10, and then the preset radius is determined using the k-distance graph method. When determining the preset radius using the k-distance diagram method Value is taken from preset By using a pre-defined clustering algorithm and feature distance, the data points in the data space are clustered to obtain multiple clusters. Patients and their corresponding treatment trajectories are divided into many types, and the treatment type corresponding to the cluster in which the target patient is located is taken as the treatment type of the target patient.

[0112] Step S25: Based on the difference in item similarity between the target patient and the reference patient corresponding to the treatment type of the target patient, calculate the feature contribution of each information dimension in each treatment stage.

[0113] As an example, based on the previous analysis, patients were divided into many types according to their clinical manifestations and changes. Among them, many patients in the same type have similar clinical manifestations. In the modern medical system, it is believed that each disease has corresponding clinical manifestations. Therefore, for patients with similar clinical manifestations, it is highly likely that they have the same disease, and their subsequent treatment and examination trajectories will also be the same.

[0114] Based on this, the commonalities in the clinical manifestations of patients with the same disease are the characteristic manifestations of that type of disease. Conversely, if the feedback in a certain examination dimension does not have commonalities or has low commonalities, it means that the dimension corresponding to that examination item is not a characteristic dimension of that disease. Based on this, the probability of constructing various information dimensions (examination items) as characteristics of the target patient type (denoted as feature contribution) can be determined.

[0115] Step S25 includes:

[0116] Calculate the first average of the item similarity between any two patients in the target patient and each reference patient.

[0117] Determine the first standard deviation between the first means, and perform negative correlation normalization on the first standard deviation to obtain the feature contribution of each information dimension in each stage of diagnosis and treatment.

[0118] As an example, based on the target patient treatment type The diagnosis and treatment stage Information Dimensions For example, feature contribution The calculation method can be:

[0119]

[0120] In the formula, Indicates target patients The Treatment stage and reference patients The diagnosis and treatment stage (with the first) (Corresponding to the diagnosis and treatment stage) in the information dimension Project similarity; This indicates the information dimension between any two patients. The standard deviation of the average similarity of the above items, also known as the first standard deviation, can be calculated as follows: First, calculate the similarity between any two patients at the corresponding stage of diagnosis and treatment in terms of information dimension. Calculate the similarity of the items and then the standard deviation of their similarity. Used to take the absolute value; Used for negative correlation normalization.

[0121] As an example, similarly, the feature contribution of various information dimensions in different stages of treatment for the target patient can be calculated. The greater the feature contribution, the greater the likelihood that the information dimension belongs to the characteristic manifestation of the disease type in that stage of treatment.

[0122] Step S30: Select the information dimensions whose feature contribution meets the preset conditions as the focus dimensions.

[0123] As an example, in a diagnosis and treatment stage, if the feature contribution of a certain information dimension is greater than or equal to the threshold (preset to 0.5), then the feature contribution is considered to meet the preset condition, indicating that it is of great use in predicting the diagnosis and treatment trajectory of patients with this type of disease in the corresponding diagnosis and treatment stage. The corresponding information dimension is recorded as the attention dimension, and its corresponding feature contribution is the attention weight of the attention dimension.

[0124] Similarly, it is possible to obtain the various attention dimensions of the target patient at different stages of the entire treatment process, as well as their corresponding attention weights.

[0125] Step S40: When an update to the treatment process data of the target patient is detected, the updated treatment trajectory feature sequence is calculated based on the attention weight corresponding to the attention dimension.

[0126] As an example, when a target patient undergoes a new examination or treatment, their medical data needs to be updated. At this time, principal component analysis needs to be performed on the patient again. The original PAC analysis requires recalculation of all data dimensions. When the updated data dimensions are all previously unseen, it can easily lead to feature combination explosion and other problems, and also cause a lot of wasted computing power.

[0127] Therefore, this application first substitutes the attention weights (feature contribution) corresponding to each information dimension obtained in the past into the initial attention weights when calculating the consistency of the condition to update the weights. By updating the initial attention weights, the corresponding diagnosis and treatment trajectory feature sequences are then updated.

[0128] Before step S40, the following are included:

[0129] Principal component analysis was used to decentralize the treatment process data for multiple patients in the target patient treatment type.

[0130] Based on the feature vectors corresponding to each dimension of interest, a new feature space is constructed after dimensionality reduction.

[0131] As an example, when constructing a new feature space, the data of numerous patients within the target patient type is first digitized. Then, the correlation (covariance) between the dimensions of interest is calculated. Afterward, the eigenvalues ​​and eigenvectors of the covariance matrix are solved and sorted according to the magnitude of the eigenvalues, with the top eigenvalues ​​being selected. The eigenvectors corresponding to the largest eigenvalues ​​are arranged in the form of a data column to form a transformation matrix, which constitutes a new feature space after dimensionality reduction.

[0132] The feature descriptors within each dimension of interest are projected into a new feature space to obtain the first feature vector of the target patient at the current stage of diagnosis and treatment, and the first feature vector is marked as the feature of the diagnosis and treatment stage.

[0133] The treatment stage characteristics of each patient in each treatment stage within the target patient treatment type are arranged in chronological order to obtain the treatment stage characteristic sequence.

[0134] As an example, the feature descriptors of the target patient in each dimension of interest at this stage of diagnosis and treatment are projected into the new feature space to obtain the feature vector of the target patient at this stage of diagnosis and treatment, which is then labeled as the features of the diagnosis and treatment stage.

[0135] Similarly, we can obtain the corresponding treatment stage characteristics of other patients in the target patient type at this treatment stage.

[0136] Similarly, we can obtain the treatment stage features of many patients (including the target patient) in the target patient type in many treatment stages, arrange them in chronological order to form a corresponding treatment stage feature sequence, which is the time sequence sequence of the treatment trajectory features of the patient.

[0137] Step S40 includes:

[0138] When an update to the treatment process data of a target patient is detected, the feature sequence of the treatment stage is updated based on the attention weight corresponding to the attention dimension, resulting in an updated treatment trajectory feature sequence.

[0139] As an example, when updating data, the feature sequence of the diagnosis and treatment stage can be updated based on the determined focus dimensions, without having to update all information dimensions. Then, the focus weights are substituted into the calculation process of the diagnosis and treatment trajectory feature sequence to obtain the updated diagnosis and treatment trajectory feature sequence.

[0140] The steps involved in updating the feature sequence of the diagnosis and treatment stages based on the attention weights corresponding to the attention dimensions, to obtain the updated diagnosis and treatment trajectory feature sequence, include:

[0141] Based on the attention weights corresponding to the attention dimensions, update the initial attention weights of each information dimension to obtain the updated weights;

[0142] The updated treatment stage feature sequence is then updated with the updated weights to obtain the updated treatment trajectory feature sequence.

[0143] As an example, this embodiment first takes the attention weights (feature contribution degrees) corresponding to each information dimension obtained in the past and substitutes them into the initial attention weights when calculating the consistency of the disease condition to update the weights. If it is an information dimension that has never appeared before, its initial attention weight is initialized to the average value of the attention weights (feature contribution degrees) corresponding to many information dimensions.

[0144] Then, the corresponding attention dimensions and corresponding attention weights are re-selected, and then PAC principal component analysis is performed to construct the treatment trajectory feature sequence for each patient, thus obtaining the updated treatment trajectory feature sequence.

[0145] Step S50: Based on the updated diagnosis and treatment trajectory feature sequence, schedule settlement resources for the target patient.

[0146] As an example, after obtaining the updated treatment trajectory feature sequence, the updated treatment trajectory feature sequence is input into the model. The model predicts the treatment trajectory of the target patient in the next time period. When the corresponding treatment trajectory is determined, settlement resources are scheduled for the target patient.

[0147] This application provides an intelligent settlement scheduling method based on patient trajectory prediction. This method acquires treatment process data of the target patient and other reference patients, divides the treatment process data into multiple treatment stages, and calculates the feature contribution of various information dimensions in each treatment stage based on the treatment similarity between the target patient and each reference patient. During the patient's treatment process, there are various examination items, and each information dimension corresponds to the data type of the examination items in the treatment process, thus determining the importance of each examination item in the treatment process. When the feature contribution meets preset conditions, a focus dimension is selected from each information dimension. When an update to the target patient's treatment process data is detected, the updated treatment trajectory feature sequence is calculated based on the focus dimension and its corresponding focus weight. Then, settlement resources are scheduled for the target patient using the updated treatment trajectory feature sequence. When the target patient updates its examination data or treatment status, the treatment trajectory feature sequence of the target patient can be updated based on the calculated focus dimension and focus weight, thus eliminating the need to recalculate the data for all information dimensions, reducing the computational load of the treatment process data, and thereby quickly and accurately predicting the patient's treatment trajectory and completing the scheduling of the patient's settlement plan.

[0148] Furthermore, referring to Figure 2 Based on the first embodiment of this application, another embodiment of this application is provided. In this embodiment, the step S50 of scheduling settlement resources for the target patient based on the updated diagnosis and treatment trajectory feature sequence includes:

[0149] Step S51: Input the updated diagnosis and treatment trajectory feature sequence into the LSTM model. Based on the LSTM model, perform prediction processing on the updated diagnosis and treatment trajectory feature sequence to obtain the diagnosis and treatment trajectory corresponding to the target patient in the next time period.

[0150] As an example, the LSTM model is a long short-term memory neural network model. After obtaining the diagnostic trajectory feature sequences corresponding to the target patient and reference patients with similar diseases, the diagnostic trajectory feature sequences corresponding to the reference patients are used as training data. The model's dataset is constructed by using two adjacent sliding windows of synchronous length to form input-output pairs, and the model is trained to obtain the treatment trajectory of the target patient type. Then, the updated treatment trajectory feature sequence corresponding to the treatment trajectory already experienced by the target patient is used as... The model takes input and predicts the treatment trajectory of the target patient in the next time period.

[0151] Step S52: Based on the treatment trajectory, the settlement resources for the target patients are scheduled using a multi-objective optimization algorithm.

[0152] As an example, after predicting the treatment trajectory of the target patient, the patient's settlement resources are scheduled by combining a multi-objective optimization algorithm with pre-defined prediction execution logic. The settlement resource scheduling results are shown below:

[0153]

[0154] In this embodiment, the updated treatment trajectory feature sequence is processed by an LSTM model to quickly and accurately predict the treatment trajectory of the target patient in the next time period, and finally complete the scheduling of the patient settlement plan.

[0155] Reference Figure 3 , Figure 3 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application.

[0156] like Figure 3 As shown, the intelligent billing and scheduling device based on patient trajectory prediction may include: a processor 1001, a memory 1003, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1003.

[0157] Optionally, the intelligent billing and scheduling device based on patient trajectory prediction may also include a user interface, a network interface, a camera, RF (Radio Frequency) circuitry, sensors, a WiFi module, etc. The user interface may include a display screen and an input submodule such as a keyboard; optional user interfaces may also include standard wired or wireless interfaces. The network interface may include standard wired or wireless interfaces (such as a Wi-Fi interface).

[0158] Those skilled in the art will understand that Figure 3 The structure of the intelligent billing and scheduling device based on patient trajectory prediction shown in the figure does not constitute a limitation on the intelligent billing and scheduling device based on patient trajectory prediction. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0159] like Figure 3 As shown, the memory 1003, serving as a storage medium, may include an operating system, a network communication module, and a patient trajectory prediction-based intelligent billing and scheduling program. The operating system is a program that manages and controls the hardware and software resources of the patient trajectory prediction-based intelligent billing and scheduling device, supporting the operation of the patient trajectory prediction-based intelligent billing and scheduling program and other software and / or programs. The network communication module is used to enable communication between the various components within the memory 1003, as well as communication with other hardware and software in the patient trajectory prediction-based intelligent billing and scheduling system.

[0160] exist Figure 3 In the intelligent billing and scheduling device based on patient trajectory prediction shown, the processor 1001 is used to execute the intelligent billing and scheduling program based on patient trajectory prediction stored in the memory 1003 to implement the steps of the intelligent billing and scheduling method based on patient trajectory prediction described above.

[0161] The specific implementation method of the intelligent settlement scheduling device based on patient trajectory prediction in this application is basically the same as the embodiments of the intelligent settlement scheduling method based on patient trajectory prediction described above, and will not be repeated here.

[0162] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0163] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0164] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0165] The above are merely preferred embodiments of this application and do not limit the scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the scope of protection of this application.

[0166] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0167] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An intelligent billing and scheduling method based on patient trajectory prediction, characterized in that, The method includes: Acquire treatment process data for the target patient and other reference patients, and divide the treatment process data into multiple treatment stages; Based on the similarity of diagnosis and treatment between the target patient and each of the reference patients at each stage of diagnosis and treatment, the feature contribution of each information dimension in each stage of diagnosis and treatment is calculated. The information dimensions correspond to the data types of the examination items during the diagnosis and treatment process, specifically including: Feature extraction is performed on the examination results corresponding to the diagnosis and treatment process data at each stage of diagnosis and treatment to obtain feature vectors of different information dimensions, and the feature vectors are marked as feature descriptors; The first feature descriptor corresponding to the multiple examination results of the target patient in each information dimension and the second feature descriptor corresponding to the reference patient are obtained respectively. Determine the maximum cosine similarity between the first feature descriptor and the second feature descriptor; The average of the maximum similarity values ​​of the examination items corresponding to each information dimension is accumulated to obtain the item similarity between the target patient and the reference patient in each information dimension. Determine the number of examinations performed by the target patient or reference patient at each stage of diagnosis and treatment, and calculate the first product between the similarity of the examination items and the initial attention weight corresponding to the information dimension. Based on the first product, the number of examination items, and the initial attention weight, the consistency of the condition between the target patient and the reference patient at the same stage of diagnosis and treatment is calculated. The reciprocal of the consistency of the disease condition is used as the sequence distance between the sequences corresponding to each of the feature descriptors; Based on the sequence distance and the preset similarity algorithm, the stage similarity between the target patient and the reference patient at each stage of diagnosis and treatment is calculated. The maximum value among the stage similarities at each stage of diagnosis and treatment is set as the similarity of the diagnosis and treatment trajectories of the target patient and the reference patient during the diagnosis and treatment process; The data space is constructed by taking the diagnosis and treatment process data of the target patient or the reference patient as data points and each examination item as a data dimension. The feature distance between each data point is determined based on the reciprocal of the similarity of the treatment trajectory. Based on the preset clustering algorithm and the feature distance, the data points in the data space are clustered to obtain multiple clusters. The treatment type corresponding to the cluster to which the target patient belongs is taken as the treatment type of the target patient; Calculate the first average of the item similarity between any two patients in the target patient and each of the reference patients; Determine the first standard deviation between the first average values, and perform negative correlation normalization on the first standard deviation to obtain the feature contribution of each information dimension in each stage of diagnosis and treatment. The information dimensions whose feature contribution meets the preset conditions are taken as the focus dimensions; When an update to the treatment process data of the target patient is detected, the updated treatment trajectory feature sequence is calculated based on the attention weight corresponding to the attention dimension. Based on the updated diagnosis and treatment trajectory feature sequence, settlement resources are allocated for target patients.

2. The intelligent billing and scheduling method based on patient trajectory prediction as described in claim 1, characterized in that, Before calculating the updated treatment trajectory feature sequence based on the attention weight corresponding to the attention dimension when the treatment process data of the target patient is updated, the method further includes: Principal component analysis was used to decentralize the treatment process data for multiple patients in the target patient treatment type. Based on the feature vectors corresponding to each dimension of interest, a new feature space is constructed after dimensionality reduction. The feature descriptors within each of the aforementioned dimensions of interest are projected into the new feature space to obtain the first feature vector of the target patient at the current stage of diagnosis and treatment, and the first feature vector is marked as the feature of the diagnosis and treatment stage. The treatment stage characteristics of each patient in each treatment stage of the target patient treatment type are arranged in chronological order to obtain the treatment stage characteristic sequence.

3. The intelligent billing and scheduling method based on patient trajectory prediction as described in claim 2, characterized in that, When an update to the treatment process data of the target patient is detected, the updated treatment trajectory feature sequence is calculated based on the attention weight corresponding to the attention dimension, including: When an update to the treatment process data of the target patient is detected, the treatment stage feature sequence is updated based on the attention weight corresponding to the attention dimension, resulting in an updated treatment trajectory feature sequence.

4. The intelligent billing and scheduling method based on patient trajectory prediction as described in claim 3, characterized in that, The step of updating the treatment stage feature sequence based on the attention weights corresponding to the attention dimensions to obtain the updated treatment trajectory feature sequence includes: Based on the attention weights corresponding to the attention dimensions, the initial attention weights of each information dimension are updated to obtain the updated weights. The updated treatment stage feature sequence is then updated using the updated weights to obtain the updated treatment trajectory feature sequence.

5. The intelligent billing and scheduling method based on patient trajectory prediction as described in claim 1, characterized in that, The process of allocating settlement resources for target patients based on the updated diagnosis and treatment trajectory feature sequence includes: The updated treatment trajectory feature sequence is input into the LSTM model. Based on the LSTM model, the updated treatment trajectory feature sequence is predicted to obtain the treatment trajectory corresponding to the target patient in the next time period. Based on the aforementioned treatment trajectory, a multi-objective optimization algorithm is used to schedule settlement resources for the target patients.

Images