WO2021151302A1 - 基于机器学习的药品质控分析方法、装置、设备及介质 - Google Patents

基于机器学习的药品质控分析方法、装置、设备及介质 Download PDF

Info

Publication number
WO2021151302A1
WO2021151302A1 PCT/CN2020/119063 CN2020119063W WO2021151302A1 WO 2021151302 A1 WO2021151302 A1 WO 2021151302A1 CN 2020119063 W CN2020119063 W CN 2020119063W WO 2021151302 A1 WO2021151302 A1 WO 2021151302A1
Authority
WO
WIPO (PCT)
Prior art keywords
drug
identification information
disease
attribute
preset
Prior art date
Application number
PCT/CN2020/119063
Other languages
English (en)
French (fr)
Inventor
李彦轩
孙行智
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021151302A1 publication Critical patent/WO2021151302A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
    • G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H70/00—ICT specially adapted for the handling or processing of medical references
    • G16H70/40—ICT specially adapted for the handling or processing of medical references relating to drugs, e.g. their side effects or intended usage
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • This application relates to the field of information technology, in particular to a method, device, equipment and medium for drug quality control analysis based on machine learning.
  • the patient's medication is usually given by the attending physician based on his diagnosis and examination results.
  • the quality of the medicine prescribed by the doctor can be monitored through machine learning. , To avoid poor treatment results for patients due to irrational use of drugs or additional drug expenditures for patients.
  • the names of diseases and drugs prescribed by doctors are usually used to monitor the quality of drugs.
  • the inventor realized that this way of quality monitoring only relying on disease names and drug names, the feature space of the machine learning model is relatively limited, and the model cannot obtain more feature information during the quality monitoring process, which leads to a relatively low quality control accuracy of the model. Low, unable to effectively judge the rationality of the doctor's medication.
  • a method for drug quality control analysis based on machine learning including:
  • the quality control analysis result of the medicine in the patient's medical record is determined according to the identification feature vector and the attribute feature vector.
  • a drug quality control analysis device based on machine learning including:
  • the obtaining unit is configured to obtain disease identification information and drug identification information in the patient's medical record, and respectively determine the disease attribute information corresponding to the disease identification information and the drug attribute information corresponding to the drug identification information;
  • the first extraction unit is configured to input the disease identification information and the drug identification information into a preset identification feature extraction model for feature extraction, to obtain an identification feature vector corresponding to the disease identification information and the drug identification information;
  • the second extraction unit is configured to input the disease attribute information and the drug attribute information features into a preset attribute feature extraction model for feature extraction, and obtain the attribute feature vector corresponding to the disease attribute information and the drug attribute information ;
  • the determining unit is configured to determine the quality control analysis result of the medicine in the patient's medical record according to the identification feature vector and the attribute feature vector.
  • a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the following steps are implemented:
  • the quality control analysis result of the medicine in the patient's medical record is determined according to the identification feature vector and the attribute feature vector.
  • a computer device including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the following steps when the program is executed:
  • the quality control analysis result of the medicine in the patient's medical record is determined according to the identification feature vector and the attribute feature vector.
  • Figure 1 shows a flow chart of a method for drug quality control analysis based on machine learning provided by an embodiment of the present application
  • FIG. 2 shows a flowchart of another method for drug quality control analysis based on machine learning provided by an embodiment of the present application
  • FIG. 3 shows a flowchart of another method for drug quality control analysis based on machine learning provided by an embodiment of the present application
  • Figure 4 shows a schematic diagram of a model structure for drug quality control analysis provided by an embodiment of the present application
  • Figure 5 shows a schematic structural diagram of another machine learning-based drug quality control analysis device provided by an embodiment of the present application.
  • FIG. 6 shows a schematic diagram of the physical structure of a computer device provided by an embodiment of the present application.
  • this method of quality monitoring only relies on disease names and drug names, and the feature space of machine learning models is relatively limited. In the process of quality monitoring, the model cannot obtain more feature information, which leads to lower quality control accuracy of the model. It is impossible to effectively judge the rationality of the doctor's medication.
  • an embodiment of the present application provides a method for drug quality control analysis based on machine learning. As shown in FIG. 1, the method includes:
  • the disease identification information and the drug identification information can be standardized disease names and standardized drug names.
  • the disease attribute information includes disease characteristics such as the incidence of the disease (high, medium, and low), the department to which the disease belongs, and the location of the disease.
  • the drug attribute information includes the drug category specified by the ATC code, the main components of the drug and other drug characteristics.
  • the embodiment of this application is mainly applied to the quality control analysis of the drug.
  • this embodiment also involves blockchain technology, which can store disease identification information and drug identification information in the patient’s medical record in the blockchain. When performing quality control analysis of the drug, select the patient's medical record to be analyzed, click the quality control analysis button, and trigger the drug quality control analysis instruction.
  • the patient's medical record to be analyzed carried in the instruction is performed Quality control analysis, specifically, first obtain the disease name and drug name recorded in the patient's medical record to be analyzed. Because the disease name and drug name recorded in the patient's medical record may be aliases for a certain disease or drug, it is not a standardized name, so it is not convenient to compare It performs quality control analysis, so it is necessary to standardize the acquired disease names and drug names to obtain standardized disease names and standardized drug names.
  • the embodiment of the application expands the feature space of the machine learning model in the quality control analysis process by introducing disease attribute information and drug attribute information, increases the information dimension in the quality control analysis process, and improves the quality control analysis accuracy of the model. Ensure the accuracy of quality control analysis results.
  • the acquired disease identification information and drug identification information are input into a preset identification feature extraction model for feature extraction, and the disease identification information and The identification feature vector corresponding to the drug identification information.
  • the preset identification feature extraction model can be, but is not limited to, pre- Suppose a text convolutional neural network model.
  • the preset text convolutional neural network model consists of an embedding layer, a convolutional layer, a pooling layer, a splicing layer and an output layer.
  • the disease identification information and drug identification information are one-dimensional data in the form of text .
  • the preset text convolutional neural network model can perform one-dimensional convolution on the input one-dimensional data in the form of text, and extract the identification feature vector corresponding to the disease identification information and the drug identification information, so as to compare the patient with the extracted identification feature vector.
  • the drugs in the medical records are analyzed for quality control.
  • the preset attribute feature extraction model may be a first preset DNN neural network model, which can process structured data.
  • the structured disease attribute information and drug attribute The information is input to the first preset DNN neural network model for feature extraction.
  • the first preset DNN neural network model contains two hidden layers for extracting the attribute feature vector corresponding to the disease attribute information and the drug attribute information.
  • the second layer The hidden layer will output the extracted attribute feature vector to combine with the identification feature vector to determine the rationality of the patient's medication.
  • the feature space of the machine learning model can be expanded, and the accuracy of the quality control analysis of the model can be improved.
  • the embodiment of the application uses a preset text convolutional neural network model to characterize the drug identification information and disease identification information in text form. Extraction and feature extraction of structured drug attribute information and disease attribute information using the first preset DNN neural network model solves the problem of processing structured data and unstructured data at the same time, and improves the data compatibility of the model sex.
  • the quality control analysis result of the drug includes two results of abnormal patient medication and normal patient medication.
  • the extracted The identification feature vector and the attribute feature vector are combined.
  • the identification feature vector and the attribute feature vector can be combined, and the quality control analysis result of the medicine in the patient’s medical record is determined according to the combined vector.
  • the method further includes: using a preset abnormality detection rule to perform abnormality detection on the disease identification information and the drug identification information to obtain an abnormality detection result; verifying the quality according to the abnormality detection result Control the accuracy of analysis results, where the preset anomaly detection rule is a rule extracted based on historical anomaly detection results, for example, based on historical anomaly detection results, it is determined that disease A and drug B cannot be used, so that multiple abnormality detections can be obtained Rules, using preset abnormality detection rules to perform abnormality detection on the disease identification information and drug identification information in the patient’s medical record, and obtain the abnormality detection result, so as to verify the accuracy of the quality control analysis result according to the abnormality detection result.
  • the preset anomaly detection rule is a rule extracted based on historical anomaly detection results, for example, based on historical anomaly detection results, it is determined that disease A and drug B cannot be used, so that multiple abnormality detections can be obtained
  • Rules using preset abnormality detection rules to perform abnormality detection on the disease identification information and drug identification information in
  • the quality control analysis result is If the medication is abnormal, and the disease identification information and drug identification information in the patient’s medical record meet the preset abnormality detection rules, it is determined that the quality control analysis result is correct and the patient's medication is indeed abnormal. In addition, if the quality control analysis result is that the patient's medication is normal , And the disease identification information and drug identification information in the patient’s medical record meet the preset abnormality detection rules, it is determined that the quality control analysis result is incorrect, and the patient’s medication may be abnormal; if the disease identification information and drug identification information in the patient’s medical record If the preset abnormality detection rule is not met, it is determined that the quality control analysis result is correct, and the patient's medication is indeed normal. Therefore, the quality control analysis result can be verified by using the preset abnormality detection rule to ensure the accuracy of the quality control analysis result sex.
  • the embodiment of the application provides a method for analyzing drug quality control based on machine learning.
  • this application can obtain information in the patient’s medical record.
  • Disease identification information and drug identification information and respectively determine the disease attribute information corresponding to the disease identification information and the drug attribute information corresponding to the drug identification information; and input the disease identification information and the drug identification information into the preset
  • the identification feature extraction model performs feature extraction to obtain the identification feature vector corresponding to the disease identification information and the drug identification information; at the same time, the disease attribute information and the drug attribute information characteristics are input into the preset attribute feature extraction model to perform Feature extraction to obtain the attribute feature vector corresponding to the disease attribute information and the drug attribute information; finally, the quality control analysis result of the drug in the patient's medical record is determined according to the identification feature vector and the attribute feature vector, thereby Based on the use of disease identification information and drug identification information, the introduction of disease attribute information and drug attribute information expands the feature space of the machine learning model,
  • the embodiment of the present application provides another method for the quality control analysis of medicines based on machine learning, as shown in FIG. 2 As shown, the method includes:
  • step 201 specifically includes: querying a preset structured disease attribute table according to the disease identification information to determine the disease Identify the disease attribute information corresponding to the identification information; query a preset structured drug attribute table according to the drug identification information to determine the drug attribute information corresponding to the drug identification information.
  • the preset structured disease attribute table stores different disease identification information and corresponding disease attribute information
  • the preset structured drug attribute table stores different drug identification information and corresponding drug attribute information.
  • the server when the server receives the drug quality control analysis instruction, it performs quality control analysis on the patient's medical record to be analyzed carried in the quality control analysis instruction, and first obtains the disease name and drug name recorded in the patient's medical record.
  • the name of the drug may be an alias for a certain disease or drug, not a standardized name.
  • the acquired disease names and drug names need to be standardized.
  • the preset disease name database and the preset drug name database respectively according to the acquired disease names and drug names, and determine the standardized names of the diseases and drugs recorded in the patient’s medical records, that is, disease identification information and drug identification information, where, The standardized names of different diseases and their corresponding aliases are stored in the preset disease name database, and the standardized names of different drugs and their corresponding aliases are stored in the preset drug name database.
  • the drug name database can standardize the disease names and drug names recorded in the patient's medical records, and obtain standardized disease names and standardized drug names in text form.
  • the disease attribute information corresponding to the disease identification information and the drug attribute information corresponding to the drug identification information query a preset structured disease attribute table according to the disease identification information to determine the structured disease attribute information corresponding to the disease identification information, In the same way, query the preset structured drug attribute table based on the drug identification information, and determine the structured drug attribute information corresponding to the drug identification information, so that the disease attribute information and drug attributes can be introduced on the basis of the disease name and drug name.
  • Information which expands the feature space of the machine learning model, enables the machine learning model to obtain more feature information, and improves the accuracy of the model’s quality control analysis.
  • disease identification information and drug identification information are one-dimensional data in text form.
  • the preset identification feature extraction model may be a preset text Convolutional neural network model.
  • the preset text convolutional neural network consists of an embedding layer, a convolutional layer, a pooling layer, a splicing layer, and an output layer.
  • the disease identification information and drug identification information are standardized, the disease name and standardization There are m types of drug names, and the one-hot vector of a standardized disease name or drug name is represented as a vector with length m, the value of the position corresponding to the name, and the value of 0 in the remaining positions.
  • the vector After being mapped by the embedding layer, the vector will be transformed into an embedding vector with length n and n much smaller than m. Compared to the 0-1 representation of the one-hot vector, the value of each position of the vector is a continuous decimal.
  • the convolution layer mainly uses multiple convolution kernels of different sizes to perform convolution operations on the embedding vector input to the layer. The formula is as follows:
  • ci represents the result of the convolution operation with the starting point convolution kernel of the i-th line of the input embedding vector
  • f represents the convolution operation
  • w represents the parameters in the convolution kernel
  • Xi:i+h-1 represents the input embedding vector Line i to i+h-1 (h is the size of the convolution kernel)
  • b is the bias term of the operation, that is, in this step, the convolution kernel will select the standardized disease name or the standardized drug name with the length of h
  • the pooling layer will splice these vectors and perform a pooling operation.
  • the formulas for splicing and pooling operations are as follows:
  • the pooling layer can obtain the largest value among the output values of the convolutional layer.
  • the splicing layer will get multiple The vector is spliced, and the information obtained is about to be integrated.
  • the formula of the output layer is as follows:
  • the softmax activation function is used to introduce non-linear factors to obtain the final probability of each category as the output, but in this application, only the output of the pooling layer is selected as the identification feature vector to represent the preset text convolutional neural network
  • the model obtains information and hidden features from standardized disease names and standardized drug names in text form.
  • a preset attribute feature extraction model is used to perform feature extraction on the disease attribute information and drug attribute information.
  • the preset attribute feature extraction model may be the first preset DNN
  • the neural network model, the specific process of extracting the attribute feature vector corresponding to the disease attribute information and the drug attribute information by using the first preset DNN neural network model is the same as step 103, and will not be repeated here.
  • the concatenate layer concatenate is used to concatenate the extracted identification feature vector and the attribute feature vector, so that the text form can be combined.
  • the implicit features corresponding to the standardized disease name and the standardized drug name are fused with the implicit features corresponding to the structured disease attribute information and the drug attribute information to obtain the spliced feature vector, so as to perform quality according to the spliced feature vector. Control analysis.
  • the spliced features are input into the preset quality control analysis model for analysis, and the quality control analysis results of the medicines in the patient's medical records are obtained, that is, whether there are any in the patient's medical records
  • the preset quality control analysis model may be the second preset DNN neural network model, and the output layer of the second preset DNN neural network model is the same as the output layer of the preset text convolutional neural network model, It calculates the spliced feature vector to get the conclusion of whether the medication in the patient's medical record is abnormal.
  • the embodiment of the present application adopts a preset identification feature extraction model and a preset attribute feature extraction model.
  • the preset quality control analysis model solves the problem of processing structured data and unorganized data at the same time, and improves the data compatibility of the model.
  • the above three models are used as Perform training as a whole to obtain historical disease identification information and historical drug identification information in each historical patient's medical record, as well as the quality control analysis results corresponding to each historical patient's medical record, and then determine the historical disease attributes corresponding to the historical disease identification information.
  • Information and historical drug attribute information corresponding to the historical drug identification information combining the historical disease identification information, the historical drug identification information, the historical disease attribute information, the historical drug attribute information, and the quality control analysis result
  • a preset neural network algorithm is used to train the sample training set to construct and construct the preset identification feature extraction model, the preset attribute feature extraction model, and the preset quality control analysis model.
  • the three models are used as a common task one, that is, by calling task one, structured disease attribute information and drug attribute information are introduced, which expands the feature space of the quality control analysis model , Increasing the information dimension of drug quality control analysis, thereby improving the accuracy of quality control analysis results.
  • this application can call task two to use disease attribute information and drug attribute information to analyze whether the medication in the electronic medical record is abnormal.
  • the model in task one does not support a certain disease or drug
  • the disease and the disease corresponding to the drug may be similar to the disease attributes and drug attributes corresponding to the types of diseases and drugs supported in the training set. Therefore, this embodiment of the application can use the existing disease attribute information and drug attribute information in the training set to analyze and deduce the patient’s drug use. rationality.
  • the method further includes: if the preset identification feature extraction model does not support the disease identification information or the drug identification information , The disease attribute information and the drug attribute information are input into a preset attribute feature extraction model for feature extraction, the attribute feature vector corresponding to the disease attribute information and the drug attribute information is obtained, and according to the attribute The feature vector determines the quality control analysis result of the drug in the patient’s medical record; if the preset identification feature extraction model supports the disease identification information and the drug identification information, then the combination of the disease identification information and the drug identification information will be triggered. The step of inputting the drug identification information into a preset identification feature extraction model for feature extraction, and obtaining the identification feature vector corresponding to the disease identification information and the drug identification information.
  • the preset quality control analysis model cannot be output Quality control analysis results, it is determined that the preset identification feature extraction model does not support the disease identification information or drug identification information; if the disease identification information or drug identification information is the existing disease identification information and drug identification information in the sample training set, or will be After analyzing the disease identification information and drug identification information in the patient's medical record and inputting it into the preset identification feature extraction model, and at the same time inputting the disease attribute information and drug attribute information into the preset attribute feature extraction model, the preset quality control analysis model can output quality control Analyze the results and determine that the preset
  • the historical disease identification information, the historical drug identification information, the historical disease feature information, the historical drug attribute information and the The quality control analysis result is used as the first sample training set; the historical disease attribute information, the historical drug attribute information, and the quality control analysis result are used as the second sample training set;
  • the first sample training set is trained to construct the preset identification feature extraction model, the preliminary attribute feature extraction model, and the preset quality control analysis model; according to the second sample training set, the preliminary attribute feature The extraction model is trained, and the preset attribute feature extraction model is constructed.
  • the first preset DNN neural network model is trained twice, and the parameters after the two trainings are used as the final parameters. Therefore, the first preset neural network model is not only used in task one. This is capable of extracting attribute feature vectors and determining the quality control results of drugs in the patient’s medical records based on the extracted attribute feature vectors. That is, the first preset DNN neural network model can support both task one and task two at the same time.
  • This application is implemented For example, take task two as a subtask of task one, so that when the input disease identification information and drug identification information are within the range supported by the task one model, the task one model is used for quality control analysis, and disease identification information and drugs are used Identification information, as well as two parts of disease attribute information and drug attribute information; and when the model of task one does not support disease identification information or drug identification information, task two is called to complete the quality control analysis, and only disease attribute information and drug attribute information are entered Therefore, the analysis range and scalability of the model can be improved.
  • the model can not only accurately analyze the supported data, but also use disease attribute information and drug attribute information for accurate analysis when the data is not supported.
  • the embodiment of this application provides another method for drug quality control analysis based on machine learning.
  • this application uses disease names and Based on the drug name, the disease attribute information and drug attribute information are introduced, and the identification feature vector corresponding to the disease name and the drug name, and the attribute feature vector corresponding to the disease attribute information and the drug attribute information are extracted, and the extracted identification features are extracted.
  • the combination of vector and attribute feature vector expands the feature space of the machine learning model, enables the machine learning model to obtain more feature information, and improves the quality control analysis accuracy of the model.
  • this application adopts a preset identification feature extraction model ,
  • the preset attribute feature extraction model and the preset quality control analysis model solve the problem of processing structured data and unstructured data at the same time, and improve the data compatibility of the model.
  • this application collects multi-task phases.
  • the quality control analysis when the task one model does not support disease identification information or drug identification information, the task two model is called, and the disease attribute information and drug attribute information are used to analyze the rationality of the patient's medication.
  • This model can not only support The data can be accurately analyzed, and disease attribute information and drug attribute information can also be used to accurately analyze unsupported data, which improves the analysis range and scalability of the model.
  • an embodiment of the present application provides a machine learning-based drug quality control analysis device.
  • the device includes: an acquisition unit 31, a first extraction unit 32, and a first extraction unit 32. Two extraction unit 33 and determination unit 33.
  • the acquiring unit 31 may be used to acquire disease identification information and drug identification information in the patient's medical record, and respectively determine the disease attribute information corresponding to the disease identification information and the drug attribute information corresponding to the drug identification information.
  • the acquiring unit 31 is the main functional module of the device to acquire disease identification information and drug identification information in the patient's medical record, and respectively determine the disease attribute information corresponding to the disease identification information and the drug attribute information corresponding to the drug identification information .
  • the first extraction unit 32 may be configured to input the disease identification information and the drug identification information into a preset identification feature extraction model for feature extraction, and obtain the common correspondence between the disease identification information and the drug identification information Identify the feature vector.
  • the first extraction unit 32 inputs the disease identification information and the drug identification information into a preset identification feature extraction model for feature extraction in this device, and obtains the common correspondence between the disease identification information and the drug identification information
  • the main functional module that identifies the feature vector is also the core module.
  • the second extraction unit 33 may be used to input the disease attribute information and the drug attribute information features into a preset attribute feature extraction model for feature extraction, and obtain the common correspondence between the disease attribute information and the drug attribute information The attribute feature vector.
  • the second extraction unit 33 inputs the disease attribute information and the drug attribute information features into a preset attribute feature extraction model for feature extraction in this device, and obtains the disease attribute information and the drug attribute information corresponding to each other.
  • the main function module of the attribute feature vector is also the core module.
  • the determining unit 34 may be configured to determine the quality control analysis result of the medicine in the patient's medical record according to the identification feature vector and the attribute feature vector.
  • the determining unit 34 is the main functional module of the device for determining the quality control analysis result of the medicine in the patient's medical record according to the identification feature vector and the attribute feature vector, and is also a core module.
  • the determination unit 34 includes: a merging module 341 and an analysis module 342.
  • the merging module 341 may be used to merge the identification feature vector and the attribute feature vector to obtain a combined feature vector.
  • the analysis module 342 may be used to input the combined feature vector into a preset quality control analysis model for quality control analysis, and obtain the quality control analysis result of the medicine in the patient's medical record.
  • the device further includes: an output unit 35.
  • the output unit 35 may be configured to input the disease attribute information and the drug attribute information into the preset attribute if the preset identification feature extraction model does not support the disease identification information or the drug identification information
  • the feature extraction model performs feature extraction to obtain the attribute feature vector corresponding to the disease attribute information and the drug attribute information, and determine the quality control analysis result of the drug in the patient's medical record according to the attribute feature vector.
  • the first extraction unit 32 may also be configured to trigger the combination of the disease identification information and the drug identification information if the preset identification feature extraction model supports the disease identification information and the drug identification information. The step of inputting to a preset identification feature extraction model for feature extraction, and obtaining identification feature vectors corresponding to the disease identification information and the drug identification information.
  • the obtaining unit 31 includes: a first determining module 311 and a second determining module 312.
  • the first determining module 311 may be used to query a preset structured disease attribute table according to the disease identification information, and determine the disease attribute information corresponding to the disease identification information.
  • the preset structured disease attribute table stores Different disease identification information and corresponding disease attribute information.
  • the second determining module 312 may be used to query a preset structured drug attribute table according to the drug identification information, and determine the drug attribute information corresponding to the drug identification information.
  • the preset structured drug attribute table stores Different drug identification information and corresponding drug attribute information.
  • the device further includes: a construction unit 36.
  • the acquiring unit 31 can also be used to acquire historical disease identification information and historical drug identification information in each historical patient's medical record, and the quality control analysis result corresponding to each historical patient's medical record.
  • the determining unit 34 may also be used to separately determine historical disease attribute information corresponding to the historical disease identification information and historical drug attribute information corresponding to the historical drug identification information.
  • the determining unit 34 may also be configured to determine a sample training set based on the historical disease identification information, the historical drug identification information, the historical disease attribute information, the historical drug attribute information, and the quality control analysis result .
  • the construction unit 36 may be used to train the sample training set using a preset neural network algorithm, and construct the preset identification feature extraction model, the preset attribute feature extraction model, and the preset quality control analysis Model. .
  • the construction unit 36 includes a training module 361 and a construction module 362.
  • the determining unit 34 may be specifically configured to use the historical disease identification information, the historical drug identification information, the historical disease attribute information, the historical drug attribute information, and the quality control analysis result as the first sample Training set; taking the historical disease attribute information, the historical drug attribute information, and the quality control analysis result as a second sample training set.
  • the training module 361 may be used to train the first sample training set by using a preset neural network algorithm to construct the preset identification feature extraction model, the preliminary attribute feature extraction model, and the preset quality Control analysis model.
  • the construction module 362 may be used to train the preliminary attribute feature extraction model according to the second sample training set to construct the preset attribute feature extraction model.
  • the device further includes: a detection unit 37 and a verification unit 38.
  • the detection unit 37 may be configured to perform abnormality detection on the disease identification information and the drug identification information by using a preset abnormality detection rule to obtain an abnormality detection result.
  • the verification unit 38 may be used to verify the accuracy of the quality control analysis result according to the abnormality detection result.
  • an embodiment of the present application also provides a computer-readable storage medium.
  • the computer-readable storage medium may be volatile or non-volatile, on which storage
  • There is a computer program when the program is executed by the processor, the following steps are realized: obtaining the disease identification information and drug identification information in the patient's medical record, and respectively determining the disease attribute information corresponding to the disease identification information and the drug corresponding to the drug identification information Attribute information; input the disease identification information and the drug identification information into a preset identification feature extraction model for feature extraction, and obtain the identification feature vector corresponding to the disease identification information and the drug identification information; The attribute information and the drug attribute information are input to a preset attribute feature extraction model for feature extraction, and the attribute feature vector corresponding to the disease attribute information and the drug attribute information is obtained; according to the identification feature vector and the attribute The feature vector determines the quality control analysis result of the medicine in the patient's medical record.
  • the computer device includes: a processor 41, The memory 42, and a computer program that is stored on the memory 42 and can run on the processor, wherein the memory 42 and the processor 41 are both set on the bus 43, and the processor 41 implements the following steps when executing the program:
  • Obtain the patient's medical record The disease identification information and the drug identification information in the identification information, and the disease attribute information corresponding to the disease identification information and the drug attribute information corresponding to the drug identification information are respectively determined; the disease identification information and the drug identification information are input into the forecast Set the identification feature extraction model to perform feature extraction to obtain identification feature vectors corresponding to the disease identification information and the drug identification information; input the disease attribute information and the drug attribute information features into a preset attribute feature extraction model to perform Feature extraction obtains the attribute feature vector corresponding to the disease attribute information and the drug attribute information; the quality
  • this application introduces disease attribute information and drug attribute information based on the use of disease name and drug name, and extracts the identification feature vector corresponding to the disease name and drug name, as well as disease attribute information and drug attribute information.
  • the attribute feature vector corresponding to the information and the combination of the extracted identification feature vector and the attribute feature vector expands the feature space of the machine learning model, enables the machine learning model to obtain more feature information, and improves the quality control analysis of the model.
  • this application solves the problem of processing structured data and unstructured data at the same time by adopting a preset identification feature extraction model, a preset attribute feature extraction model, and a preset quality control analysis model, and improves the model’s data performance. Compatibility.
  • this application collects multi-task quality control analysis.
  • the task one model does not support disease identification information or drug identification information
  • the task two model is invoked, and the disease attribute information and drug attribute information are used to analyze and derive The rationality of the patient's medication, so the model can not only accurately analyze the supported data, but also use disease attribute information and drug attribute information to accurately analyze the unsupported data, which improves the analysis range and scalability of the model.
  • modules or steps of this application can be implemented by a general computing device, and they can be concentrated on a single computing device or distributed in a network composed of multiple computing devices.
  • they can be implemented with program codes executable by a computing device, so that they can be stored in a storage device for execution by the computing device, and in some cases, can be executed in a different order than here.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Medical Informatics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Epidemiology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Chemical & Material Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Medicinal Chemistry (AREA)
  • Molecular Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Primary Health Care (AREA)
  • Public Health (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Pharmacology & Pharmacy (AREA)
  • Toxicology (AREA)
  • Medical Treatment And Welfare Office Work (AREA)

Abstract

一种基于机器学习的药品质控分析方法、装置、设备及介质，涉及信息技术领域，主要在于能够能够扩展机器学习模型的特征空间，引入了疾病属性特征和药品属性特征，从而提高了模型的质控分析精度。其中方法包括：确定所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；确定所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。涉及人工智能中的机器学习技术，适用于药品的质控分析，同时适用于智慧医疗领域，从而可进一步推动智慧城市的建设。另外，还涉及区块链技术。

Description

基于机器学习的药品质控分析方法、装置、设备及介质
本申请要求于2020年7月16日提交中国专利局、申请号为CN202010685223.1、名称为“基于机器学习的药品质控分析方法、装置、设备及介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及信息技术领域，尤其是涉及一种基于机器学习的药品质控分析方法、装置、设备及介质。
背景技术
在患者的诊疗过程中，患者的用药通常由主治医师根据其诊断和检查结果给出，为了确保医生所开药品的合理性，可以通过机器学习的方式对医生所开处方中的药品进行质量监控，避免由于不合理用药导致患者治疗结果不佳或者给患者造成额外的药品开支。
目前，通常使用医生所开处方中的疾病名称和药品名称对药品进行质量监控。发明人意识到这种仅依赖疾病名称和药品名称进行质量监控的方式，机器学习模型的特征空间较为局限，在质量监控过程中模型无法获取更多的特征信息，进而导致模型的质控精度较低，无法有效的判断医生用药的合理性。
发明内容
根据本申请的第一个方面，提供一种基于机器学习的药品质控分析方法，包括：
获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；
将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；
将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；
根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
根据本申请的第二个方面，提供一种基于机器学习的药品质控分析装置，包括：
获取单元，用于获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；
第一提取单元，用于将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；
第二提取单元，用于将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；
确定单元，用于根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
根据本申请的第三个方面，提供一种计算机可读存储介质，其上存储有计算机程序，该程序被处理器执行时实现以下步骤：
获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；
将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；
将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；
根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
根据本申请的第四个方面，提供一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现以下步骤：
获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；
将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；
将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；
根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
附图说明
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：
图1示出了本申请实施例提供的一种基于机器学习的药品质控分析方法流程图；
图2示出了本申请实施例提供的另一种基于机器学习的药品质控分析方法流程图；
图3示出了本申请实施例提供的另一种基于机器学习的药品质控分析方法流程图；
图4示出了本申请实施例提供的用于药品质控分析的模型结构示意图；
图5示出了本申请实施例提供的另一种基于机器学习的药品质控分析装置的结构示意图；
图6示出了本申请实施例提供的一种计算机设备的实体结构示意图。
具体实施方式
下文中将参考附图并结合实施例来详细说明本申请。需要说明的是，在不冲突的情况下，本申请中的实施例及实施例中的特征可以相互组合。
目前，这种仅依赖疾病名称和药品名称进行质量监控的方式，机器学习模型的特征空间较为局限，在质量监控过程中模型无法获取更多的特征信息，进而导致模型的质控精度较低，无法有效的判断医生用药的合理性。
为了解决上述问题，本申请实施例提供了一种基于机器学习的药品质控分析方法，如图1所示，所述方法包括：
101、获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息。
其中，疾病标识信息和药品标识信息可以为标准化的疾病名称和标准化的药品名称，疾病属性信息包括疾病的发病率(高、中、低)，疾病所属的科室以及疾病的发病部位等疾病特征，药品属性信息包括ATC编码指定的药品类别、药品主要成分等药品特征，本申请实施例主要应用于药品的质控分析，本申请实施例的执行主体为能够对患者病历中药品进行质控分析的装置或设备，与此同时，本实施例还涉及区块链技术，可将患者病历中的 疾病标识信息和药品标识信息存储与区块链中，对于本申请实施例，当需要对患者病历中的药品进行质控分析时，选择待分析的患者病历，点击质控分析按钮，触发药品质控分析指令，服务器接收到药品质控分析指令后，对该指令中携带的待分析的患者病历进行质控分析，具体地，首先获取待分析的患者病历中记载的疾病名称和药品名称，由于患者病历中记载的疾病名称和药品名称可能为某疾病或者药品的别名，不是标准化名称，不便于对其进行质控分析，因此需要对获取的疾病名称和药品名称进行标准化处理，得到标准化的疾病名称和标准化的药品名称。
进一步地，根据获取的患者病历中的疾病标识信息和药品标识信息，分别查询预设结构化疾病属性表和预设结构化药品属性表，确定疾病标识信息对应的疾病属性信息和药品标识信息对应的药品属性信息，其中，疾病标识信息和药品标识信息为文本形式，疾病属性信息和药品属性信息为结构化数据，预设结构化疾病属性表和预设结构化药品属性表可以存储与区块链中，本申请实施例通过引入疾病属性信息和药品属性信息，扩展了质控分析过程中机器学习模型的特征空间，增加质控分析过程中的信息维度，提高了模型的质控分析精度，确保质控分析结果的准确性。
102、将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量。
对于本申请实施例，为了提取疾病标识信息和药品标识信息共同对应的标识特征向量，将获取的疾病标识信息和药品标识信息输入至预设标识特征提取模型进行特征提取，得到该疾病标识信息和药品标识信息共同对应的标识特征向量，其中，由于疾病标识信息和药品标识信息为文本形式，为了能够对疾病标识信息和药品标识信息进行处理，预设标识特征提取模型可以为但不局限于预设文本卷积神经网络模型，该预设文本卷积神经网络模型由嵌入层、卷积层、池化层、拼接层和输出层组成，疾病标识信息和药品标识信息为文本形式的一维数据，该预设文本卷积神经网络模型能够对输入的文本形式的一维数据进行一维卷积，提取疾病标识信息和药品标识信息共同对应的标识特征向量，以便根据提取的标识特征向量对患者病历中的药品进行质控分析。
103、将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量。
其中，预设属性特征提取模型可以为第一预设DNN神经网络模型，该第一预设DNN神经网络模型能够处理结构化数据，对于本申请实施例，将结构化的疾病属性信息和药品属性信息输入至第一预设DNN神经网络模型进行特征提取，该第一预设DNN神经网络模型包含两层隐藏层，用于提取疾病属性信息和药品属性信息共同对应的属性特征向量，第二层隐藏层会输出提取的属性特征向量，以便与标识特征向量相结合，确定患者用药的合理性。由此能够扩展机器学习模型的特征空间，提升模型的质控分析精度，与此同时，本申请实施例通过采用预设文本卷积神经网络模型对文本形式的药品标识信息和疾病标识信息进行特征提取，以及采用第一预设DNN神经网络模型对结构化的药品属性信息和疾病属性信息进行特征提取，解决了同时处理结构化数据和非结构化数据的问题，提高了模型对数据的可兼容性。
104、根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
其中，该药品的质控分析结果包括患者用药异常和患者用药正常两种结果，对于本申请实施例，为了根据标识特征向量和属性特征向量确定患者病历中药品的质控分析结果，将提取的标识特征向量和属性特征向量相结合，具体可以将标识特征向量和属性特征向量合并，根据合并后的向量确定患者病历中药品的质控分析结果，为了确保质控分析结果的准确性，需要对质控分析结果进行验证，所述方法还包括：利用预设异常检测规则对所述疾病标识信息和所述药品标识信息进行异常检测，得到异常检测结果；根据所述异常检测结果验证所述质控分析结果的准确性，其中，该预设异常检测规则为根据历史异常检测结果提取的规则，例如，根据历史异常检测结果，确定疾病A，不能使用药品B，由此能够获取多条异常检测规则，利用预设异常检测规则对患者病历中的疾病标识信息和药品标识信息进行异常检测，得到异常检测结果，以便根据该异常检测结果验证质控分析结果的准确性，如果该质控分析结果为用药异常，且患者病历中的疾病标识信息和药品标识信息符合预设异常检测规则，则确定该质控分析结果无误，患者用药确实存在异常，此外，如果该质控分析结果为患者用药正常，且患者病历中的疾病标识信息和药品标识信息符合该预设异常检测规则，则确定该质控分析结果有误，患者用药有可能存在异常；若患者病历中的疾病标识信息和药品标识信息不符合该预设异常检测规则，则确定该质控分析结果无误，患者用药确实正常，由此通过利用该预设异常检测规则能够对该质控分析结果进行验证，确保质控分析结果的准确性。
本申请实施例提供的一种基于机器学习的药品质控分析方法，与目前使用医生所开处方中的疾病名称和药品名称对药品进行质量监控的方式相比，本申请能够获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；并将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；同时将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；最终根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果，由此在使用疾病标识信息和药品标识信息的基础上，引入疾病属性信息和药品属性信息，扩展了机器学习模型的特征空间，使机器学习模型能够获取更多的特征信息，提高了模型的质控分析精度，此外，本申请通过采用预设标识特征提取模型对文本形式的药品标识信息和疾病标识信息进行特征提取，以及采用预设属性特征提取模型对结构化的药品属性信息和疾病属性信息进行特征提取，解决了同时处理结构化数据和非结构化数据的问题，提高了模型对数据的可兼容性。
进一步的，为了更好的说明上述质控分析结果的确定过程，作为对上述实施例的细化和扩展，本申请实施例提供了另一种基于机器学习的药品质控分析方法，如图2所示，所述方法包括：
201、获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息。
对于本申请实施例，为了确定疾病标识信息对应的疾病属性信息和药品标识信息对应的药品属性信息，步骤201具体包括：根据所述疾病标识信息查询预设结构化疾病属性表，确定所述疾病标识信息对应的疾病属性信息；根据所述药品标识信息查询预设结构化药品属性表，确定所述药品标识信息对应的药品属性信息。其中，所述预设结构化疾病属性表中存储有不同疾病标识信息及其对应的疾病属性信息，所述预设结构化药品属性表中存储有不同药品标识信息及其对应的药品属性信息。
具体地，当服务器接收到药品质控分析指令时，对该质控分析指令中携带的待分析的患者病历进行质控分析，首先获取患者病历中记载的疾病名称和药品名称，由于该疾病名称和药品名称可能是某种疾病或者药品的别名，并不是标准化名称，为了便于对患者病历中记载的所有疾病名称和药品名称进行统一处理，需要对获取的疾病名称和药品名称进行标准化处理，具体地，可以根据获取的疾病名称和药品名称，分别查询预设疾病名称库和预设药品名称库，确定患者病历中所记载疾病和药品的标准化名称，即疾病标识信息和药品标识信息，其中，预设疾病名称库中存储有不同疾病的标准化名称及其对应的别名，预设药品名称库中存储有不同药品的标准化名称及其对应的别名，由此通过该预设疾病名称库和预设药品名称库，能够对患者病历中记载的疾病名称和药品名称进行标准化处理，得到文本形式的标准化的疾病名称和标准化的药品名称。
进一步地，为了获取疾病标识信息对应的疾病属性信息和药品标识信息对应的药品属性信息，根据该疾病标识信息查询预设结构化疾病属性表，确定该疾病标识信息对应的结构化疾病属性信息，同理根据该药品标识信息查询预设结构化药品属性表，确定该药品标识信息对应的结构化药品属性信息，由此能够在使用疾病名称和药品名称的基础上，引入疾病属性信息和药品属性信息，扩展了机器学习模型的特征空间，使机器学习模型能够获取更多的特征信息，提高模型的质控分析精度。
202、将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量。
对于本申请实施例，疾病标识信息和药品标识信息为文本形式的一维数据，为了能够对文本形式的疾病标识信息和药品标识信息进行质控分析，预设标识特征提取模型可以为预设文本卷积神经网络模型，该预设文本卷积神经网络由嵌入层、卷积层、池化层、拼接层和输出层组成，具体地，若疾病标识信息和药品标识信息中标准化疾病名称和标准化药品名称共有m种，则某一标准化疾病名称或药品名称的独热向量表示为长度为m、与该名称对应位置值为1且其余位置值为0的一条向量。在经过嵌入层的映射后，该向量则会转化成长度为n且n远小于m的嵌入向量。相比于独热向量的0-1表示，该向量的每个位置的值都为一个连续的小数。卷积层主要用多个不同大小的卷积核对输入该层的嵌入向量做卷积运算，其公式如下所示：
c i＝f(w·x i:i+h-1+b)
其中，ci表示以输入嵌入向量的第i行为起点卷积核进行卷积运算的结果，f表示卷积运算，w表示卷积核中的参数，Xi:i+h-1表示以输入嵌入向量的第i到i+h-1行(h为卷积核的大小)，b为运算的偏置项，即在该步骤中卷积核会选取标准化疾病名称或标准化药品名称中长度为h的一部分参与运算，其意义就在于提取这部分中各名称间的相互关系。
进一步地，在卷积层输出多个向量ci之后，池化层会将这部分向量进行拼接并进行池化操作，拼接和池化操作的公式分别如下所示：
c＝[c 1,c 2,…,c n-h+1]
Figure PCTCN2020119063-appb-000001
在该步操作中，由于取了向量c的最大值，所以池化层可获得卷积层各个输出值中的值最大，此外，由于一个文本卷积神经网络模型中一般会有多个卷积核及其对应的池化层， 所以拼接层会将其得到的多个
Figure PCTCN2020119063-appb-000002
向量进行拼接，即将获得的信息进行整合，输出层的公式如下所示：
y＝softmax(w·x+b)
在线性运算的基础上使用softmax激活函数引入非线性因子获得最终各类别的概率作为输出，但在本申请中仅选取池化层的输出作为标识特征向量，用来表示预设文本卷积神经网络模型从文本形式的标准化疾病名称和标准化药品名称中获取到的信息和隐含特征。
203、将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量。
对于本申请实施例，在引入疾病属性信息和药品属性信息之后，利用预设属性特征提取模型对疾病属性信息和药品属性信息进行特征提取，该预设属性特征提取模型可以为第一预设DNN神经网络模型，利用第一预设DNN神经网络模型提取疾病属性信息和药品属性信息共同对应的属性特征向量的具体过程与步骤103相同，在此不再赘述。
204、将所述标识特征向量和所述属性特征向量进行合并，得到合并后的特征向量。
对于本申请实施例，为了利用标识特征向量和属性特征向量对患者病历中的药品进行质控分析，利用拼接层concatenate将提取的标识特征向量和属性特征向量进行拼接，由此能够将文本形式的标准化疾病名称和标准化药品名称共同对应的隐含特征与结构化的疾病属性信息和药品属性信息共同对应的隐含特征相融合，得到拼接后的特征向量，以便根据该拼接后的特征向量进行质控分析。
205、将合并后的特征向量输入至预设质控分析模型进行质控分析，得到所述患者病历中药品的质控分析结果。
进一步地，为了确定患者病历中药品的质控分析结果，将拼接后的特征输入至预设质控分析模型进行分析，得到该患者病历中药品的质控分析结果，即该患者病历中是否存在用药异常的情况，其中，预设质控分析模型可以为第二预设DNN神经网络模型，第二预设DNN神经网络模型的输出层与预设文本卷积神经网络模型中的输出层相同，其对拼接后的特征向量进行计算，得到患者病历中用药是否异常的结论。由此，通过引入疾病属性信息和药品属性信息，扩展了模型的特征空间，提高了模型的质控分析精度，此外，本申请实施例通过采用预设标识特征提取模型、预设属性特征提取模型和预设质控分析模型，解决了同时处理结构化数据和非机构化数据的问题，提高了模型对数据的可兼容性。
进一步地，在利用上述方法对患者病历中药品进行质控分析之前，需要构建预设标识特征提取模型、预设属性特征提取模型和预设质控分析模型，具体地，将上述三个模型作为一个整体进行训练，获取各个历史患者病历中的历史疾病标识信息和历史药品标识信息，以及所述各个历史患者病历对应的质控分析结果，之后分别确定所述历史疾病标识信息对应的历史疾病属性信息和所述历史药品标识信息对应的历史药品属性信息，将所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果作为样本训练集，利用预设神经网络算法对样本训练集进行训练，构建构建所述预设标识特征提取模型、所述预设属性特征提取模型和所述预设质控分析模型。如图3中的虚线框所示，将三个模型作为共同的任务一进行使用，即通过调用任务一，引 入了结构化的疾病属性信息和药品属性信息，扩展了质控分析模型的特征空间，增加了药品质控分析的信息维度，从而提高了质控分析结果的准确性。
进一步地，由于任务一中的模型是针对有限的数据进行训练得到，即任务一模型能够分析的疾病和药品的种类有限，若患者电子病历中出现的疾病和药品种类不在任务一种模型的训练集内，会导致任务一无法支持患者病历中的疾病和药品种类，从而无法对该患者病历中的药品进行质控分析，为了克服上述缺陷，如图3中的实线框所示，本申请实施例可调用任务二，利用疾病属性信息和药品属性信息分析推导该电子病历中用药是否异常，例如，任务一中的模型虽然不支持某种疾病或者药品，但是该种疾病和药品对应的疾病属性信息和药品属性信息可能与训练集中支持的疾病和药品种类对应的疾病属性和药品属性相似，由此本申请实施例可以利用训练集中已有的疾病属性信息和药品属性信息分析推导患者用药的合理性。
由此针对上述内容，为了将任务一与任务二相结合对药品进行质控分析，所述方法还包括：若所述预设标识特征提取模型不支持所述疾病标识信息或所述药品标识信息，则将所述疾病属性信息和所述药品属性信息输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量，并根据所述属性特征向量确定所述患者病历中药品的质控分析结果；若所述预设标识特征提取模型支持所述疾病标识信息和所述药品标识信息，则触发所述将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量的步骤。
具体判断预设标识特征提取模型是否支持疾病标识信息或者药品标识信息时，若待分析患者病历中的疾病标识信息或者药品标识信息不为样本训练集中已有的疾病标识信息和药品标识信息，或者将待分析患者病历中的疾病标识信息和药品标识信息输入至预设标识特征提取模型，同时将疾病属性信息和药品属性信息输入至预设属性特征提取模型后，预设质控分析模型无法输出质控分析结果，则确定预设标识特征提取模型不支持该疾病标识信息或药品标识信息；若疾病标识信息或者药品标识信息为样本训练集中已有的疾病标识信息和药品标识信息，或者将待分析患者病历中的疾病标识信息和药品标识信息输入至预设标识特征提取模型，同时将疾病属性信息和药品属性信息输入至预设属性特征提取模型后，预设质控分析模型能够输出质控分析结果，确定预设标识特征提取模型支持该疾病标识信息和药品标识信息。
对于本申请实施例，在对任务一和任务二中的模型进行训练时，将所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病特征信息、所述历史药品属性信息和所述质控分析结果作为第一样本训练集；将所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果作为第二样本训练集；利用预设神经网络算法对所述第一样本训练集进行训练，构建所述预设标识特征提取模型、所述初步属性特征提取模型和所述预设质控分析模型；根据所述第二样本训练集对所述初步属性特征提取模型进行训练，构建所述预设属性特征提取模型。
由此可知，在上述训练过程中，第一预设DNN神经网络模型共训练两次，并将两次训练之后的参数作为最终的参数进行应用，因此第一预设神经网络模型不仅在任务一种能够提取属性特征向量，还能够根据提取的属性特征向量确定患者病历中药品的质控结果，即第一预设DNN神经网络模型既能够支持任务一，同时也能够支持任务二，本申请实施例将任务二作为任务一的子任务，这样当输入的疾病标识信息和药品标识信息在任务一模 型支持的范围内时，则采用任务一种的模型进行质控分析，使用疾病标识信息和药品标识信息，以及疾病属性信息和药品属性信息两部分内容；而当任务一的模型不支持疾病标识信息或药品标识信息时，则调用任务二完成质控分析，只输入疾病属性信息和药品属性信息，由此模型的分析范围和扩展性都可得到提高，该模型不仅能够对支持的数据进行精确分析，还能够在不支持数据时，利用疾病属性信息和药品属性信息进行精确分析。
本申请实施例提供的另一种基于机器学习的药品质控分析方法，与目前使用医生所开处方中的疾病名称和药品名称对药品进行质量监控的方式相比，本申请在使用疾病名称和药品名称的基础上，引入疾病属性信息和药品属性信息，通过提取疾病名称和药品名称共同对应的标识特征向量，以及疾病属性信息和药品属性信息共同对应的属性特征向量，并将提取的标识特征向量和属性特征向量相结合，扩展了机器学习模型的特征空间，使机器学习模型能够获取更多的特征信息，提高了模型的质控分析精度，此外，本申请通过采用预设标识特征提取模型、预设属性特征提取模型和预设质控分析模型，解决了同时处理结构化数据和非结构化数据的问题，提高了模型对数据的可兼容性，与此同时，本申请采集多任务相结合的质控分析，当任务一模型不支持疾病标识信息或药品标识信息时，调用任务二模型，利用疾病属性信息和药品属性信息分析推导患者用药的合理性，由此该模型不仅能够对支持的数据进行精确分析，还能够利用疾病属性信息和药品属性信息对不支持的数据进行精确分析，提高了模型的分析范围和扩展性。
进一步地，作为图1的具体实现，本申请实施例提供了一种基于机器学习的药品质控分析装置，如图4所示，所述装置包括：获取单元31、第一提取单元32、第二提取单元33和确定单元33。
所述获取单元31，可以用于获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息。所述获取单元31是本装置中获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息的主要功能模块。
所述第一提取单元32，可以用于将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量。所述第一提取单元32是本装置中将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量的主要功能模块，也是核心模块。
所述第二提取单元33，可以用于将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量。所述第二提取单元33是本装置中将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量的主要功能模块，也是核心模块。
所述确定单元34，可以用于根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。所述确定单元34是本装置中根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果的主要功能模块，也是核心模块。
对于本申请实施例，如图5所示，为了利用标识特征向量和属性特征向量进行质控分析，所述确定单元34，包括：合并模块341和分析模块342。
所述合并模块341，可以用于将所述标识特征向量和所述属性特征向量进行合并，得到合并后的特征向量。
所述分析模块342，可以用于将合并后的特征向量输入至预设质控分析模型进行质控分析，得到所述患者病历中药品的质控分析结果。
进一步地，为了调用多任务对患者病历中的药品进行质控分析，所述装置还包括：输出单元35。
所述输出单元35，可以用于若所述预设标识特征提取模型不支持所述疾病标识信息或所述药品标识信息，则将所述疾病属性信息和所述药品属性信息输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量，并根据所述属性特征向量确定所述患者病历中药品的质控分析结果。
所述第一提取单元32，还可以用于若所述预设标识特征提取模型支持所述疾病标识信息和所述药品标识信息，则触发所述将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量的步骤。
进一步地，为了获取疾病标识信息对应的疾病属性信息和药品标识信息对应的药品属性信息，所述获取单元31，包括：第一确定模块311和第二确定模块312。
所述第一确定模块311，可以用于根据所述疾病标识信息查询预设结构化疾病属性表，确定所述疾病标识信息对应的疾病属性信息，所述预设结构化疾病属性表中存储有不同疾病标识信息及其对应的疾病属性信息。
所述第二确定模块312，可以用于根据所述药品标识信息查询预设结构化药品属性表，确定所述药品标识信息对应的药品属性信息，所述预设结构化药品属性表中存储有不同药品标识信息及其对应的药品属性信息。
进一步地，为了构建预设标识特征提取模型、预设属性特征提取和预设质控分析模型，所述装置还包括：构建单元36。
所述获取单元31，还可以用于获取各个历史患者病历中的历史疾病标识信息和历史药品标识信息，以及所述各个历史患者病历对应的质控分析结果。
所述确定单元34，还可以用于分别确定所述历史疾病标识信息对应的历史疾病属性信息和所述历史药品标识信息对应的历史药品属性信息。
所述确定单元34，还可以用于根据所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果，确定样本训练集。
所述构建单元36，可以用于利用预设神经网络算法对所述样本训练集进行训练，构建所述预设标识特征提取模型、所述预设属性特征提取模型和所述预设质控分析模型。。
进一步地，为了构建任务一和任务二中的模型，所述构建单元36，包括：训练模块361和构建模块362。
所述确定单元34，具体可以用于将所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果作为第一样本训练集；将所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果作为第二样本训练集。
所述训练模块361，可以用于利用预设神经网络算法对所述第一样本训练集进行训练，构建所述预设标识特征提取模型、所述初步属性特征提取模型和所述预设质控分析模型。
所述构建模块362，可以用于根据所述第二样本训练集对所述初步属性特征提取模型进行训练，构建所述预设属性特征提取模型。
进一步地，为了对质控分析结果进行验证，所述装置还包括：检测单元37和验证单元38。
所述检测单元37，可以用于利用预设异常检测规则对所述疾病标识信息和所述药品标识信息进行异常检测，得到异常检测结果。
所述验证单元38，可以用于根据所述异常检测结果验证所述质控分析结果的准确性。
需要说明的是，本申请实施例提供的一种基于机器学习的药品质控分析装置所涉及各功能模块的其他相应描述，可以参考图1所示方法的对应描述，在此不再赘述。
基于上述如图1所示方法，相应的，本申请实施例还提供了一种计算机可读存储介质，所述计算机可读存储介质可以是易失性，也可以是非易失性，其上存储有计算机程序，该程序被处理器执行时实现以下步骤：获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
基于上述如图1所示方法和如图4所示装置的实施例，本申请实施例还提供了一种计算机设备的实体结构图，如图6所示，该计算机设备包括：处理器41、存储器42、及存储在存储器42上并可在处理器上运行的计算机程序，其中存储器42和处理器41均设置在总线43上所述处理器41执行所述程序时实现以下步骤：获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
通过本申请的技术方案，本申请在使用疾病名称和药品名称的基础上，引入疾病属性信息和药品属性信息，通过提取疾病名称和药品名称共同对应的标识特征向量，以及疾病属性信息和药品属性信息共同对应的属性特征向量，并将提取的标识特征向量和属性特征向量相结合，扩展了机器学习模型的特征空间，使机器学习模型能够获取更多的特征信息， 提高了模型的质控分析精度，此外，本申请通过采用预设标识特征提取模型、预设属性特征提取模型和预设质控分析模型，解决了同时处理结构化数据和非结构化数据的问题，提高了模型对数据的可兼容性，与此同时，本申请采集多任务相结合的质控分析，当任务一模型不支持疾病标识信息或药品标识信息时，调用任务二模型，利用疾病属性信息和药品属性信息分析推导患者用药的合理性，由此该模型不仅能够对支持的数据进行精确分析，还能够利用疾病属性信息和药品属性信息对不支持的数据进行精确分析，提高了模型的分析范围和扩展性。
显然，本领域的技术人员应该明白，上述的本申请的各模块或各步骤可以用通用的计算装置来实现，它们可以集中在单个的计算装置上，或者分布在多个计算装置所组成的网络上，可选地，它们可以用计算装置可执行的程序代码来实现，从而，可以将它们存储在存储装置中由计算装置来执行，并且在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。这样，本申请不限制于任何特定的硬件和软件结合。
以上所述仅为本申请的优选实施例而已，并不用于限制本申请，对于本领域的技术人员来说，本申请可以有各种更改和变化。凡在本申请的精神和原则之内，所作的任何修改、等同替换、改进等，均应包括在本申请的保护范围之内。

Claims (20)

  1. 一种基于机器学习的药品质控分析方法，其中，包括：
    获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；
    将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；
    将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；
    根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
  2. 根据权利要求1所述的方法，其中，所述根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果，包括：
    将所述标识特征向量和所述属性特征向量进行合并，得到合并后的特征向量；
    将合并后的特征向量输入至预设质控分析模型进行质控分析，得到所述患者病历中药品的质控分析结果。
  3. 根据权利要求1所述的方法，其中，在所述分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息之后，所述方法还包括：
    若所述预设标识特征提取模型不支持所述疾病标识信息或所述药品标识信息，则将所述疾病属性信息和所述药品属性信息输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量，并根据所述属性特征向量确定所述患者病历中药品的质控分析结果；
    若所述预设标识特征提取模型支持所述疾病标识信息和所述药品标识信息，则触发所述将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量的步骤。
  4. 根据权利要求1所述的方法，其中，所述分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息，包括：
    根据所述疾病标识信息查询预设结构化疾病属性表，确定所述疾病标识信息对应的疾病属性信息，所述预设结构化疾病属性表中存储有不同疾病标识信息及其对应的疾病属性信息；
    根据所述药品标识信息查询预设结构化药品属性表，确定所述药品标识信息对应的药品属性信息，所述预设结构化药品属性表中存储有不同药品标识信息及其对应的药品属性信息。
  5. 根据权利要求2所述的方法，其中，在所述获取患者病历中的疾病标识信息和药品标识信息之前，所述方法还包括：
    获取各个历史患者病历中的历史疾病标识信息和历史药品标识信息，以及所述各个历史患者病历对应的质控分析结果；
    分别确定所述历史疾病标识信息对应的历史疾病属性信息和所述历史药品标识信息对应的历史药品属性信息；
    根据所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果，确定样本训练集；
    利用预设神经网络算法对所述样本训练集进行训练，构建所述预设标识特征提取模型、所述预设属性特征提取模型和所述预设质控分析模型。
  6. 根据权利要求5所述的方法，其中，所述样本训练集包括第一样本训练集和第二样本训练集，所述根据所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属 性信息、所述历史药品属性信息和所述质控分析结果，确定样本训练集，包括：
    将所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果作为第一样本训练集；
    将所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果作为第二样本训练集；
    所述利用预设神经网络算法对所述样本训练集进行训练，构建所述预设标识特征提取模型、所述预设属性特征提取模型和所述预设质控分析模型，包括：
    利用预设神经网络算法对所述第一样本训练集进行训练，构建所述预设标识特征提取模型、所述初步属性特征提取模型和所述预设质控分析模型；
    根据所述第二样本训练集对所述初步属性特征提取模型进行训练，构建所述预设属性特征提取模型。
  7. 根据权利要求1-6任一项所述的方法，其中，在所述根据所述第一特征向量和所述第二特征向量确定所述患者病历中药品的质控分析结果之后，所述方法还包括：
    利用预设异常检测规则对所述疾病标识信息和所述药品标识信息进行异常检测，得到异常检测结果；
    根据所述异常检测结果验证所述质控分析结果的准确性。
  8. 一种基于机器学习的药品质控分析装置，其中，包括：
    获取单元，用于获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；
    第一提取单元，用于将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；
    第二提取单元，用于将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；
    确定单元，用于根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
  9. 一种计算机设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其中，所述计算机程序被处理器执行时实现如下步骤：
    获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；
    将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；
    将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；
    根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
  10. 根据权利要求9所述的计算机设备，其中，所述根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果，包括：
    将所述标识特征向量和所述属性特征向量进行合并，得到合并后的特征向量；
    将合并后的特征向量输入至预设质控分析模型进行质控分析，得到所述患者病历中药品的质控分析结果。
  11. 根据权利要求9所述的计算机设备，其中，在所述分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息之后，所述计算机程序被处理器执行时还实现如下步骤：
    若所述预设标识特征提取模型不支持所述疾病标识信息或所述药品标识信息，则将所述疾病属性信息和所述药品属性信息输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量，并根据所述属性特征向量确定所述患者病历中药品的质控分析结果；
    若所述预设标识特征提取模型支持所述疾病标识信息和所述药品标识信息，则触发所述将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量的步骤。
  12. 根据权利要求9所述的计算机设备，其中，所述分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息，包括：
    根据所述疾病标识信息查询预设结构化疾病属性表，确定所述疾病标识信息对应的疾病属性信息，所述预设结构化疾病属性表中存储有不同疾病标识信息及其对应的疾病属性信息；
    根据所述药品标识信息查询预设结构化药品属性表，确定所述药品标识信息对应的药品属性信息，所述预设结构化药品属性表中存储有不同药品标识信息及其对应的药品属性信息。
  13. 根据权利要求10所述的计算机设备，其中，在所述获取患者病历中的疾病标识信息和药品标识信息之前，所述计算机程序被处理器执行时还实现如下步骤：
    获取各个历史患者病历中的历史疾病标识信息和历史药品标识信息，以及所述各个历史患者病历对应的质控分析结果；
    分别确定所述历史疾病标识信息对应的历史疾病属性信息和所述历史药品标识信息对应的历史药品属性信息；
    根据所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果，确定样本训练集；
    利用预设神经网络算法对所述样本训练集进行训练，构建所述预设标识特征提取模型、所述预设属性特征提取模型和所述预设质控分析模型。
  14. 根据权利要求12所述的计算机设备，其中，所述样本训练集包括第一样本训练集和第二样本训练集，所述根据所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果，确定样本训练集，包括：
    将所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果作为第一样本训练集；
    将所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果作为第二样本训练集；
    所述利用预设神经网络算法对所述样本训练集进行训练，构建所述预设标识特征提取模型、所述预设属性特征提取模型和所述预设质控分析模型，包括：
    利用预设神经网络算法对所述第一样本训练集进行训练，构建所述预设标识特征提取模型、所述初步属性特征提取模型和所述预设质控分析模型；
    根据所述第二样本训练集对所述初步属性特征提取模型进行训练，构建所述预设属性特征提取模型。
  15. 根据权利要求9-14任一项所述的计算机设备，其中，在所述根据所述第一特征向量和所述第二特征向量确定所述患者病历中药品的质控分析结果之后，所述计算机程序被处理器执行时还实现如下步骤：
    利用预设异常检测规则对所述疾病标识信息和所述药品标识信息进行异常检测，得到异常检测结果；
    根据所述异常检测结果验证所述质控分析结果的准确性。
  16. 一种计算机可读存储介质，其上存储有计算机程序，其中，所述计算机程序被处 理器执行时实现如下步骤：
    获取患者病历中的疾病标识信息和药品标识信息，并分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息；
    将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量；
    将所述疾病属性信息和所述药品属性信息特征输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量；
    根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果。
  17. 根据权利要求16所述的计算机可读存储介质，其中，所述根据所述标识特征向量和所述属性特征向量确定所述患者病历中药品的质控分析结果，包括：
    将所述标识特征向量和所述属性特征向量进行合并，得到合并后的特征向量；
    将合并后的特征向量输入至预设质控分析模型进行质控分析，得到所述患者病历中药品的质控分析结果。
  18. 根据权利要求16所述的计算机可读存储介质，其中，在所述分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息之后，所述计算机程序被处理器执行时还实现如下步骤：
    若所述预设标识特征提取模型不支持所述疾病标识信息或所述药品标识信息，则将所述疾病属性信息和所述药品属性信息输入至预设属性特征提取模型进行特征提取，得到所述疾病属性信息和所述药品属性信息共同对应的属性特征向量，并根据所述属性特征向量确定所述患者病历中药品的质控分析结果；
    若所述预设标识特征提取模型支持所述疾病标识信息和所述药品标识信息，则触发所述将所述疾病标识信息和所述药品标识信息输入至预设标识特征提取模型进行特征提取，得到所述疾病标识信息和所述药品标识信息共同对应的标识特征向量的步骤。
  19. 根据权利要求16所述的计算机可读存储介质，其中，所述分别确定所述疾病标识信息对应的疾病属性信息和所述药品标识信息对应的药品属性信息，包括：
    根据所述疾病标识信息查询预设结构化疾病属性表，确定所述疾病标识信息对应的疾病属性信息，所述预设结构化疾病属性表中存储有不同疾病标识信息及其对应的疾病属性信息；
    根据所述药品标识信息查询预设结构化药品属性表，确定所述药品标识信息对应的药品属性信息，所述预设结构化药品属性表中存储有不同药品标识信息及其对应的药品属性信息。
  20. 根据权利要求17所述的计算机可读存储介质，其中，在所述获取患者病历中的疾病标识信息和药品标识信息之前，所述计算机程序被处理器执行时还实现如下步骤：
    获取各个历史患者病历中的历史疾病标识信息和历史药品标识信息，以及所述各个历史患者病历对应的质控分析结果；
    分别确定所述历史疾病标识信息对应的历史疾病属性信息和所述历史药品标识信息对应的历史药品属性信息；
    根据所述历史疾病标识信息、所述历史药品标识信息、所述历史疾病属性信息、所述历史药品属性信息和所述质控分析结果，确定样本训练集；
    利用预设神经网络算法对所述样本训练集进行训练，构建所述预设标识特征提取模型、所述预设属性特征提取模型和所述预设质控分析模型。
PCT/CN2020/119063 2020-07-16 2020-09-29 基于机器学习的药品质控分析方法、装置、设备及介质 WO2021151302A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010685223.1A CN111833984B (zh) 2020-07-16 2020-07-16 基于机器学习的药品质控分析方法、装置、设备及介质
CN202010685223.1 2020-07-16

Publications (1)

Publication Number Publication Date
WO2021151302A1 true WO2021151302A1 (zh) 2021-08-05

Family

ID=72924113

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/119063 WO2021151302A1 (zh) 2020-07-16 2020-09-29 基于机器学习的药品质控分析方法、装置、设备及介质

Country Status (2)

Country Link
CN (1) CN111833984B (zh)
WO (1) WO2021151302A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116300622A (zh) * 2023-03-22 2023-06-23 北京市永康药业有限公司 一种用于药品生产的出料智能监控方法及系统
CN117995346A (zh) * 2024-04-07 2024-05-07 北京惠每云科技有限公司 病历质控优化方法、装置、电子设备及存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112700168B (zh) * 2021-01-14 2021-12-07 北京赛而生物药业有限公司 一种质检胶囊类药品的方法和装置
CN116631573A (zh) * 2023-07-25 2023-08-22 讯飞医疗科技股份有限公司 一种处方用药审核方法、装置、设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104615753A (zh) * 2015-02-13 2015-05-13 杜雨阳 获取药品与疾病之间适用关系的方法和系统
CN104933324A (zh) * 2015-07-10 2015-09-23 庞健 处方用药识别方法及系统
CN109801694A (zh) * 2018-12-18 2019-05-24 北京仁泽健康服务中心 一种疾病用药方案智能管控方法及系统
CN109920508A (zh) * 2018-12-28 2019-06-21 安徽省立医院 处方审核方法及系统
CN110444267A (zh) * 2019-07-24 2019-11-12 卓尔智联(武汉)研究院有限公司 基于电子病例系统的药物监测装置、方法及可读存储介质

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110444288A (zh) * 2019-07-24 2019-11-12 卓尔智联(武汉)研究院有限公司 基于复杂网络的辅助诊断装置、方法及可读存储介质
CN110880361B (zh) * 2019-10-16 2023-02-28 平安科技（深圳）有限公司 一种个性化精准用药推荐方法及装置
CN111191020B (zh) * 2019-12-27 2023-09-22 江苏省人民医院（南京医科大学第一附属医院） 基于机器学习和知识图谱的处方推荐方法和系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104615753A (zh) * 2015-02-13 2015-05-13 杜雨阳 获取药品与疾病之间适用关系的方法和系统
CN104933324A (zh) * 2015-07-10 2015-09-23 庞健 处方用药识别方法及系统
CN109801694A (zh) * 2018-12-18 2019-05-24 北京仁泽健康服务中心 一种疾病用药方案智能管控方法及系统
CN109920508A (zh) * 2018-12-28 2019-06-21 安徽省立医院 处方审核方法及系统
CN110444267A (zh) * 2019-07-24 2019-11-12 卓尔智联(武汉)研究院有限公司 基于电子病例系统的药物监测装置、方法及可读存储介质

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116300622A (zh) * 2023-03-22 2023-06-23 北京市永康药业有限公司 一种用于药品生产的出料智能监控方法及系统
CN116300622B (zh) * 2023-03-22 2024-01-26 北京市永康药业有限公司 一种用于药品生产的出料智能监控方法及系统
CN117995346A (zh) * 2024-04-07 2024-05-07 北京惠每云科技有限公司 病历质控优化方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
CN111833984B (zh) 2023-06-20
CN111833984A (zh) 2020-10-27

Similar Documents

Publication Publication Date Title
WO2021151302A1 (zh) 基于机器学习的药品质控分析方法、装置、设备及介质
CN112071425B (zh) 一种数据处理方法、装置、计算机设备及存储介质
US20180322954A1 (en) Method and device for constructing medical knowledge graph and assistant diagnosis method
CN109887596A (zh) 基于知识图谱的慢阻肺疾病诊断方法、装置和计算机设备
CN111863267B (zh) 数据信息获取方法、数据分析方法、装置以及存储介质
JP6907831B2 (ja) コンテキストベースの患者類似性の方法及び装置
CN108899064A (zh) 电子病历生成方法、装置、计算机设备和存储介质
WO2022267678A1 (zh) 视频问诊方法、装置、设备及存储介质
CN110069779B (zh) 医疗文本的症状实体识别方法及相关装置
WO2020172607A1 (en) Systems and methods for using deep learning to generate acuity scores for critically ill or injured patients
CN111627512A (zh) 一种相似病历的推荐方法、装置，电子设备及存储介质
CN112507138B (zh) 专病知识图谱构建方法及装置、介质及电子设备
CN111883251A (zh) 医疗误诊检测方法、装置、电子设备及存储介质
CN112530550A (zh) 影像报告生成方法、装置、计算机设备和存储介质
WO2021008601A1 (zh) 一种医学数据的检验方法
CN112447270A (zh) 一种用药推荐方法、装置、设备及存储介质
WO2022227171A1 (zh) 关键信息提取方法、装置、电子设备及介质
JP7007469B2 (ja) 医療文書作成支援装置、方法およびプログラム、学習済みモデル、並びに学習装置、方法およびプログラム
KR102461647B1 (ko) 뇌파검사보고서에 대한 의료 유형정보를 제공하는 인공지능 시스템 및 그 구현방법
US20230377697A1 (en) System and a way to automatically monitor clinical trials - virtual monitor (vm) and a way to record medical history
Permanasari et al. A web-based decision support system of patient time prediction using iterative dichotomiser 3 algorithm
CN112711579A (zh) 医疗数据的质量检测方法及装置、存储介质及电子设备
CN113470775A (zh) 信息采集方法、装置、设备及存储介质
CN111798948B (zh) 病历信息处理方法、装置、设备和存储介质
CN110910980A (zh) 一种脓毒血症的预警装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20916436

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 10/03/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20916436

Country of ref document: EP

Kind code of ref document: A1