Abnormality detection method and device for treatment scheme

Abstract

The invention provides a treatment scheme anomaly detection method which can be applied to the technical field of artificial intelligence and medical treatment. The method comprises the following steps: inputting a treatment scheme into a first feature extraction module in an anomaly detection model to obtain an anomaly classification feature sequence comprising at least one anomaly classification feature sorted according to time information; classifying the treatment schemes by using the abnormal classification feature sequence to obtain a classification result; under the condition that the classification result represents that the treatment scheme is abnormal, inputting the treatment scheme into a second feature extraction module in an anomaly detection model to obtain a transaction feature sequence comprising at least one transaction feature sorted according to the time information; the transaction feature sequence is input into a transaction prediction module in an anomaly detection model, normal transaction prediction information is obtained, and the transaction prediction module is obtained by training a normal treatment scheme in historical treatment schemes; and detecting treatment transaction information in the treatment scheme by using the normal transaction prediction information to obtain a detection result.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of artificial intelligence and medical technology, in particular to an abnormality detection method and device of a treatment plan. BACKGROUND

[0002] Electronic prescription cost management is a core link in the management of medical institutions. It has a profound impact on cost control, patient satisfaction and the rational allocation of medical resources of medical institutions.

[0003] However, in the electronic prescription cost management, excessive reliance on manual review, simple rule checking or statistical tools, such method is easily affected by subjective factors and low in efficiency, leading to the problem that abnormal costs are difficult to identify in time. SUMMARY

[0004] In view of the above problems, the present disclosure provides an abnormality detection method of a treatment plan, an abnormality detection device of medical information, equipment, a medium and a program product.

[0005] According to a first aspect of the present disclosure, an abnormality detection method of a treatment plan is provided, comprising: inputting a treatment plan into a first feature extraction module of an abnormality detection model to obtain an abnormality classification feature sequence including at least one abnormality classification feature sorted according to time information; classifying the treatment plan using the abnormality classification feature sequence to obtain a classification result; in the case that the classification result represents that the treatment plan has an abnormality, inputting the treatment plan into a second feature extraction module of the abnormality detection model to obtain a transaction feature sequence including at least one transaction feature sorted according to time information; inputting the transaction feature sequence into a transaction prediction module of the abnormality detection model to obtain normal transaction prediction information, the transaction prediction module being trained using normal treatment plans in historical treatment plans; and detecting treatment transaction information in the treatment plan using the normal transaction prediction information to obtain a detection result.

[0006] The second aspect of the present disclosure provides an abnormality detection device of medical information, comprising: a first input module configured to input a treatment plan into a first feature extraction module of an abnormality detection model to obtain an abnormality classification feature sequence comprising at least one abnormality classification feature sorted according to time information; a classification module configured to classify the treatment plan using the abnormality classification feature sequence to obtain a classification result; a second input module configured to, in a case where the classification result indicates that the treatment plan is abnormal, input the treatment plan into a second feature extraction module of the abnormality detection model to obtain a transaction feature sequence comprising at least one transaction feature sorted according to time information; a third input module configured to input the transaction feature sequence into a transaction prediction module of the abnormality detection model to obtain normal transaction prediction information, the transaction prediction module being trained using normal treatment plans in historical treatment plans; and a detection module configured to detect treatment transaction information in the treatment plan using the normal transaction prediction information to obtain a detection result.

[0007] The third aspect of the present disclosure provides an electronic device, comprising: one or more processors; a memory configured to store one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method.

[0008] The fourth aspect of the present disclosure further provides a computer-readable storage medium having stored thereon a computer program or instructions, wherein the computer program or instructions, when executed by a processor, implement the steps of the method.

[0009] The fifth aspect of the present disclosure further provides a computer program product comprising a computer program or instructions, wherein the computer program or instructions, when executed by a processor, implement the steps of the method.

[0010] According to an embodiment of the present disclosure, by inputting the treatment scheme into the first feature extraction module in the anomaly detection model, an anomaly classification feature sequence including at least one anomaly classification feature sorted according to time information is obtained; the treatment scheme is classified by using the anomaly classification feature sequence to obtain a classification result, so that the classification result is obtained by comprehensively considering the change of the treatment scheme with time information, and the hidden relationship between various information in the treatment scheme can be captured, and the accuracy of anomaly detection can be improved. In the case where the classification result indicates that the treatment scheme has an anomaly, the treatment scheme is input into the second feature extraction module in the anomaly detection model to obtain a transaction feature sequence including at least one transaction feature sorted according to time information; in order to accurately capture the abnormal transaction data, the transaction feature sequence can be input into the transaction prediction module in the anomaly detection model to obtain normal transaction prediction information, and since the transaction prediction module is trained by using the normal treatment scheme in the historical treatment scheme, the normal transaction prediction information is based on the treatment scheme to predict normal transaction data, but not transaction data in the future time; the treatment transaction information in the treatment scheme is detected by using the normal transaction prediction information to obtain a detection result, and intelligent detection of the treatment scheme is realized, manual review is reduced, and the efficiency of anomaly detection is improved. BRIEF DESCRIPTION OF DRAWINGS

[0011] The above and other objects, features and advantages of the present disclosure will become more apparent from the following description when taken in conjunction with the accompanying drawings, in which:

[0012] Figure 1 An application scenario diagram of a treatment scheme anomaly detection method, a medical information anomaly detection apparatus, a device, a medium and a program product according to an embodiment of the present disclosure is schematically shown;

[0013] Figure 2 A flowchart of a treatment scheme anomaly detection method according to an embodiment of the present disclosure is schematically shown;

[0014] Figure 3 A structure diagram of an anomaly detection model according to an embodiment of the present disclosure is schematically shown;

[0015] Figure 4 A process of training the first feature extraction module and the second feature extraction module according to an embodiment of the present disclosure is schematically shown;

[0016] Figure 5 A process of anomaly detection of a treatment scheme according to an embodiment of the present disclosure is schematically shown;

[0017] Figure 6 A structure block diagram of a medical information anomaly detection apparatus according to an embodiment of the present disclosure is schematically shown; and

[0018] Figure 7A block diagram of an electronic device suitable for implementing an abnormality detection method of a treatment plan according to an embodiment of the disclosure is schematically shown. DETAILED DESCRIPTION

[0019] Hereinafter, embodiments of the disclosure will be described with reference to the accompanying drawings. However, it is to be understood that these descriptions are merely exemplary and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of embodiments of the disclosure. However, it will be apparent to one skilled in the art that one or more embodiments can be practiced without these specific details. In other instances, well-known structures and functions have been omitted to avoid unnecessarily complicating the disclosure with details that will be readily understood by those skilled in the art.

[0020] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the disclosure. As used herein, the term "including" "comprising" and the like are meant to be inclusive, but not limiting to the components, steps, operations and / or features that were listed. The use of "including" "comprising" and "having" also modifies the term "comprises" to include a combination of elements or ingredients unless specified otherwise.

[0021] All terms used herein, including technical and scientific terms, have the same meanings as those generally understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having meanings that are consistent with the context of the specification, and should not be interpreted in an idealized or overly formal manner.

[0022] In the case of using expressions similar to "at least one of A, B, and C, etc.", it should generally be interpreted to include any of them alone, any combination of two or more of them, etc. unless otherwise specified.

[0023] In the technical solutions of the disclosure, the user information (including but not limited to user personal information, user image information, user health information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved are information and data authorized by the user or authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of related data comply with relevant laws, regulations and standards, necessary security measures are taken, do not violate public order and good customs, and provide corresponding operation portal for user to choose authorization or refusal.

[0024] In the management of electronic prescriptions, at present, the historical data accumulated in the process of electronic prescriptions has not been fully tapped, and the influence of external factors such as patient characteristics and drug price fluctuations on costs has not been fully considered. At the same time, the existing methods still rely on certain artificial experience in the judgment process, which leads to the subjectivity and limitations of the judgment results. Therefore, how to fully utilize historical data and experience under the premise of reducing subjective interference as much as possible, and build an advanced technology that can overcome the limitations of traditional methods and adapt to the needs of medical institutions in the new era to improve the intelligent level of electronic prescription cost management has become a problem to be solved.

[0025] Therefore, the embodiments of the present disclosure provide an abnormality detection method of a treatment plan, which can be applied to the fields of artificial intelligence and medical technology. The method comprises: inputting the treatment plan into a first feature extraction module in an abnormality detection model to obtain an abnormality classification feature sequence comprising at least one abnormality classification feature sorted according to time information; classifying the treatment plan by using the abnormality classification feature sequence to obtain a classification result; in the case that the classification result represents that the treatment plan is abnormal, inputting the treatment plan into a second feature extraction module in the abnormality detection model to obtain a transaction feature sequence comprising at least one transaction feature sorted according to time information; inputting the transaction feature sequence into a transaction prediction module in the abnormality detection model to obtain normal transaction prediction information, the transaction prediction module being trained by using normal treatment plans in historical treatment plans; and detecting treatment transaction information in the treatment plan by using the normal transaction prediction information to obtain a detection result.

[0026] Figure 1 An application scenario diagram of the abnormality detection method of a treatment plan, the abnormality detection device of medical information, the equipment, the medium and the program product according to the embodiments of the present disclosure is schematically shown.

[0027] As Figure 1 shown, the application scenario 100 according to the embodiments can include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104 and a server 105. The network 104 is a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103 and the server 105. The network 104 can include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0028] The user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0029] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smartphones, tablet computers, laptop computers, desktop computers, etc.

[0030] The server 105 can be a server providing various services, such as a background management server supporting websites browsed by the user using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (only as examples). The background management server can analyze and process received user requests, etc., and feed back the processing results (such as web pages, information, or data, etc. obtained or generated according to user requests) to the terminal device.

[0031] It should be noted that the treatment scheme anomaly detection method provided by the embodiments of the present disclosure can generally be executed by the server 105. Accordingly, the medical information anomaly detection apparatus provided by the embodiments of the present disclosure can generally be arranged in the server 105. The treatment scheme anomaly detection method provided by the embodiments of the present disclosure can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105. Accordingly, the medical information anomaly detection apparatus provided by the embodiments of the present disclosure can also be arranged in a server or a server cluster different from the server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or the server 105.

[0032] It should be understood that Figure 1 The number of terminal devices, networks, and servers in the above description is only illustrative. According to the needs of implementation, there can be any number of terminal devices, networks, and servers.

[0033] Figure 2 A flowchart of a treatment scheme anomaly detection method according to an embodiment of the present disclosure is schematically shown.

[0034] As Figure 2 shown, the treatment scheme anomaly detection method of this embodiment includes operations S210-S250.

[0035] In operation S210, the first feature extraction module in the abnormality detection model is input with the treatment scheme, to obtain an abnormality classification feature sequence including at least one abnormality classification feature sorted according to time information.

[0036] In an embodiment of the present disclosure, the treatment scheme can be an electronic prescription, and the abnormality classification feature can represent that there is an abnormality in a drug or a transaction value of the treatment scheme. For example, the abnormality classification feature can be that a patient continuously prescribes the same high-priced drug for two consecutive days, a doctor prescribes multiple drugs with conflicting indications for the same patient, or the same patient continuously prescribes high-priced drugs through different hospitals for multiple days, etc. The at least one abnormality classification feature in the treatment scheme is sorted according to time, to obtain the abnormality classification feature sequence.

[0037] In an embodiment of the present disclosure, the first feature extraction module can be constructed based on a feature extraction neural network, and the first feature extraction module can be used to extract features of abnormal drugs or fees in the treatment scheme. For example, the first feature extraction module can include multiple convolution layers.

[0038] In operation S220, the treatment scheme is classified using the abnormality classification feature sequence, to obtain a classification result.

[0039] In an embodiment of the present disclosure, the classification result can be classified into an abnormality classification result representing that there is an abnormality in the treatment scheme and a normal classification result representing that there is no abnormality in the treatment scheme. The abnormality classification feature can be used to determine whether there is an abnormality in the treatment scheme.

[0040] For example, the classification module in the prediction model can be used to classify the treatment scheme using the abnormality classification feature sequence. The classification module can be a binary classification-based classification sub-model.

[0041] In operation S230, in a case where the classification result represents that there is an abnormality in the treatment scheme, the second feature extraction module in the abnormality detection model is input with the treatment scheme, to obtain a transaction feature sequence including at least one transaction feature sorted according to time information.

[0042] In an embodiment of the present disclosure, the transaction feature can be a feature that a transaction amount of at least one drug in the treatment scheme is a normal amount. The transaction amount can be the overall fee of the treatment scheme, or the fee of each drug. The transaction feature is sorted according to time information, to obtain the transaction feature sequence.

[0043] In an embodiment of the present disclosure, the second feature extraction module can be constructed based on a feature extraction neural network, and the second feature extraction module can be used to extract features of normal drug transaction values in the treatment scheme.

[0044] In operation S240, the transaction feature sequence is input into a transaction prediction module in the anomaly detection model to obtain normal transaction prediction information, and the transaction prediction module is trained by using normal treatment schemes in historical treatment schemes.

[0045] In an embodiment of the present disclosure, the normal transaction prediction information can be a predicted value of a normal transaction amount of the medicine. The normal treatment scheme can be a historical transaction value of the medicine transaction value corresponding to the same disease and the same type of medicine. The normal transaction value of the corresponding treatment scheme is predicted by using the transaction feature sequence of the treatment scheme.

[0046] In operation S250, the treatment transaction information in the treatment scheme is detected by using the normal transaction prediction information to obtain a detection result.

[0047] In an embodiment of the present disclosure, the treatment transaction information can be an actual transaction value of the medicine in the treatment scheme. The actual transaction value is detected according to the predicted normal transaction value of the medicine to obtain the detection result. The detection result includes that the treatment scheme has an abnormal transaction value and that the treatment scheme does not have an abnormal transaction value. The existence of the abnormal transaction value indicates that the transaction value of at least one medicine in the treatment scheme is abnormal. In the case of the existence of the abnormal transaction value, the detection result further includes the name of the medicine with the abnormal transaction value and the actual transaction value of the medicine.

[0048] According to an embodiment of the present disclosure, by inputting the treatment scheme into the first feature extraction module in the anomaly detection model, an anomaly classification feature sequence including at least one anomaly classification feature sorted according to time information is obtained. The treatment scheme is classified by using the anomaly classification feature sequence to obtain a classification result. Therefore, the classification result is obtained by comprehensively considering the change of the treatment scheme with time information, and the hidden relationship between various information in the treatment scheme can be captured, which can improve the accuracy of anomaly detection. In the case where the classification result indicates that the treatment scheme has an abnormality, the treatment scheme is input into the second feature extraction module in the anomaly detection model to obtain a transaction feature sequence including at least one transaction feature sorted according to time information. In order to accurately capture the abnormal transaction data, the transaction feature sequence can be input into the transaction prediction module in the anomaly detection model to obtain normal transaction prediction information. Since the transaction prediction module is trained by using normal treatment schemes in historical treatment schemes, the normal transaction prediction information is predicted based on the normal transaction data of the treatment scheme, rather than the transaction data of the future time. The treatment transaction information in the treatment scheme is detected by using the normal transaction prediction information to obtain a detection result, which realizes intelligent detection of the treatment scheme, reduces manual review, and improves the efficiency of anomaly detection.

[0049] According to an embodiment of the present disclosure, the historical treatment scheme further includes historical patient information, historical doctor information, historical transaction information, and historical treatment medicine information.

[0050] In embodiments of the present disclosure, the historical patient information can include patient medical history, card number or other patient identification information, etc. The historical doctor information can include the doctor's specialty, the average transaction value of the electronic prescriptions issued by the doctor and the average transaction value of the doctor for each drug, etc. The historical transaction information can include the actual transaction value of each electronic prescription, the actual transaction value of each drug, the treatment cost corresponding to each disease, etc. The historical treatment drug information can include the prescriptions issued corresponding to each disease.

[0051] In embodiments of the present disclosure, the time information can be sorted by day, and the treatment plan can be represented by the electronic prescription number. For example, the data sequence of the historical treatment drug information is , is the number of recorded days, is the electronic prescription number, represents the drug information on the prescription in the first day; the data sequence of the historical patient information is , is the number of recorded days, is the electronic prescription number, represents the patient information on the prescription in the first day; the data sequence of the historical doctor information is , is the number of recorded days, is the electronic prescription number, represents the doctor information on the prescription in the first day; the data sequence of the historical transaction information is , is the number of recorded days, is the electronic prescription number, represents the cost data on the prescription in the first day.

[0052] In embodiments of the present disclosure, for example, the historical patient information, the historical doctor information, the historical transaction information and the historical treatment drug information can also be integrated to form a multi-dimensional electronic prescription data set , the composition of which is , represents the electronic prescription data and cost on the prescription in the first day, and specifically includes , where indicates whether there is a cost anomaly on the prescription in the first day, indicates whether there is a cost anomaly on the prescription in the first Possible abnormal amount of expenses.

[0053] In embodiments of the present disclosure, for example, the multi-dimensional electronic prescription data set is normalized to reduce the influence of different feature value ranges on algorithm performance, and the normalized data is .

[0054] According to embodiments of the present disclosure, the historical patient information, historical doctor information, historical transaction information and historical treatment drug information of the two adjacent days (or multiple consecutive days) are integrated, the hidden relationship between various information in the treatment scheme can be captured, and the accuracy of the anomaly detection can be improved. For example, some abnormal behaviors (such as “decomposition of prescription” and “repeated prescription”) can be identified only through data of consecutive time; the same high-priced drug is prescribed by a patient on the tth day and the t+1th day (the amount is not exceeded on a single day, but the amount is exceeded on two days); a doctor prescribes multiple drugs with conflicting indications (such as antibiotics + antiviral drugs) for the same patient on two adjacent days; the single-day data may have a temporary peak in expenses due to a reasonable reason (such as emergency treatment or special treatment), resulting in false positives. Fraud chains and collaborative behaviors can also be identified, for example, the same doctor prescribes the same high-priced drug for different patients on two adjacent days; the same patient uses multiple hospitals to prescribe drugs on multiple consecutive days.

[0055] According to embodiments of the present disclosure, the first feature extraction module is trained based on the following manner: according to the time information, the historical treatment schemes are processed to obtain treatment time sequence features; the treatment time sequence features are input into the first feature extraction module to obtain abnormal classification training features; the first feature extraction module is trained using the abnormal classification training features until the classification error value between the training classification result and the true classification result satisfies a preset condition, and the training classification result is obtained by classifying the historical treatment schemes using the abnormal classification training features.

[0056] According to embodiments of the present disclosure, the treatment time sequence features can represent the changes of the historical patient information, historical doctor information, historical transaction information and historical treatment drug information with the time information in the process of treating each disease.

[0057] In embodiments of the present disclosure, for example, the first feature extraction module can be trained by: extracting the treatment time sequence features in the historical treatment schemes using the feature extraction layer; inputting the treatment time sequence features into the first feature extraction module to extract abnormal classification features; obtaining a probability value P by classification using a plurality of fully connected layers; the probability value P ∈ [0, 1], a threshold value (for example, 0.5) can be set to determine whether it is abnormal, and a training result is obtained. The classification error value between the training result and the true classification result is calculated, and the preset condition is satisfied, for example, the training is stopped when the classification error value is less than 0.3.

[0058] According to an embodiment of the present disclosure, the second feature extraction module is trained by using associated information of historical treatment transaction information in historical treatment schemes, the associated information including same-type drug transaction information, same-type disease treatment transaction information, preset drug transaction rules and historical treatment transaction information of the same patient.

[0059] In an embodiment of the present disclosure, the historical treatment transaction information can be transaction values of at least one drug in each historical treatment scheme, or can be transaction values of at least one drug in all historical treatment schemes. The same-type drug transaction information can be a single transaction value, a multiple transaction value or an overall transaction value of other drugs of the same type as each drug in the historical treatment scheme, for example, if aspirin is included in the treatment scheme, the same-type drug transaction information can include a single transaction value of ibuprofen, an overall transaction value of indomethacin, etc. The same-type disease treatment transaction information can be a transaction value of at least one drug in other treatment schemes for the disease suffered by the patient, or a transaction value of at least one drug in treatment schemes for other diseases of the same type as the disease suffered by the patient, or a transaction value of a drug accounting for a certain percentile of the same-type disease treatment transaction information, etc. For example, if the drug in the treatment scheme is for treating tendonitis of the patient, the same-type disease treatment transaction information can include a transaction value of a drug in a treatment scheme for muscle strain. The preset drug transaction rule can be a policy, for example, an increase in drug reimbursement ratio, a maximum value of a specified number of drugs purchased per month, etc. The historical treatment transaction information of the same patient can be the treatment cost spent by the same patient for the same disease within a specified period of time in the past, or the treatment cost spent by the same patient for different diseases within a specified period of time in the past, etc., wherein the specified period of time can be one month, one year, etc., and can be adjusted according to actual conditions.

[0060] According to an embodiment of the present disclosure, by training the second feature extraction module by using the same-type drug transaction information, the same-type disease treatment transaction information, the preset drug transaction rule and the historical treatment transaction information of the same patient, the second feature extraction module can learn the association between the patient, the disease, the drug and the cost, and improve the optimization effect of training on the second feature extraction module.

[0061] According to an embodiment of the present disclosure, the second feature extraction module is trained based on the following manner: extracting features from historical treatment transaction information according to homologous drug transaction information, homologous disease treatment transaction information, preset drug transaction rules and historical treatment transaction information of the same patient to obtain drug transaction features; determining a doctor dosage threshold of each of the plurality of drugs according to a historical cumulative dosage of each of the plurality of drugs by the doctor; performing vectorization processing on a reference dosage of each of the plurality of drugs and the doctor dosage threshold to obtain drug dosage features; splicing the drug transaction features and the drug dosage features to obtain prediction transaction training features; and training the second feature extraction module by using the prediction transaction training features.

[0062] In an embodiment of the present disclosure, the drug transaction features can represent transaction values of each drug for different diseases of different patients, and can also represent transaction values of a combination of a plurality of drugs for corresponding diseases. The historical cumulative dosage can represent the dosage that has been prescribed by the doctor for each drug. The doctor dosage threshold can be how many dosages the doctor can still prescribe for each drug, for example, it can be per time, or it can be the total dosage within a specified time. The reference dosage of each of the plurality of drugs can be the maximum dosage that the doctor can prescribe per time for each drug, or the maximum dosage that the doctor can prescribe within a specified time for each drug. The drug dosage features can represent the dosage that the doctor prescribes per time for each drug, or the dosage that the doctor prescribes within a specified time for each drug. The prediction transaction training features can represent that the transaction amount of at least one drug in the historical treatment scheme is a normal amount.

[0063] In an embodiment of the present disclosure, for example, the patient has a headache in the historical treatment scheme, and aspirin is prescribed. According to the transaction value of aspirin in the historical treatment scheme, the single transaction value of the drug of the same type as aspirin, the transaction value of other drugs for treating headache such as amcaine tablets, and the treatment cost of the patient for treating headache in the past three months, the transaction value feature of aspirin for the patient and other patients with headache is determined. According to the dosage of aspirin that the doctor has currently prescribed, the dosage of aspirin that the doctor can still prescribe is determined. According to the maximum dosage of aspirin that can be prescribed and the dosage of aspirin that the doctor can still prescribe, the prediction transaction training features are determined.

[0064] According to an embodiment of the present disclosure, determining the prediction transaction training features by the doctor dosage threshold and the reference dosage of each of the plurality of drugs can enable the second feature extraction module to learn the difference in the dosage prescribed by each doctor and the maximum value of the dosage that each drug can be prescribed in the training process, so that the second feature extraction module is more accurate when performing transaction feature extraction.

[0065] According to an embodiment of the present disclosure, the historical treatment transaction information is feature extracted according to the same drug transaction information, the same disease treatment transaction information, the preset drug transaction rule and the historical treatment transaction information of the same patient to obtain drug transaction features, including: determining a drug transaction range of a plurality of drugs in the historical treatment transaction information according to the preset drug transaction rule and the same drug transaction information; determining a treatment transaction range of the historical treatment transaction according to the same disease treatment transaction information and the historical treatment transaction information of the same patient; and performing vectorization processing on the drug transaction value and the drug transaction range of each of the plurality of drugs, the historical treatment transaction information of the same patient and the treatment transaction range to obtain the drug transaction features.

[0066] In an embodiment of the present disclosure, the drug transaction range can be a drug standard transaction value range, and the transaction value of each drug should theoretically be within the drug transaction range. The treatment transaction range can be a treatment cost range of the same type of disease.

[0067] In an embodiment of the present disclosure, for example, in a specified region, the medical insurance reimbursement ratio of aspirin is 80%, and the transaction value of 100 yuan of aspirin is 20 yuan. The transaction value of the drugs similar to aspirin is between 30 and 40 yuan, so the drug transaction range of aspirin is 20 to 40 yuan. The treatment cost range of the diseases of the same type as headache is 30-70 yuan, and the treatment cost spent by the patient with headache in the past three months for headache is 80 yuan, so the treatment transaction range of the diseases of the same type as headache is 30-80 yuan.

[0068] According to an embodiment of the present disclosure, by determining the standard transaction value range of the drug and the treatment cost range of the disease of the same type, it can be determined whether the treatment cost spent by the patient is within the standard transaction range. If yes, it can be determined that the treatment scheme has no abnormal transaction value, and if not, it can be determined that the treatment scheme has an abnormal transaction value.

[0069] According to an embodiment of the present disclosure, the preset condition includes that a weighted sum value between the fitting transaction deviation and the classification error value reaches a target value.

[0070] In an embodiment of the present disclosure, the weighted sum value Loss can be calculated by the fitting transaction deviation Loss 拟合 and the classification error value Loss 分类

[0071]

[0072] Wherein, α is the weight of the classification error value, and β is the weight of the fitting transaction deviation.

[0073] In an embodiment of the present disclosure, the target value can be preset in advance, for example, it can be 0.3, or it can be adjusted according to the actual situation.​

[0074] Figure 3 A structural diagram of an anomaly detection model according to an embodiment of the present disclosure is schematically shown.

[0075] As shown in Figure 3 , the anomaly detection model includes a first feature extraction module 310, a classification module 320, a second feature extraction module 330, a transaction prediction module 340, and a detection module 350. The first feature extraction module 310 can extract anomaly classification features and sort the anomaly classification features according to time information to obtain an anomaly column feature sequence, which is input to the classification module 320. The classification module 320 classifies the treatment plan using the anomaly column feature sequence to obtain a classification result including a case where the treatment plan is abnormal and a case where the treatment plan is not abnormal. In the case where the classification result represents that the treatment plan is abnormal, the treatment plan is input to the second feature extraction module 330 to extract transaction features, and the transaction features are sorted according to time to obtain a transaction feature sequence, which is input to the transaction prediction module 340 to obtain normal transaction prediction information and input to the detection module 350 to obtain a detection result.

[0076] In an embodiment of the present disclosure, for example, time sequence features can be extracted by a multi-layer long short-term memory neural network (LSTM), to mine features in electronic prescriptions with time sequence correlation. For example, the process can be as follows: the normalized electronic prescription data is input into a multi-layer LSTM network, and the LSTM dynamically adjusts the memory cell state through a gating mechanism (forget gate, input gate, and output gate) to retain long-term dependencies. For example:

[0077] Forget gate: decides which historical information to discard (e.g., whether the patient's past medication record is related to the current treatment cost).

[0078] Input gate: updates the key features of the current day (e.g., new high-priced drugs) to the memory cell.

[0079] Output gate: generates the feature representation of the day, which implies the time sequence rule (e.g., periodic fluctuation of the cost of a certain type of drug).

[0080] Further high-order time sequence features are extracted by the multi-layer LSTM, for example, the first layer can capture short-term cost changes within a single prescription (e.g., the same drug is opened for several consecutive days). The second layer can find long-term patterns of different electronic prescriptions (e.g., a doctor tends to open a high-priced drug combination at the beginning of each month).

[0081] In embodiments of the present disclosure, the features of normal prescriptions are clustered in the latent space (e.g., the features of regular medication of chronic patients), and the features of abnormal prescriptions deviate from the cluster center (e.g., a combination of high-priced drugs suddenly appears). Through supervised learning, the model constructs a separation hyperplane in the latent space to separate the abnormal features from the normal features. For example, 90% of the prescriptions of a certain doctor are clustered in a certain area of the latent space (normal cluster), and a prescription suddenly appears far from the area (abnormal point), and the classification module will give a high abnormal probability.

[0082] According to embodiments of the present disclosure, by setting the preset condition by combining the fitting transaction deviation and the classification error value, the training result of the transaction prediction module can be more accurate, and in actual application, the accuracy of abnormal detection can be improved, manual review can be reduced, and the efficiency of abnormal detection can be improved.

[0083] According to embodiments of the present disclosure, the normal transaction prediction information is used to detect the treatment transaction information in the treatment scheme to obtain a detection result, including: determining a transaction deviation between the normal transaction prediction information and the treatment transaction information in the treatment scheme; in a case where the transaction deviation satisfies a deviation threshold condition, determining a detection result representing that there is no abnormal transaction value in the treatment transaction information; and in a case where the transaction deviation does not satisfy the deviation threshold condition, determining a detection result representing that there is an abnormal transaction value in the treatment transaction information.

[0084] In embodiments of the present disclosure, the deviation threshold condition represents a maximum deviation between the predicted normal transaction value of the treatment scheme and the actual transaction value, which can be preset in advance, for example, can be less than 0.5, or can be adjusted according to actual conditions. The deviation between the predicted normal transaction value of the treatment scheme and the actual transaction value is determined, for example, in a case where the deviation is less than 0.5, it is determined that there is no abnormal transaction value in the actual transaction value; and in a case where the deviation is greater than 0.5, it is determined that there is an abnormal transaction value in the actual transaction value.

[0085] Figure 4 The process of performing abnormal detection on a treatment scheme according to embodiments of the present disclosure is schematically shown.

[0086] As Figure 4As shown, the first feature extraction 401 obtains an abnormal classification feature sequence from the abnormal classification features, classifies the treatment scheme 402 according to the abnormal classification feature sequence, and obtains a classification result. It is determined whether there is an abnormality 403 according to the classification result. In the case of no, it is determined that the treatment scheme is normal 410. In the case of yes, the treatment scheme is input to the second feature extraction module. The second feature extraction 404 obtains a transaction feature, and obtains a transaction feature sequence. The transaction prediction 405 is performed on the transaction feature sequence, and normal transaction prediction information is obtained. The treatment transaction information is detected 406 according to the normal transaction prediction information, and a transaction deviation between the normal transaction prediction information and the treatment transaction information in the treatment scheme is obtained. According to the transaction deviation, it is determined whether the deviation threshold is met 407. In the case of meeting the deviation threshold, it is determined that there is no abnormal transaction value 409. In the case of not meeting the deviation threshold, it is determined that there is an abnormal transaction value 408.

[0087] According to an embodiment of the present disclosure, the transaction prediction module is trained based on the following manner: according to the mapping relationship between the transaction feature and the transaction value, the normal transaction feature sequence is mapped to obtain a normal transaction prediction value, the normal transaction feature sequence is obtained by inputting the normal treatment scheme in the historical treatment scheme to the second feature extraction module; the normal transaction prediction value is input to the linear layer of the transaction prediction module to obtain a normal transaction fitting curve representing the change of the normal transaction prediction value over time; the parameters in the transaction prediction module are updated until the fitting transaction deviation between the normal transaction fitting curve and the real transaction fitting curve meets the preset condition.

[0088] In an embodiment of the present disclosure, the normal transaction feature sequence can represent the normal transaction value in the normal treatment scheme. The normal transaction prediction value can be a quantization of the normal transaction feature sequence.

[0089] In an embodiment of the present disclosure, a mapping table between the transaction feature and the transaction value can be preset. According to the mapping table, a plurality of normal transaction values changing over time corresponding to the normal transaction feature sequence are determined. The plurality of normal transaction values changing over time are input to the linear layer of the transaction prediction module to obtain a fitting curve of the normal transaction value. In the case that the fitting transaction deviation between the normal transaction fitting curve and the real transaction fitting curve meets the preset condition, the training process of the transaction prediction module is ended, and the trained prediction transaction module is obtained.

[0090] According to an embodiment of the present disclosure, by quantizing the normal transaction feature sequence and training the transaction prediction module based on the normal transaction prediction value, the specific difference between the actual transaction value and the normal transaction value can be determined, and the prediction result of the transaction prediction module is more accurate and reliable.

[0091] In embodiments of the present disclosure, in the case of an abnormal treatment scheme, for example, the LSTM time series feature shared with the first feature extraction module can also be input, and the output is the predicted abnormal amount. The supervision signal can use the mean squared error loss (MSE) to optimize the prediction accuracy.

[0092] In embodiments of the present disclosure, for example, the time series feature mining layer can be composed of multiple feature extraction modules, which extract high-dimensional features from the input dimension to the output dimension . .

[0093] For example, the feature extraction modules are stacked to realize feature extraction of data with an input dimension of , and output time series features with an output dimension of . , where is the number of time steps required by the LSTM module.

[0094]

[0095] where W represents the number of stacked feature extraction modules (i.e., the depth of the network), and its core role is to enhance the abstract ability of time series features layer by layer. (For example, W=2 represents two layers of modules in series), each layer of feature extraction module performs nonlinear transformation on the input data, and gradually extracts higher-order time series features. Increasing W can improve the model's ability to capture complex time series patterns.

[0096] In embodiments of the present disclosure, for example, the time series features can also be further extracted by an X-layer multi-layer perception (MLP) fully connected neural network, and the output dimension of the extracted features is . , , where represents the number of abnormal parameters in the abnormal features:

[0097]

[0098] , where X represents the number of layers of the multi-layer perception.

[0099] Based on the extracted cost abnormality features , the result of whether there is a cost abnormality is obtained .

[0100] For example, the time series features can also be further extracted by a V-layer MLP fully connected neural network, and the output dimension of the extracted features is . :

[0101]

[0102] wherein V represents the number of layers of the MLP.

[0103] Using the regression layer, based on the extracted cost normal amount features and the cost actual amount, the cost abnormal amount is obtained .

[0104] Figure 5 The process of training the first feature extraction module and the second feature extraction module according to the embodiments of the present disclosure is schematically shown.

[0105] As Figure 5 shown, the input historical treatment plan 501, the time sequence feature 502 is extracted, and the time sequence feature includes the treatment time sequence feature, the same kind of drug transaction information, the same kind of disease treatment transaction information, the preset drug transaction rule and the historical treatment transaction information of the same patient to the historical treatment transaction information. The treatment time sequence feature is input into the first feature extraction module 503, and the abnormal classification training feature 506 is obtained. The first feature extraction module 508 is trained by using the abnormal classification training feature, and the classification error value 510 is determined.

[0106] The drug transaction feature 504 is determined by the same kind of drug transaction information, the same kind of disease treatment transaction information, the preset drug transaction rule and the historical treatment transaction information of the same patient to the historical treatment transaction information. After determining the drug dose feature 505, the drug transaction feature and the drug dose feature are spliced to determine the prediction transaction training feature 507. The second feature extraction module 509 is trained by using the prediction transaction training feature, and the fitting transaction deviation 511 is determined. The weighted sum value 512 is determined according to the classification error value and the fitting deviation value, and it is judged whether the weighted sum value reaches the target value.

[0107] Figure 6 The structural block diagram of the medical information anomaly detection device according to the embodiments of the present disclosure is schematically shown.

[0108] As Figure 6 shown, the medical information anomaly detection device 600 of the embodiments includes a first input module 610, a classification module 620, a second input module 630, a third input module 640 and a detection module 650.

[0109] The first input module 610 is used for inputting the treatment plan into the first feature extraction module in the anomaly detection model to obtain an abnormal classification feature sequence including at least one abnormal classification feature sorted according to time information. In an embodiment, the first input module 610 can be used to perform the operation S210 described in the foregoing, which will not be described here again.

[0110] The classification module 620 is configured to classify the treatment scheme by using the abnormal classification feature sequence, to obtain a classification result. In an embodiment, the classification module 620 can be configured to perform the operation S220 described above, and details are not repeated here.

[0111] The second input module 630 is configured to input the treatment scheme into the second feature extraction module in the abnormal detection model when the classification result indicates that the treatment scheme is abnormal, to obtain a transaction feature sequence including at least one transaction feature sorted according to time information. In an embodiment, the second input module 630 can be configured to perform the operation S230 described above, and details are not repeated here.

[0112] The third input module 640 is configured to input the transaction feature sequence into a transaction prediction module in the abnormal detection model, to obtain normal transaction prediction information, the transaction prediction module being trained by using normal treatment schemes in historical treatment schemes. In an embodiment, the third input module 640 can be configured to perform the operation S240 described above, and details are not repeated here.

[0113] The detection module 650 is configured to detect treatment transaction information in the treatment scheme by using the normal transaction prediction information, to obtain a detection result. In an embodiment, the detection module 650 can be configured to perform the operation S250 described above, and details are not repeated here.

[0114] According to an embodiment of the present disclosure, by inputting the treatment scheme into the first feature extraction module in the abnormal detection model, an abnormal classification feature sequence including at least one abnormal classification feature sorted according to time information is obtained; the treatment scheme is classified by using the abnormal classification feature sequence, to obtain a classification result, so that the classification result is obtained by comprehensively considering the situation that the treatment scheme changes with time information, and the hidden relationship between various information in the treatment scheme can be captured, and the accuracy of abnormal detection can be improved. When the classification result indicates that the treatment scheme is abnormal, the treatment scheme is input into the second feature extraction module in the abnormal detection model, to obtain a transaction feature sequence including at least one transaction feature sorted according to time information; in order to accurately capture abnormal transaction data, the transaction feature sequence can be input into the transaction prediction module in the abnormal detection model, to obtain normal transaction prediction information, and since the transaction prediction module is trained by using normal treatment schemes in historical treatment schemes, the normal transaction prediction information is based on the treatment scheme to predict normal transaction data, rather than transaction data in future time; the treatment transaction information in the treatment scheme is detected by using the normal transaction prediction information, to obtain a detection result, and intelligent detection of the treatment scheme is realized, manual review is reduced, and the efficiency of abnormal detection is improved.

[0115] According to an embodiment of the present disclosure, any multiple modules of the first input module 610, the classification module 620, the second input module 630, the third input module 640 and the detection module 650 can be combined in one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the function of one or more of the modules can be combined with at least part of the function of other modules, and implemented in one module. According to an embodiment of the present disclosure, at least one of the first input module 610, the classification module 620, the second input module 630, the third input module 640 and the detection module 650 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on substrate, a system on package, an application specific integrated circuit (ASIC), or any other reasonable manner that can be integrated or packaged by a circuit, etc. hardware or firmware, or any one of software, hardware and firmware or any appropriate combination of any of them. Alternatively, at least one of the first input module 610, the classification module 620, the second input module 630, the third input module 640 and the detection module 650 can be at least partially implemented as a computer program module which can perform corresponding functions when executed.

[0116] According to an embodiment of the present disclosure, the historical treatment scheme further includes historical patient information, historical doctor information, historical transaction information and historical treatment drug information.

[0117] According to an embodiment of the present disclosure, the medical information anomaly detection device 600 further includes a feature processing module, a treatment time sequence feature input module and a first training module. The feature processing module is configured to perform feature processing on the plurality of historical treatment schemes according to the time information to obtain treatment time sequence features. The treatment time sequence feature input module is configured to input the treatment time sequence features into the first feature extraction module to obtain anomaly classification training features. The first training module is configured to train the first feature extraction module using the anomaly classification training features until a classification error value between a training classification result and a true classification result satisfies a preset condition, the training classification result being obtained by classifying the historical treatment schemes using the anomaly classification training features.

[0118] According to an embodiment of the present disclosure, the second feature extraction module is trained using associated information of the historical treatment transaction information in the historical treatment scheme, the associated information including similar drug transaction information, similar disease treatment transaction information, a preset drug transaction rule and historical treatment transaction information of the same patient.

[0119] According to an embodiment of the present disclosure, the abnormality detection device 600 of medical information further comprises a feature extraction module, a dose threshold determination module, a vectorization processing module, a feature splicing module and a second training module. The feature extraction module is configured to extract features from the historical treatment transaction information according to the same kind of drug transaction information, the same kind of disease treatment transaction information, the preset drug transaction rule and the historical treatment transaction information of the same patient, to obtain drug transaction features. The dose threshold determination module is configured to determine the doctor dose threshold of each of the plurality of drugs according to the historical cumulative dose of each of the plurality of drugs by the doctor in the historical doctor information. The vectorization processing module is configured to perform vectorization processing on the reference dose and the doctor dose threshold of each of the plurality of drugs to obtain drug dose features. The feature splicing module is configured to splice the drug transaction features and the drug dose features to obtain prediction transaction training features. The second training module is configured to train the second feature extraction module using the prediction transaction training features.

[0120] According to an embodiment of the present disclosure, the feature extraction module comprises a drug transaction range determination sub-module, a treatment transaction range determination sub-module and a range vectorization sub-module. The drug transaction range determination sub-module is configured to determine the drug transaction range of the plurality of drugs in the historical treatment transaction information according to the preset drug transaction rule and the same kind of drug transaction information. The treatment transaction range determination sub-module is configured to determine the treatment transaction range of the historical treatment transaction information according to the same kind of disease treatment transaction information and the historical treatment transaction information of the same patient. The range vectorization sub-module is configured to perform vectorization processing on the drug transaction value and the drug transaction range of each of the plurality of drugs, the historical treatment transaction information of the same patient and the treatment transaction range to obtain drug transaction features.

[0121] According to an embodiment of the present disclosure, the detection module 650 comprises a transaction deviation determination sub-module, a first detection result determination sub-module and a second detection result determination sub-module. The transaction deviation determination sub-module is configured to determine the transaction deviation between the normal transaction prediction information and the treatment transaction information in the treatment scheme. The first detection result determination sub-module is configured to determine a detection result representing that there is no abnormal transaction value in the treatment transaction information when the transaction deviation satisfies the deviation threshold condition. The second detection result determination sub-module is configured to determine a detection result representing that there is an abnormal transaction value in the treatment transaction information when the transaction deviation does not satisfy the deviation threshold condition.

[0122] According to an embodiment of the present disclosure, the abnormality detection apparatus 600 of medical information further comprises a mapping module, a predicted value input module and a parameter updating module. The mapping module is configured to map the normal transaction feature sequence according to the mapping relationship between the transaction features and the transaction values, to obtain the normal transaction predicted values, the normal transaction feature sequence being obtained by inputting the normal treatment scheme in the historical treatment schemes into the second feature extraction module. The predicted value input module is configured to input the normal transaction predicted values into the linear layer of the transaction prediction module, to obtain the normal transaction fitting curve representing the change of the normal transaction predicted values over time. The parameter updating module is configured to update the parameters in the transaction prediction module until the fitting transaction deviation between the normal transaction fitting curve and the real transaction fitting curve meets the preset condition.

[0123] According to an embodiment of the present disclosure, the preset condition comprises that the weighted sum value between the fitting transaction deviation and the classification error value reaches a target value.

[0124] Figure 7 A block diagram of an electronic device suitable for implementing the abnormality detection method of treatment schemes according to an embodiment of the present disclosure is schematically shown.

[0125] As shown in Figure 7 , the electronic device 700 according to an embodiment of the present disclosure comprises a processor 701 which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 702 or loaded into a random access memory (RAM) 703 from a storage portion 708. The processor 701 may, for example, include a general-purpose microprocessor (e.g., a CPU), an instruction set processor, and / or a related chipset, and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), and the like. The processor 701 can also include an on-board memory for cache use. The processor 701 can include a single processing unit or a plurality of processing units for performing different actions of the method processes according to embodiments of the present disclosure.

[0126] In the RAM 703, various programs and data required for the operation of the electronic device 700 are stored. The processor 701, the ROM 702 and the RAM 703 are connected to each other through a bus 704. The processor 701 performs various operations of the method processes according to embodiments of the present disclosure by executing the programs in the ROM 702 and / or the RAM 703. It should be noted that the programs can also be stored in one or more memories other than the ROM 702 and the RAM 703. The processor 701 can also perform various operations of the method processes according to embodiments of the present disclosure by executing the programs stored in the one or more memories.

[0127] According to an embodiment of the present disclosure, the electronic device 700 can further include an input / output (I / O) interface 705 also connected to the bus 704. The electronic device 700 can further include one or more of the following components connected to the input / output (I / O) interface 705: an input part 706 including a keyboard, a mouse, etc.; an output part 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage part 708 including a hard disk, etc.; and a communication part 709 including a network interface card such as a LAN card, a modem, etc. The communication part 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output (I / O) interface 705 as necessary. A removable medium 711 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is mounted on the drive 710 as necessary, so that a computer program read out therefrom is installed in the storage part 708 as necessary.

[0128] The present disclosure also provides a computer readable storage medium, which can be included in the device / apparatus / system described in the above embodiments, or can exist separately without being assembled into the device / apparatus / system. The above computer readable storage medium carries one or more programs, which when executed, implement the method according to the embodiments of the present disclosure.

[0129] According to an embodiment of the present disclosure, the computer readable storage medium can be a non-volatile computer readable storage medium, which can include, but is not limited to, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any appropriate combination thereof. In the present disclosure, the computer readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, the computer readable storage medium can include one or more of the above-described ROM 702 and / or RAM 703 and / or one or more memories other than the ROM 702 and the RAM 703.

[0130] The embodiments of the present disclosure also include a computer program product, which includes a computer program containing program codes for executing the methods shown in the flowcharts. When the computer program product is run in a computer system, the program codes are used to make the computer system implement the abnormality detection method of the treatment scheme provided by the embodiments of the present disclosure.

[0131] The above-described functions of the system / device defined in the system / apparatus of the embodiments of the present disclosure are performed when the computer program is executed by the processor 701. According to the embodiments of the present disclosure, the system, apparatus, module, unit, etc. described above can be implemented by the computer program modules.

[0132] In one embodiment, the computer program can be stored in a tangible storage medium, such as an optical, magnetic, or other memory on a hard disk drive, solid-state drive, or other storage device. In another embodiment, the computer program can be transmitted over a network, using a wireless or wired transmission medium, and downloaded and installed by a communication portion 709, and / or installed from a removable medium 711. The computer program code contained in the computer program can be transmitted using any appropriate medium, including but not limited to wireless, wired, optical fiber cable, or any suitable combination of the foregoing.

[0133] In such an embodiment, the computer program can be downloaded and installed from a network via the communication portion 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, the above-described functions of the system defined in the embodiments of the present disclosure are performed. According to the embodiments of the present disclosure, the system, apparatus, device, module, unit, etc. described above can be implemented by the computer program modules.

[0134] According to the embodiments of the present disclosure, the program code for carrying out the computer program provided by the embodiments of the present disclosure can be written in any combination of one or more programming languages, and specifically, these computer programs can be implemented using a high-level procedural and / or object-oriented programming language, and / or an assembly / machine language. The programming language includes, but is not limited to, a programming language such as Java, C++, Python, "C" language, or a similar programming language. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In the case involving a remote computing device, the remote computing device can be connected to the user computing device through any kind of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, connected to the Internet through an Internet service provider).

[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0136] Those skilled in the art will appreciate that the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in the present disclosure. In particular, the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways without departing from the spirit and teachings of the present disclosure. All such combinations and / or couplings fall within the scope of the present disclosure.

[0137] The above describes the embodiments of the present disclosure. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be advantageously used in combination. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for detecting abnormalities in a treatment plan, characterized in that: The method comprises: Inputting the treatment plan into a first feature extraction module in the anomaly detection model to obtain an anomaly classification feature sequence including at least one anomaly classification feature sorted according to time information; Classifying the treatment plan using the abnormal classification feature sequence to obtain a classification result; If the classification result indicates that the treatment plan is abnormal, inputting the treatment plan into a second feature extraction module in the anomaly detection model to obtain a transaction feature sequence including at least one transaction feature sorted according to the time information; Inputting the transaction feature sequence into a transaction prediction module in the anomaly detection model to obtain normal transaction prediction information, wherein the transaction prediction module is trained using normal treatment plans in historical treatment plans; The normal transaction prediction information is used to detect the treatment transaction information in the treatment plan to obtain a detection result.

2. The method according to claim 1, characterized in that The historical treatment plan also includes historical patient information, historical doctor information, historical transaction information and historical treatment drug information.

3. The method according to claim 2, characterized in that The first feature extraction module is trained based on the following method: Performing feature processing on the plurality of historical treatment plans according to the time information to obtain treatment time series features; Inputting the treatment time series features into the first feature extraction module to obtain abnormal classification training features; The first feature extraction module is trained using the abnormal classification training feature until a classification error value between a training classification result and a true classification result meets a preset condition, wherein the training classification result is obtained by classifying the historical treatment plan using the abnormal classification training feature.

4. The method according to claim 1, wherein The second feature extraction module is trained using the associated information of the historical treatment transaction information in the historical treatment plan, where the associated information includes transaction information of similar drugs, transaction information of treatments of similar diseases, preset drug transaction rules, and historical treatment transaction information of the same patient.

5. The method according to claim 4, characterized in that The second feature extraction module is trained based on the following method: Extracting features from the historical treatment transaction information based on the transaction information of similar drugs, the transaction information of treatments of similar diseases, the preset drug transaction rules, and the historical treatment transaction information of the same patient to obtain drug transaction features; determining a doctor's dosage threshold for each of the plurality of drugs based on the doctor's historical cumulative dosage for each of the plurality of drugs in the historical doctor information; Performing vector quantization on the respective benchmark doses of the plurality of drugs and the doctor's dose threshold to obtain drug dose characteristics; splicing the drug transaction feature with the drug dosage feature to obtain a predicted transaction training feature; The predicted transaction training features are used to train a second feature extraction module.

6. The method according to claim 5, characterized in that The extracting features of the historical treatment transaction information based on the transaction information of similar drugs, the transaction information of treatments of similar diseases, the preset drug transaction rules, and the historical treatment transaction information of the same patient to obtain drug transaction features includes: determining a drug transaction range of the plurality of drugs in the historical treatment transaction information according to the preset drug transaction rules and the transaction information of similar drugs; determining a treatment transaction scope of the historical treatment transactions based on the treatment transaction information of the same disease and the historical treatment transaction information of the same patient; The drug transaction values ​​and the drug transaction ranges of the respective drugs, the historical treatment transaction information of the same patient, and the treatment transaction range are vectorized to obtain the drug transaction features.

7. The method according to claim 3, characterized in that The detecting of the treatment transaction information in the treatment plan using the normal transaction prediction information to obtain the detection result includes: determining a transaction deviation between the normal transaction prediction information and the treatment transaction information in the treatment plan; In a case where the transaction deviation satisfies a deviation threshold condition, determining the detection result indicating that no abnormal transaction value exists in the treatment transaction information; In a case where the transaction deviation does not satisfy a deviation threshold condition, determining the detection result indicating that an abnormal transaction value exists in the treatment transaction information.

8. The method according to claim 7, characterized in that The transaction prediction module is trained based on the following method: Mapping a normal transaction feature sequence based on a mapping relationship between transaction features and transaction values ​​to obtain a normal transaction prediction value, wherein the normal transaction feature sequence is obtained by inputting a normal treatment plan in the historical treatment plan into the second feature extraction module; Inputting the normal transaction prediction value into the linear layer of the transaction prediction module to obtain a normal transaction fitting curve representing the change of the normal transaction prediction value over time; The parameters in the transaction prediction module are updated until the fitted transaction deviation between the normal transaction fitting curve and the real transaction fitting curve meets a preset condition.

9. The method according to claim 8, characterized in that The preset condition includes: a weighted sum of the fitting transaction deviation and the classification error value reaches a target value.

10. A device for detecting abnormalities in a treatment plan, characterized in that: The device comprises: A first input module, configured to input the treatment plan into a first feature extraction module in the anomaly detection model to obtain an anomaly classification feature sequence including at least one anomaly classification feature sorted according to time information; A classification module, configured to classify the treatment plan using the abnormal classification feature sequence to obtain a classification result; a second input module configured to input the treatment plan into a second feature extraction module in the anomaly detection model when the classification result indicates that the treatment plan is abnormal, to obtain a transaction feature sequence including at least one transaction feature sorted according to the time information; a third input module, configured to input the transaction feature sequence into a transaction prediction module in the anomaly detection model to obtain normal transaction prediction information, wherein the transaction prediction module is trained using normal treatment plans in historical treatment plans; The detection module is used to detect the treatment transaction information in the treatment plan using the normal transaction prediction information to obtain a detection result.

Images