WO2021073277A1 - 一种个性化精准用药推荐方法及装置 - Google Patents
一种个性化精准用药推荐方法及装置 Download PDFInfo
- Publication number
- WO2021073277A1 WO2021073277A1 PCT/CN2020/112186 CN2020112186W WO2021073277A1 WO 2021073277 A1 WO2021073277 A1 WO 2021073277A1 CN 2020112186 W CN2020112186 W CN 2020112186W WO 2021073277 A1 WO2021073277 A1 WO 2021073277A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- patient
- drug
- information
- patients
- feature information
- Prior art date
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
- 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
- G16H10/00—ICT specially adapted for the handling or processing of patient-related medical or healthcare data
- G16H10/60—ICT specially adapted for the handling or processing of patient-related medical or healthcare data for patient-specific data, e.g. for electronic patient records
-
- 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
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
Definitions
- This application relates to the field of artificial intelligence technology, and in particular to a method and device for personalized and accurate medication recommendation.
- Precision medicine is the core component of precision medicine. In the process of disease diagnosis and treatment, the accurate medication plan and the on-time and proper medication process play a decisive role in improving the patient's therapeutic effect. In order to prescribe the right medicine, it needs to be different from person to person and from disease to disease, and comprehensively consider the patient's current condition, past medical history, medication history, and family medical history.
- EHR Electronic Health Record
- the embodiments of the present application provide a personalized accurate medication recommendation method and device to solve the problem of low accuracy in medication medication of patients in the prior art, and difficulty in meeting patients' personalized medication needs.
- a personalized and accurate medication recommendation method includes:
- the named entity recognition algorithm obtains the patient’s medication information from the text data of each patient; the drug-based collaborative filtering algorithm selects the medication information of multiple historical patients to obtain the target patient’s medication information.
- the first drug recommendation result; the patient’s medical record data is merged to obtain the patient’s condition feature information; the patient-based collaborative filtering algorithm selects the target from the multiple historical patients At least one similar patient whose current condition feature information of the patient is similar; generating a second drug recommendation result according to the medication information of the similar patient; performing fusion processing on the first drug recommendation result and the second drug recommendation result , Obtain the personalized medicine recommendation result of the target patient.
- a personalized accurate medication recommendation device comprising:
- the acquiring unit is used to acquire medical record data of multiple patients suffering from the same disease, the medical record data including structured data, text data, and image data, wherein the patients include historical patients and those who currently need to be recommended for medication Target patients; identification unit, used to use named entity recognition algorithm to obtain the patient’s medication information from the text data of each patient; the first screening unit, used for drug-based collaborative filtering algorithm from multiple The first drug recommendation result of the target patient is obtained by screening the medication information of the historical patient; the processing unit is configured to merge the medical record data of the patient to obtain the characteristic information of the patient's condition; The second screening unit is used to screen out at least one similar patient that is similar to the current condition feature information of the target patient from the multiple historical patients based on the patient-based collaborative filtering algorithm; the generating unit is used to select according to The medication information of the similar patients generates a second drug recommendation result; the fusion unit is used to perform fusion processing on the first drug recommendation result and the second drug recommendation result to obtain the personalized medicine of the target patient Recommended results.
- a computer non-volatile storage medium includes a stored program, and when the program runs, the device where the storage medium is located is controlled to perform the following steps:
- the named entity recognition algorithm obtains the patient’s medication information from the text data of each patient; the drug-based collaborative filtering algorithm selects the medication information of multiple historical patients to obtain the target patient’s medication information.
- the first drug recommendation result; the patient’s medical record data is merged to obtain the patient’s condition feature information; the patient-based collaborative filtering algorithm selects the target from the multiple historical patients At least one similar patient whose current condition feature information of the patient is similar; generating a second drug recommendation result according to the medication information of the similar patient; performing fusion processing on the first drug recommendation result and the second drug recommendation result , Obtain the personalized medicine recommendation result of the target patient.
- a computer device including a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor executes all When the computer program is described, the following steps are implemented:
- the named entity recognition algorithm obtains the patient’s medication information from the text data of each patient; the drug-based collaborative filtering algorithm selects the medication information of multiple historical patients to obtain the target patient’s medication information.
- the first drug recommendation result; the patient’s medical record data is merged to obtain the patient’s condition feature information; the patient-based collaborative filtering algorithm selects the target from the multiple historical patients At least one similar patient whose current condition feature information of the patient is similar; generating a second drug recommendation result according to the medication information of the similar patient; performing fusion processing on the first drug recommendation result and the second drug recommendation result , Obtain the personalized medicine recommendation result of the target patient.
- This program considers the dynamic change process of the patient's condition more comprehensively, provides the accuracy of medication, and meets the patient's personalized medication needs.
- FIG. 1 is a flowchart of an optional personalized and precise medication recommendation method provided by an embodiment of the present application
- FIG. 2 is a schematic diagram of an optional personalized precise medication recommendation device provided by an embodiment of the present application.
- Fig. 3 is a schematic diagram of an optional computer device provided by an embodiment of the present application.
- the technical solution of this application can be applied to the fields of artificial intelligence, big data and/or digital medical technology, and the data involved can be stored in a database, or can be distributed storage through a blockchain, which is not limited by this application.
- Fig. 1 is a flowchart of a method for personalized accurate medication recommendation according to an embodiment of the present application. As shown in Fig. 1, the method includes:
- Step S01 Obtain medical record data of multiple patients suffering from the same disease.
- the medical record data includes structured data, text data, and image data.
- patients include historical patients and target patients who currently need to be recommended for medication;
- Step S02 using a named entity recognition algorithm to obtain the patient's medication information from the text data of each patient;
- Step S03 the drug-based collaborative filtering algorithm selects the drug use information of multiple historical patients to obtain the first drug recommendation result of the target patient;
- Step S04 merge the patient's medical record data to obtain the patient's condition characteristic information
- step S05 the patient-based collaborative filtering algorithm selects at least one similar patient that is similar to the current condition feature information of the target patient from multiple historical patients;
- Step S06 generating a second drug recommendation result based on the medication information of similar patients
- step S07 the first drug recommendation result and the second drug recommendation result are fused to obtain the personalized drug recommendation result of the target patient.
- Step S01 Obtain medical record data of multiple patients suffering from the same disease.
- the medical record data includes structured data, text data, and image data.
- the patients include historical patients and target patients who currently need to be recommended for medication.
- the structured data is the numerical data in the patient’s Electronic Health Record (EHR), such as heart rate, blood pressure, blood sugar, urine volume and other test data, which are stored in the electronic health record in numerical form.
- EHR Electronic Health Record
- Text data such as medication records, discharge summary, nursing records, ward round records, etc., are stored in electronic health records in text form, and image data such as CT images, MRI images, X-ray images, etc., are stored in electronic health in the form of pictures File.
- step S02 a named entity recognition algorithm is used to obtain the patient's medication information from the text data of each patient.
- the text data may be, for example, a medication record, which may be a scanned image file or a text format.
- the text data of each patient is divided into words to obtain multiple words; the named entity recognition algorithm is used to identify each patient's medication information from multiple words.
- the medication information of each patient is expressed in the form of a patient-drug coding matrix.
- Named Entity Recognition (Named Entity Recognition; hereinafter referred to as: NER) refers to the recognition of entities with specific meanings in the text, mainly including names of persons, names of diseases, names of drugs, and/or proper nouns.
- a named entity can be used to identify the patient’s social security card number, such as social security card number: 6123456.
- the social security card number is the patient’s unique identification code
- the patient’s social security card number is used to represent the patient
- the drug code is based on The mapping relationship between drugs and codes recorded in the drug knowledge base can be queried and obtained by inputting the identified drug name into the drug knowledge base.
- the information integration to obtain patient A’s medication information is ⁇ 6123456, D120, D130 ⁇ .
- the medication information can also be obtained from text data by means such as keyword recognition, which is not limited here.
- step S03 the drug-based collaborative filtering algorithm selects the drug use information of multiple historical patients to obtain the first drug recommendation result of the target patient.
- the drug-based collaborative filtering algorithm refers to finding the scores of certain drugs by historical patients by searching for the similarity between drugs and drugs, and then recommending several similar drugs with the highest scores to the target patients.
- the first drug recommendation result includes at least one drug.
- the drugs included in the first drug recommendation result are drugs with a score higher than a preset value by filtering from the medication information of multiple historical patients suffering from the same disease.
- Historical patients refer to patients who have visited a doctor who have the same disease as the target patient.
- collaborative filtering is a method of predicting the medications of target patients by collecting medication information from many patients. For example, if drug a is used to treat a certain disease with a high score, the system will consider that drug a is suitable for the treatment of the disease.
- step S04 the patient's medical record data is merged to obtain the patient's condition characteristic information.
- the patient's condition feature information includes first feature information, second feature information, and third feature information.
- the condition feature information is a piece of multi-dimensional record data, which is obtained by combining the first feature information, the second feature information, and the third feature information of each patient based on the time dimension information.
- the first characteristic information is extracted from the text data of the medical record data
- the second characteristic information is extracted from the structured data of the medical record data
- the third characteristic information is extracted from the image data of the medical record data.
- step S04 specifically includes:
- step S041 the medication information of each identified patient is converted into corresponding numerical data according to the preset mapping table. Understandably, the drug is mapped to a value to facilitate subsequent calculations, such as mapping "nifedipine” to "D130", mapping “levoamlodipine besylate” to “D131” and so on.
- step S042 the structured data in the medical record data of each patient is formed into a sparse matrix according to the time sequence.
- a Variational Autoencoder VAE is used to process the structured data and compress it into a sparse matrix according to the timing information.
- VAE is a kind of self-encoder, and the coding result of VAE can reduce the dimensionality of high-dimensional data.
- Step S043 Use a variational autoencoder to compress and encode the numerical data and the sparse matrix to obtain the first code and the second code of the patient.
- the first code includes first feature information and time dimension information derived from text data.
- the second code includes second feature information and time dimension information derived from structured data.
- time dimension information here is not only the date of treatment, but the time used to indicate the course of the disease, such as the time of the first visit, the time of the first follow-up visit, and the time of the second follow-up visit, so as to determine the stage of the patient's condition.
- step S044 the image data is pooled using a preset convolutional neural network to obtain a third code of the patient.
- the pooling process first obtain the three primary color values of each pixel in the image data; then use the preset convolutional neural network to extract the characteristic part according to the three primary color values of all pixels to form the third code of the patient.
- the convolutional neural network is used to pool it to highlight the key information, such as the area of pneumonia.
- the convolutional neural network outputs a third code, which includes the third feature information and time dimension information derived from the image data .
- the patchy fuzzy area of the lower right lung in the lung X-ray of patient a on 2019-9-30 is 1cm*2cm; the right side of the lung X-ray of patient a on 2019-10-02
- the patchy fuzzy area of the lower lung is 0.5cm*0.5cm.
- the third code of the patient includes the third feature information and time dimension information derived from the image data.
- step S045 the first code, the second code, and the third code are combined to obtain disease characteristic information of each patient. Specifically, the first feature information, the second feature information, and the third feature information of each patient are combined and processed based on the time dimension information to obtain the condition feature information of each patient.
- the condition feature information is a piece of multi-dimensional record data. .
- step S05 the patient-based collaborative filtering algorithm selects at least one similar patient that is similar to the current condition feature information of the target patient from multiple historical patients.
- the patient-based collaborative filtering algorithm refers to finding the similarity of the feature information between the patient and the patient, and then confirming the patient with the highest similarity of the feature information as a similar patient. For example, if the similarity between the condition characteristic information of one patient at a certain stage and the current condition characteristic information of the target patient among multiple historical patients is 92%, which is greater than a preset threshold of 90%, then the patient is a similar patient.
- step S05 The specific steps of step S05 include:
- Step S051 Input the condition feature information of each patient into the preset word vector representation model to obtain the feature information vector of each patient.
- the preset word vector representation model can be, for example, the Word2vea model.
- other word vector representation models can also be used, so that the multi-dimensional record data can be converted into word vectors to facilitate subsequent similarity calculations. .
- Step S052 Calculate the Euclidean distance between the feature information vector of each historical patient and the feature information vector of the target patient. Understandably, by calculating the similarity of each patient’s condition, we can find patients who are similar to the target patient’s current condition. Because each person’s disease course changes, physical fitness, etc., are different, so when giving medication recommendations, it should be sufficient Consider individual differences.
- step S053 the Euclidean distance is confirmed as the degree of similarity between the historical patient and the target patient.
- x represents the feature information vector of the condition of the historical patient
- y represents the feature information vector of the condition of the target patient
- d(x,y) represents the Euclidean distance between the vector x and the vector y
- n represents the total number of dimensions of the vector .
- step S054 at least one similar patient is screened from multiple historical patients according to the similarity of the condition, where the similarity of the condition between the similar patient and the target patient is greater than a preset threshold.
- the preset threshold is 90%. If the condition similarity between the historical patient Jia and the target patient is 92%, then the historical patient Jia can be confirmed as a similar patient to the target patient.
- Step S06 Generate a second drug recommendation result based on the medication information of similar patients.
- the medication records of similar patients are first obtained from a preset database, and the medication records of similar patients are segmented to obtain multiple vocabulary; the named entity recognition algorithm is used to recognize the medication records from the multiple vocabularies. Medication information for similar patients.
- the database here can be a hospital's case database. For example, if the medication information of similar patient B is ⁇ 6123457, D120, D130 ⁇ , then the second drug recommendation results are D120 and D130.
- Step S06 specifically includes:
- Step S061 find the disease-related diseases from the preset disease-drug directed connection graph; step S062, obtain the related diseases and the medication information of the diseases according to the directed connection graph; step S063, according to the related diseases and the medication information of the diseases And the medication information of similar patients to generate a second drug recommendation result.
- the disease-drug directed connection graph uses NER mining to mark the association between the disease and the corresponding drug preparation (for example, statins can be used to reduce blood lipids), forming points and edges, and fusion with the disease VS disease graph network , And finally form a directed connection graph with diseases and corresponding drugs as the vertices and the associations as edges to record the associations between diseases, such as complications.
- NER mining to mark the association between the disease and the corresponding drug preparation (for example, statins can be used to reduce blood lipids), forming points and edges, and fusion with the disease VS disease graph network .
- step S07 the first drug recommendation result and the second drug recommendation result are fused to obtain the personalized drug recommendation result of the target patient.
- the first drug recommendation result is first merged with the second drug recommendation result, and then duplicate drugs are deleted from the merged multiple drugs to obtain a personalized drug recommendation result.
- the first drug recommendation result is (x1, x2)
- the second drug recommendation result is (x2, y1, z1)
- the fusion is (x1, x2, x2, y1, z1)
- a duplicate x2 is deleted
- the recommended result of chemical medicine is (x1, x2, y1, z1).
- step S07 The specific steps of step S07 include:
- step S071 the first drug recommendation result and the second drug recommendation result are fused to obtain the fusion drug recommendation result; step S072, the fusion drug recommendation result is compared with the preset mutually exclusive drug group to determine the fusion drug recommendation Whether there is a mutually exclusive drug group in the result; step S073, if it exists, adopt the same-drug substitution strategy to adjust the fusion drug recommendation result to eliminate the mutually exclusive drug group; step S074, generate the target patient’s profile based on the adjusted fusion drug recommendation result Personalized drug recommendation results.
- chloramphenicol in antibiotics and sulfonylurea hypoglycemic agents can cause hypoglycemia. Therefore, it is a mutually exclusive drug group and cannot be taken at the same time; aspirin and indomethacin are also mutually exclusive drug groups.
- step S075 is further included. If it does not exist, the personalized medicine recommendation result of the target patient is generated according to the fusion medicine recommendation result.
- the embodiment of the present application provides a personalized accurate medication recommendation device, which is used to implement the above-mentioned personalized accurate medication recommendation method.
- the device includes: an acquisition unit 10, an identification unit 20, and a first screening unit 30.
- the acquiring unit 10 is used to acquire medical record data of multiple patients suffering from the same disease, the medical record data including structured data, text data, and image data, wherein the patients include historical patients and current medications that need to be recommended Target patients.
- the structured data is the numerical data in the patient’s Electronic Health Record (EHR), such as heart rate, blood pressure, blood sugar, urine volume and other test data, which are stored in the electronic health record in numerical form.
- EHR Electronic Health Record
- Text data such as medication records, discharge summary, nursing records, ward round records, etc., are stored in electronic health records in text form, and image data such as CT images, MRI images, X-ray images, etc., are stored in electronic health in the form of pictures File.
- the recognition unit 20 is configured to obtain the medication information of the patient from the text data of each patient by using a named entity recognition algorithm.
- the text data may be, for example, a medication record, which may be a scanned image file or a text format.
- the text data of each patient is divided into words to obtain multiple words; the named entity recognition algorithm is used to identify each patient's medication information from multiple words.
- the medication information of each patient is expressed in the form of a patient-drug coding matrix.
- Named Entity Recognition (Named Entity Recognition; hereinafter referred to as: NER) refers to the recognition of entities with specific meanings in the text, mainly including names of persons, names of diseases, names of drugs, and/or proper nouns.
- a named entity can be used to identify the patient’s social security card number, such as social security card number: 6123456.
- the social security card number is the patient’s unique identification code
- the patient’s social security card number is used to represent the patient
- the drug code is based on The mapping relationship between drugs and codes recorded in the drug knowledge base can be queried and obtained by inputting the identified drug name into the drug knowledge base.
- the information integration to obtain patient A’s medication information is ⁇ 6123456, D120, D130 ⁇ .
- the medication information can also be obtained from text data by means such as keyword recognition, which is not limited here.
- the first screening unit 30 is configured to obtain the first drug recommendation result of the target patient by screening the medication information of the multiple historical patients based on the drug-based collaborative filtering algorithm.
- the drug-based collaborative filtering algorithm refers to finding the scores of certain drugs by historical patients by searching for the similarity between drugs and drugs, and then recommending several similar drugs with the highest scores to the target patients.
- the first drug recommendation result includes at least one drug.
- the drugs included in the first drug recommendation result are drugs with a score higher than a preset value by filtering from the medication information of multiple historical patients suffering from the same disease.
- Historical patients refer to patients who have visited a doctor who have the same disease as the target patient.
- collaborative filtering is a method of predicting the medications of target patients by collecting medication information from many patients. For example, if drug a is used to treat a certain disease with a high score, the system will consider that drug a is suitable for the treatment of the disease. For example, antihypertensive drugs with better efficacy for drugs suitable for high blood pressure.
- the processing unit 40 is configured to merge the medical record data of the patient to obtain the characteristic information of the patient's condition.
- the patient's condition feature information includes first feature information, second feature information, and third feature information.
- the condition feature information is a piece of multi-dimensional record data, which is obtained by combining the first feature information, the second feature information, and the third feature information of each patient based on the time dimension information.
- the first characteristic information is extracted from the text data of the medical record data
- the second characteristic information is extracted from the structured data of the medical record data
- the third characteristic information is extracted from the image data of the medical record data.
- the second screening unit 50 is configured to screen out at least one similar patient that is similar to the current condition feature information of the target patient from the multiple historical patients based on the patient's collaborative filtering algorithm.
- the patient-based collaborative filtering algorithm refers to finding the similarity of the feature information between the patient and the patient, and then confirming the patient with the highest similarity of the feature information as a similar patient. For example, if the similarity between the condition characteristic information of one patient at a certain stage and the current condition characteristic information of the target patient among multiple historical patients is 92%, which is greater than a preset threshold of 90%, then the patient is a similar patient.
- the generating unit 60 is configured to generate a second drug recommendation result according to the medication information of the similar patients.
- the medication records of similar patients are first obtained from a preset database, and the medication records of similar patients are segmented to obtain multiple vocabulary; the named entity recognition algorithm is used to recognize the medication records from the multiple vocabularies. Medication information for similar patients.
- the database here can be a hospital's case database. For example, if the medication information of similar patient B is ⁇ 6123457, D120, D130 ⁇ , then the second drug recommendation results are D120 and D130.
- the fusion unit 70 is configured to perform fusion processing on the first drug recommendation result and the second drug recommendation result to obtain the personalized drug recommendation result of the target patient.
- the first drug recommendation result is first merged with the second drug recommendation result, and then duplicate drugs are deleted from the merged multiple drugs to obtain a personalized drug recommendation result.
- the first drug recommendation result is (x1, x2)
- the second drug recommendation result is (x2, y1, z1)
- the fusion is (x1, x2, x2, y1, z1)
- a duplicate x2 is deleted
- the recommended result of chemical medicine is (x1, x2, y1, z1).
- the processing unit 40 includes a conversion sub-unit, a processing sub-unit, a compression sub-unit, a pooling sub-unit, and a merging sub-unit.
- the transformation subunit is used to transform the medication information of each identified patient into corresponding numerical data according to the preset mapping table. Understandably, the drug is mapped to a value to facilitate subsequent calculations, such as mapping "nifedipine” to "D130", mapping “levoamlodipine besylate” to “D131” and so on.
- the processing subunit is used to form a sparse matrix according to the time sequence of the structured data in the medical record data of each patient.
- a Variational Autoencoder VAE is used to process the structured data and compress it into a sparse matrix according to the timing information.
- VAE is a kind of self-encoder, and the coding result of VAE can reduce the dimensionality of high-dimensional data.
- each variable in X represents an input vector
- the elements of the vector are factors related to the patient, such as heart rate, blood pressure,
- X represents a disease.
- the compression subunit is used to compress and encode the numerical data and the sparse matrix by using the variational autoencoder to obtain the first code and the second code of the patient.
- the first code includes first feature information and time dimension information derived from text data.
- the second code includes second feature information and time dimension information derived from structured data.
- time dimension information here is not only the date of treatment, but the time used to indicate the course of the disease, such as the time of the first visit, the time of the first follow-up visit, and the time of the second follow-up visit, so as to determine the stage of the patient's condition.
- the pooling subunit is used to pool the image data using a preset convolutional neural network to obtain the third code of the patient.
- the pooling process first obtain the three primary color values of each pixel in the image data; then use the preset convolutional neural network to extract the characteristic part according to the three primary color values of all pixels to form the third code of the patient.
- the convolutional neural network is used to pool it to highlight the key information, such as the area of pneumonia.
- the convolutional neural network outputs a third code, which includes the third feature information and time dimension information derived from the image data .
- the patchy fuzzy area of the lower right lung in the lung X-ray of patient a on 2019-9-30 is 1cm*2cm; the right side of the lung X-ray of patient a on 2019-10-02
- the patchy fuzzy area of the lower lung is 0.5cm*0.5cm.
- the third code of the patient includes the third feature information and time dimension information derived from the image data.
- the merging subunit is used for merging the first code, the second code, and the third code to obtain disease characteristic information of each patient. Specifically, the first characteristic information, the second characteristic information, and the third characteristic information of each patient are combined and processed based on the time dimension information to obtain the disease characteristic information of each patient.
- the disease characteristic information is a piece of multi-dimensional record data. .
- the second screening unit 50 includes a preprocessing subunit, a first calculation subunit, a confirmation subunit, and a screening subunit.
- the pre-processing subunit is used to input the disease feature information of each patient into the preset word vector representation model to obtain the feature information vector of each patient.
- the preset word vector representation model can be, for example, the Word2vea model.
- other word vector representation models can also be used, so that the multi-dimensional record data can be converted into word vectors to facilitate subsequent similarity calculations. .
- the first calculation subunit is used to calculate the Euclidean distance between the feature information vector of each historical patient and the feature information vector of the target patient. Understandably, by calculating the similarity of each patient’s condition, we can find patients who are similar to the target patient’s current condition. Because each person’s disease course changes, physical fitness, etc., are different, so when giving medication recommendations, it should be sufficient Consider individual differences.
- the confirmation subunit is used to confirm the Euclidean distance as the similarity between the historical patient and the target patient.
- x represents the feature information vector of the condition of the historical patient
- y represents the feature information vector of the condition of the target patient
- d(x,y) represents the Euclidean distance between the vector x and the vector y
- n represents the total number of dimensions of the vector .
- the screening subunit is used to screen out at least one similar patient from multiple historical patients according to the similarity of the condition, where the similarity between the condition of the similar patient and the target patient is greater than a preset threshold.
- the preset threshold is 90%. If the condition similarity between the historical patient Jia and the target patient is 92%, then the historical patient Jia can be confirmed as a similar patient to the target patient.
- the generation unit 60 includes a search subunit, an acquisition subunit, and a first generation subunit.
- the search subunit is used to search the disease-related diseases from the preset disease-drug directed connection graph; the acquisition subunit is used to obtain the related diseases and the medication information of the disease according to the directed connection graph; the first generation subunit , Used to generate the second drug recommendation result based on the related disease, the medication information of the disease, and the medication information of similar patients.
- the disease-drug directed connection graph uses NER mining to mark the association between the disease and the corresponding drug preparation (for example, statins can be used to reduce blood lipids), forming points and edges, and fusion with the disease VS disease graph network , And finally form a directed connection graph with diseases and corresponding drugs as the vertices and the associations as edges to record the associations between diseases, such as complications.
- NER mining to mark the association between the disease and the corresponding drug preparation (for example, statins can be used to reduce blood lipids), forming points and edges, and fusion with the disease VS disease graph network .
- the fusion unit 70 includes a fusion subunit, a comparison subunit, an adjustment subunit, and a second generation subunit.
- the fusion subunit is used to merge the first drug recommendation result with the second drug recommendation result to obtain the fusion drug recommendation result;
- the comparison subunit is used to compare the fusion drug recommendation result with the preset mutually exclusive drug group Yes, to determine whether there is a mutually exclusive drug group in the fusion drug recommendation result;
- the adjustment sub-unit is used to adjust the fusion drug recommendation result by using the same drug replacement strategy if it exists, so as to eliminate the mutually exclusive drug group;
- the second generation sub-unit It is used to generate personalized drug recommendation results for target patients based on the adjusted fusion drug recommendation results.
- chloramphenicol in antibiotics and sulfonylurea hypoglycemic agents can cause hypoglycemia. Therefore, it is a mutually exclusive drug group and cannot be taken at the same time; aspirin and indomethacin are also mutually exclusive drug groups.
- the fusion unit 70 also includes a third generation unit, which is configured to generate a personalized drug recommendation result of the target patient according to the fusion drug recommendation result if it does not exist.
- the embodiment of the present application provides a storage medium, and the storage medium includes a stored program.
- the storage medium involved can be a computer-readable storage medium.
- the storage medium such as a computer-readable storage medium, can be nonvolatile (such as a computer nonvolatile storage medium) or volatile (such as a computer volatile Storage media).
- the device where the storage medium is located is controlled to perform the following steps:
- the medical record data includes structured data, text data and image data.
- patients include historical patients and target patients who currently need to be recommended for medication; use named entity recognition algorithm
- Obtain the patient's medication information from the text data of each patient the drug-based collaborative filtering algorithm selects the first drug recommendation result of the target patient from the medication information of multiple historical patients; combines the patient's medical history data Perform merge processing to obtain the patient’s condition feature information; the patient-based collaborative filtering algorithm selects from multiple historical patients at least one similar patient that is similar to the target patient’s current condition feature information;
- the medication information generates a second drug recommendation result; the first drug recommendation result and the second drug recommendation result are fused to obtain the personalized drug recommendation result of the target patient.
- the step of controlling the device where the storage medium is located to merge the patient's medical record data to obtain the patient's condition feature information includes:
- the medication information of each identified patient is transformed into corresponding numerical data; the structured data in the medical record data of each patient is formed into a sparse matrix according to the time sequence; the use of variational auto-encoding
- the device compresses and encodes the numerical data and sparse matrix to obtain the patient's first code and second code; uses the preset convolutional neural network to pool the image data to obtain the patient's third code; The first code, the second code, and the third code are combined to obtain the disease characteristic information of each patient.
- the first code includes first feature information and time dimension information derived from text data
- the second code includes second feature information and time dimension information derived from structured data
- the third code includes information derived from image data.
- the third characteristic information and time dimension information; when the program is running, the step of controlling the device where the storage medium is located to perform the combined processing of the first code, the second code, and the third code to obtain the disease characteristic information of each patient includes:
- the condition feature information is a piece of multi-dimensional record data.
- controlling the device where the storage medium is located to execute a patient-based collaborative filtering algorithm to select at least one similar patient that is similar to the target patient’s current condition feature information from multiple historical patients includes :
- controlling the device where the storage medium is located to execute the step of generating the second drug recommendation result based on the medication information of similar patients includes:
- Fig. 3 is a schematic diagram of a computer device provided by an embodiment of the present application.
- the computer device 100 of this embodiment includes: a processor 101, a memory 102, and a computer program 103 stored in the memory 102 and running on the processor 101.
- the processor 101 executes the computer program 103 when the computer program 103 is executed.
- the personalized and accurate medication recommendation method in the example will not be repeated here.
- the function of each model/unit in the personalized accurate medication recommendation device in the embodiment is realized. In order to avoid repetition, it will not be repeated here.
- the computer device 100 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a cloud server.
- the computer device may include, but is not limited to, a processor 101 and a memory 102.
- FIG. 3 is only an example of the computer device 100 and does not constitute a limitation on the computer device 100. It may include more or less components than those shown in the figure, or a combination of certain components, or different components.
- computer equipment may also include input and output devices, network access devices, buses, and so on.
- the so-called processor 101 may be a central processing unit (Central Processing Unit, CPU), other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
- the general-purpose processor may be a microprocessor or the processor may also be any conventional processor or the like.
- the memory 102 may be an internal storage unit of the computer device 100, such as a hard disk or a memory of the computer device 100.
- the memory 102 may also be an external storage device of the computer device 100, such as a plug-in hard disk equipped on the computer device 100, a smart media card (SMC), a secure digital (SD) card, and a flash memory card (Flash). Card) and so on.
- the memory 102 may also include both an internal storage unit of the computer device 100 and an external storage device.
- the memory 102 is used to store computer programs and other programs and data required by the computer equipment.
- the memory 102 can also be used to temporarily store data that has been output or will be output.
- the disclosed system, device, and method may be implemented in other ways.
- the device embodiments described above are only illustrative.
- the division of the units is only a logical function division, and there may be other divisions in actual implementation, for example, multiple units or components may be combined. Or it can be integrated into another system, or some features can be ignored or not implemented.
- the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.
- the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
- the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
- the above-mentioned integrated unit may be implemented in the form of hardware, or may be implemented in the form of hardware plus software functional units.
- the above-mentioned integrated unit implemented in the form of a software functional unit may be stored in a computer readable storage medium.
- the above-mentioned software functional unit is stored in a storage medium and includes several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (Processor) execute the method described in each embodiment of the present application. Part of the steps.
- the aforementioned storage media include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program code .
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Public Health (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Pathology (AREA)
- Chemical & Material Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Medicinal Chemistry (AREA)
- Medical Treatment And Welfare Office Work (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
一种个性化精准用药推荐方法及装置,涉及人工智能技术领域,方法包括:获取患同一疾病的多个病患的病历数据,包括结构化数据、文本数据及影像数据(S01);从文本数据中得到病患的用药信息(S02);从多个历史病患的用药信息中筛选得到目标病患的第一药品推荐结果(S03);将病患的病历数据进行合并处理,得到病患的病情特征信息(S04);从多个历史病患中筛选出与目标病患的当前病情特征信息相似的至少一个相似病患(S05);根据相似病患的用药信息生成第二药品推荐结果(S06);根据第一药品推荐结果及第二药品推荐结果得到目标病患的个性化药品推荐结果(S07)。该方法及装置能够解决现有技术中病患用药精准度低的问题。
Description
本申请要求于2019年10月16日提交中国专利局、申请号为201910983855.3,发明名称为“一种个性化精准用药推荐方法及装置”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。
本申请涉及人工智能技术领域,尤其涉及一种个性化精准用药推荐方法及装置。
精准用药是精准医疗的核心构成。在疾病诊疗过程中,准确的用药方案,按时、按量的用药过程对提升病患的治疗效果起到了决定性作用。而要做到对症下药,需要因人、因病而异,综合考虑病患的当前情况、过往病史用药史及家族病史。
发明人意识到,现有智能系统对病患EHR(Electronic Health Record)数据的使用,一般限于对其中数值型、结构化部分的利用。而对于个体病患,尤其是患有慢性病的个体病患,其诊疗医疗记录具有跨度时间长,仅靠结构化数值型EHR数据来进行药品推荐,用药的精准度低,难以满足病患的个性化用药需求。
发明内容
有鉴于此,本申请实施例提供了一种个性化精准用药推荐方法及装置,用以解决现有技术中病患用药的精准度低,难以满足病患的个性化用药需求的问题。
为了实现上述目的,根据本申请的一个方面,提供了一种个性化精准用药推荐方法,所述方法包括:
获取患同一疾病的多个病患的病历数据,所述病历数据包括结构化数据、文本数据及影像数据,其中,所述病患包括历史病患及当前需要被推荐用药的目标病患;利用命名实体识别算法从每个所述病患的文本数据中得到所述病患的用药信息;基于药品的协同过滤算法从多个所述历史病患的用药信息中筛选得到所述目标病患的第一药品推荐结果;将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息;基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患;根据所述相似病患的用药信息生成第二药品推荐结果;将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果。
为了实现上述目的,根据本申请的一个方面,提供了一种个性化精准用药推荐装置,所述装置包括:
获取单元,用于获取患同一疾病的多个病患的病历数据,所述病历数据包括结构化数据、文本数据及影像数据,其中,所述病患包括历史病患及当前需要被推荐用药的目标病患;识别单元,用于利用命名实体识别算法从每个所述病患的文本数据中得到所述病患的用药信息;第一筛选单元,用于基于药品的协同过滤算法从多个所述历史病患的用药信息中筛选得到所述目标病患的第一药品推荐结果;处理单元,用于将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息;第二筛选单元,用于基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患;生成单元,用于根据所述相似病患的用药信息生成第二药品推荐结果;融合单元,用于将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果。
为了实现上述目的,根据本申请的一个方面,提供了一种计算机非易失性存储介质,所述存储介质包括存储的程序,在所述程序运行时控制所述存储介质所在设备执行以下步骤:
获取患同一疾病的多个病患的病历数据,所述病历数据包括结构化数据、文本数据及影像数据,其中,所述病患包括历史病患及当前需要被推荐用药的目标病患;利用命名实 体识别算法从每个所述病患的文本数据中得到所述病患的用药信息;基于药品的协同过滤算法从多个所述历史病患的用药信息中筛选得到所述目标病患的第一药品推荐结果;将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息;基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患;根据所述相似病患的用药信息生成第二药品推荐结果;将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果。
为了实现上述目的,根据本申请的一个方面,提供了一种计算机设备,包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序,所述处理器执行所述计算机程序时实现以下步骤:
获取患同一疾病的多个病患的病历数据,所述病历数据包括结构化数据、文本数据及影像数据,其中,所述病患包括历史病患及当前需要被推荐用药的目标病患;利用命名实体识别算法从每个所述病患的文本数据中得到所述病患的用药信息;基于药品的协同过滤算法从多个所述历史病患的用药信息中筛选得到所述目标病患的第一药品推荐结果;将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息;基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患;根据所述相似病患的用药信息生成第二药品推荐结果;将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果。
本方案更加综合全面的考虑病患病情的动态变化过程,提供用药的精准度,满足病患的个性化用药需求。
为了更清楚地说明本申请实施例的技术方案,下面将对实施例中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其它的附图。
图1是本申请实施例提供的一种可选的个性化精准用药推荐方法的流程图;
图2是本申请实施例提供的一种可选的个性化精准用药推荐装置的示意图;
图3是本申请实施例提供的一种可选的计算机设备的示意图。
为了更好的理解本申请的技术方案,下面结合附图对本申请实施例进行详细描述。
本申请的技术方案可应用于人工智能、大数据和/或数字医疗技术领域,涉及的数据可存储于数据库中,或者可以通过区块链分布式存储,本申请不做限定。
在本申请实施例中使用的术语是仅仅出于描述特定实施例的目的,而非旨在限制本申请。在本申请实施例和所附权利要求书中所使用的单数形式的“一种”、“所述”和“该”也旨在包括多数形式,除非上下文清楚地表示其他含义。
应当理解,本文中使用的术语“和/或”仅仅是一种描述关联对象的关联关系,表示可以存在三种关系,例如,A和/或B,可以表示:单独存在A,同时存在A和B,单独存在B这三种情况。另外,本文中字符“/”,一般表示前后关联对象是一种“或”的关系。
图1是根据本申请实施例的一种个性化精准用药推荐方法的流程图,如图1所示,该方法包括:
步骤S01,获取患同一疾病的多个病患的病历数据,病历数据包括结构化数据、文本数据及影像数据,其中,病患包括历史病患及当前需要被推荐用药的目标病患;
步骤S02,利用命名实体识别算法从每个病患的文本数据中得到病患的用药信息;
步骤S03,基于药品的协同过滤算法从多个历史病患的用药信息中筛选得到目标病患的第一药品推荐结果;
步骤S04,将病患的病历数据进行合并处理,得到病患的病情特征信息;
步骤S05,基于病患的协同过滤算法从多个历史病患中筛选出与目标病患的当前病情特征信息相似的至少一个相似病患;
步骤S06,根据相似病患的用药信息生成第二药品推荐结果;
步骤S07,将第一药品推荐结果及第二药品推荐结果进行融合处理,得到目标病患的个性化药品推荐结果。
在本方案中,通过将同一疾病的病患的病历数据进行合并处理,筛选出与目标病患的当前病情特征信息相似的病患,并进一步根据相似病患的用药记录得到第二药品推荐结果,最后将第一药品推荐结果与第二药品推荐结果进行融合,相比于现有的仅利用第一药品推荐系结果来给病患提供用药推荐,更加综合全面的考虑病患病情的动态变化过程,提供用药的精准度,满足病患的个性化用药需求。
下面对本实施例提供的图像处理方法的具体技术方案进行详细的说明。
步骤S01,获取患同一疾病的多个病患的病历数据,病历数据包括结构化数据、文本数据及影像数据,其中,病患包括历史病患及当前需要被推荐用药的目标病患。
其中,历史病患是指曾经患上述疾病的病患,目标病患是当前正在患病的病患,目标病患是用药推荐的目标对象。具体地,结构化数据为病患的电子健康档案(Electronic Health Record,EHR)中的数值型数据,例如:心率、血压、血糖、尿量等各项检验数据,其以数值形式存储在电子健康档案中。文本数据例如用药记录、出院总结、护理记录、查房记录等,其以文本形式存储在电子健康档案中,影像数据例如CT图像、MRI图像、X光影像等,其以图片形式存储在电子健康档案中。
步骤S02,利用命名实体识别算法从每个病患的文本数据中得到病患的用药信息。
其中,文本数据例如可以是用药记录,其可能是扫描图档或文本格式。具体包括:将每个病患的文本数据进行分词处理,得到多个词汇;利用命名实体识别算法从多个词汇中识别得到每个病患的用药信息。在本实施方式中,每个病患的用药信息以病患-药品编码矩阵的形式表达。命名实体识别(Named Entity Recognition;以下简称:NER)是指识别文本中具有特定意义的实体,主要包括人名、疾病名、药品名和/或专有名词等。其中,可以利用命名实体识别获取病患的社保卡号,例如社保卡号:6123456,由于社保卡号是病患的唯一身份识别码,将病患的社保卡号用来表示所述病患,药品编码是根据药品知识库中记录的药品及编码的映射关系,通过将识别到的药品名输入药品知识库,可以查询得到药品编码。
例如,病患的社保卡号:6123456,识别得到药品(药品a:D120;药品b:D130;),那么信息整合得到病患A的用药信息为{6123456,D120,D130}。
在其他实施方式中,也可以用关键词识别等手段从文本数据中获取用药信息,在此不做限定。
步骤S03,基于药品的协同过滤算法从多个历史病患的用药信息中筛选得到目标病患的第一药品推荐结果。
其中,基于药品的协同过滤算法是指,通过查找药品和药品之间的相似度,找到历史病患对某些药品的评分,进而将评分最高的若干个相似药品推荐给目标病患。第一药品推荐结果至少包括一种药品。在一种实施方式中,第一药品推荐结果中包括的药品是根据患有相同疾病的多个历史病患的用药信息中过滤得到评分高于预设值的药品。历史病患是指与目标病患患有相同疾病的已就诊的病患。
可以理解地,协同过滤是通过收集来自很多病患的用药信息来预测目标病患的用药的方法。例如药品a用来治疗某疾病的评分很高,那么系统就会认为药品a适用于该疾病的治疗应用。
步骤S04,将病患的病历数据进行合并处理,得到病患的病情特征信息。
其中,病患的病情特征信息包括第一特征信息、第二特征信息及第三特征信息。病情 特征信息是一条多维记录数据,是以时间维度信息为基准将每个病患的第一特征信息、第二特征信息及第三特征信息合并处理得到的。第一特征信息是从病历数据的文本数据中提取得到的,第二特征信息是从病历数据的结构化数据中提取得到的,第三特征性是从病历数据的影像数据中提取得到的。
进一步地,步骤S04具体包括:
步骤S041,根据预设的映射表将识别到的每个病患的用药信息转变为对应的数值型数据。可以理解地,将药品映射为一个数值,方便后续计算处理,例如将“硝苯地平”映射为“D130”,将“苯磺酸左旋氨氯地平”映射为“D131”等等。
步骤S042,将每个病患的病历数据中的结构化数据根据时间顺序形成稀疏矩阵。具体地,利用变分自编码器(Variational Autoencoder,VAE)来处理结构化数据,使其按照时序信息压缩成稀疏矩阵。VAE是一种自编码器,VAE的编码结果可以将高维数据进行降维。每个病患的结构化数据例如Xvae={x1,x2,…xi,…xn},X中的每一个变量代表一个输入向量,向量的元素是与病患相关的因子,如心率、血压、实验检验肌酐、血糖、尿素,X表示一种疾病。
步骤S043,利用变分自编码器对数值型数据、稀疏矩阵进行压缩编码处理,得到病患的第一编码及第二编码。第一编码包括来源于文本数据的第一特征信息及时间维度信息,同样地,第二编码包括来源于结构化数据的第二特征信息及时间维度信息。
进一步地,这里时间维度信息不仅仅是治疗日期,而是用于表示病程的时间,例如初诊时间,第一次复诊时间,第二次复诊时间,从而判断病患的病情所处的阶段。
步骤S044,利用预设的卷积神经网络将影像数据进行池化处理,得到病患的第三编码。
池化处理时,先获取影像数据中的每个像素点的三原色值;再利用预设的卷积神经网络并根据所有像素点的三原色值提取特征部分,形成病患的第三编码。
具体地,通过卷积神经网络将其池化,突出其中的重点信息,例如肺炎的区域,卷积神经网络输出第三编码,第三编码包括来源于影像数据的第三特征信息及时间维度信息。例如,病患a于2019-9-30的肺部X光片中的右下肺斑片状模糊面积为1cm*2cm;病患a于2019-10-02的肺部X光片中的右下肺斑片状模糊面积为0.5cm*0.5cm。病患的第三编码包括来源于影像数据的第三特征信息及时间维度信息。
步骤S045,将第一编码、第二编码及第三编码进行合并处理,得到每个病患的病情特征信息。具体地,以时间维度信息为基准将每个病患的第一特征信息、第二特征信息及第三特征信息合并处理,得到每个病患的病情特征信息,病情特征信息是一条多维记录数据。
步骤S05,基于病患的协同过滤算法从多个历史病患中筛选出与目标病患的当前病情特征信息相似的至少一个相似病患。
其中,基于病患的协同过滤算法是指,通过查找病患和病患之间的病情特征信息的相似度,进而将病情特征信息的相似度最高的病患确认为相似病患。例如,多个历史病患中有一个病患某一阶段的病情特征信息与目标病患当前病情特征信息的相似度为92%,大于预设阈值90%,那么该病患就是相似病患。
可以理解地,通过筛选相似病患,突显目标病患当前病情的特征重点,更加综合全面地考虑病人病情的动态变化过程,使得用药推荐更加精准,更加贴合病患当前的病情。
步骤S05的具体步骤包括:
步骤S051,将每个病患的病情特征信息输入预设的词向量表示模型,得到每个病患的特征信息向量。在本实施方式中,预设的词向量表示模型例如可以是Word2vea模型,在其他实施方式中,也可以采用其他的词向量表示模型,使得多维记录数据转变为词向量,方便后续的相似度计算。
步骤S052,计算每个历史病患的特征信息向量与目标病患的特征信息向量之间的欧式 距离。可以理解地,通过计算各个病患的病情相似度,从而找到与目标病患的当前病情相似的病患,因为每个人的病程变化、体质等都不相同,因此在给予用药推荐时,应该充分考虑个体的差异性。
步骤S053,将欧式距离确认为历史病患与目标病患的病情相似度。
在其他实施方式中,也可以采取其他的相似度计算方法来计算历史病患与目标病患之间病情的相似度,例如余弦距离、编辑距离等,在此不做限定。
步骤S054,根据病情相似度从多个历史病患中筛选出至少一个相似病患,其中,相似病患与目标病患的病情相似度大于预设阈值。
例如预设阈值为90%,如果历史病患贾某与目标病患的病情相似度为92%,那么就可以将历史病患贾某确认为是目标病患的相似病患。
步骤S06,根据相似病患的用药信息生成第二药品推荐结果。
在一种实施方式中,先从预设的数据库中获取相似病患的用药记录,将相似病患的用药记录进行分词处理,得到多个词汇;利用命名实体识别算法从多个词汇中识别得到相似病患的用药信息。这里的数据库可以是医院的病例数据库。例如,相似病患B的用药信息{6123457,D120,D130},那么第二药品推荐结果为D120及D130。
步骤S06具体包括:
步骤S061,从预设的疾病-药品的有向连接图中查找疾病的关联疾病;步骤S062,根据有向连接图获取关联疾病及疾病的用药信息;步骤S063,根据关联疾病、疾病的用药信息及相似病患的用药信息生成第二药品推荐结果。
在本实施方式中,疾病-药品的有向连接图,是通过NER挖掘标注出疾病与对应药物制剂的关联(如他汀可用于降低血脂),形成点和边,和疾病VS疾病图网络进行融合,最终形成以疾病、对应药物为顶点,之间关联为边的有向连接图,以记录疾病之间的关联,例如并发症等。
步骤S07,将第一药品推荐结果及第二药品推荐结果进行融合处理,得到目标病患的个性化药品推荐结果。
在一种实施方式中,第一药品推荐结果先与第二药品推荐结果进行合并,然后在合并后的多个药品中删除重复的药品,得到个性化药品推荐结果。
例如,第一药品推荐结果为(x1,x2),第二药品推荐结果为(x2,y1,z1),融合得到(x1,x2,x2,y1,z1),删除一个重复的x2,最后个性化药品推荐结果为(x1,x2,y1,z1)。
步骤S07具体步骤包括:
步骤S071,将第一药品推荐结果与第二药品推荐结果进行融合处理,得到融合药品推荐结果;步骤S072,将融合药品推荐结果与预设的互斥药品组进行比对,以判断融合药品推荐结果中是否存在互斥药品组;步骤S073,如存在,采用同药性药品替换策略调整融合药品推荐结果,以消除互斥药品组;步骤S074,根据调整后的融合药品推荐结果生成目标病患的个性化药品推荐结果。
例如抗生素中的“氯霉素”与磺脲类降血糖药,合用会引起低血糖。因此是互斥药品组,不能同时服用;阿司匹林与消炎痛也是互斥药品组。
进一步地,在步骤S072之后,还包括步骤S075,如不存在,根据所述融合药品推荐结果生成所述目标病患的个性化药品推荐结果。
在本方案中,通过将同一疾病的病患的病历数据进行合并处理,筛选出与目标病患的当前病情特征信息相似的病患,并进一步根据相似病患的用药记录得到第二药品推荐结果,最后将第一药品推荐结果与第二药品推荐结果进行融合,相比于现有的仅利用第一药品推荐系结果来给病患提供用药推荐,更加综合全面的考虑病患病情的动态变化过程,提供用药的精准度,满足病患的个性化用药需求。
本申请实施例提供了一种个性化精准用药推荐装置,该装置用于执行上述个性化精准用药推荐方法,如图2所示,该装置包括:获取单元10、识别单元20、第一筛选单元30、处理单元40、第二筛选单元50、生成单元60及融合单元70。
获取单元10,用于获取患同一疾病的多个病患的病历数据,所述病历数据包括结构化数据、文本数据及影像数据,其中,所述病患包括历史病患及当前需要被推荐用药的目标病患。
其中,历史病患是指曾经患上述疾病的病患,目标病患是当前正在患病的病患,目标病患是用药推荐的目标对象。具体地,结构化数据为病患的电子健康档案(Electronic Health Record,EHR)中的数值型数据,例如:心率、血压、血糖、尿量等各项检验数据,其以数值形式存储在电子健康档案中。文本数据例如用药记录、出院总结、护理记录、查房记录等,其以文本形式存储在电子健康档案中,影像数据例如CT图像、MRI图像、X光影像等,其以图片形式存储在电子健康档案中。
识别单元20,用于利用命名实体识别算法从每个所述病患的文本数据中得到所述病患的用药信息。
其中,文本数据例如可以是用药记录,其可能是扫描图档或文本格式。具体包括:将每个病患的文本数据进行分词处理,得到多个词汇;利用命名实体识别算法从多个词汇中识别得到每个病患的用药信息。在本实施方式中,每个病患的用药信息以病患-药品编码矩阵的形式表达。命名实体识别(Named Entity Recognition;以下简称:NER)是指识别文本中具有特定意义的实体,主要包括人名、疾病名、药品名和/或专有名词等。其中,可以利用命名实体识别获取病患的社保卡号,例如社保卡号:6123456,由于社保卡号是病患的唯一身份识别码,将病患的社保卡号用来表示所述病患,药品编码是根据药品知识库中记录的药品及编码的映射关系,通过将识别到的药品名输入药品知识库,可以查询得到药品编码。
例如,病患的社保卡号:6123456,识别得到药品(药品a:D120;药品b:D130;),那么信息整合得到病患A的用药信息为{6123456,D120,D130}。
在其他实施方式中,也可以用关键词识别等手段从文本数据中获取用药信息,在此不做限定。
第一筛选单元30,用于基于药品的协同过滤算法从多个所述历史病患的用药信息中筛选得到所述目标病患的第一药品推荐结果。其中,基于药品的协同过滤算法是指,通过查找药品和药品之间的相似度,找到历史病患对某些药品的评分,进而将评分最高的若干个相似药品推荐给目标病患。第一药品推荐结果至少包括一种药品。在一种实施方式中,第一药品推荐结果中包括的药品是根据患有相同疾病的多个历史病患的用药信息中过滤得到评分高于预设值的药品。历史病患是指与目标病患患有相同疾病的已就诊的病患。
可以理解地,协同过滤是通过收集来自很多病患的用药信息来预测目标病患的用药的方法。例如药品a用来治疗某疾病的效果评分很高,那么系统就会认为药品a适用于该疾病的治疗应用。例如,适用于高血压的药物疗效较好的降压药。
处理单元40,用于将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息。其中,病患的病情特征信息包括第一特征信息、第二特征信息及第三特征信息。病情 特征信息是一条多维记录数据,是以时间维度信息为基准将每个病患的第一特征信息、第二特征信息及第三特征信息合并处理得到的。第一特征信息是从病历数据的文本数据中提取得到的,第二特征信息是从病历数据的结构化数据中提取得到的,第三特征性是从病历数据的影像数据中提取得到的。
第二筛选单元50,用于基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患。其中,基于病患的协同过滤算法是指,通过查找病患和病患之间的病情特征信息的相似度,进而将病情特征信息的相似度最高的病患确认为相似病患。例如,多个历史病患中有一个病患某一阶段的病情特征信息与目标病患当前病情特征信息的相似度为92%,大于预设阈值90%,那么该病患就是相似病患。
可以理解地,通过筛选相似病患,突显目标病患当前病情的特征重点,更加综合全面地考虑病人病情的动态变化过程,使得用药推荐更加精准,更加贴合病患当前的病情。
生成单元60,用于根据所述相似病患的用药信息生成第二药品推荐结果。
在一种实施方式中,先从预设的数据库中获取相似病患的用药记录,将相似病患的用药记录进行分词处理,得到多个词汇;利用命名实体识别算法从多个词汇中识别得到相似病患的用药信息。这里的数据库可以是医院的病例数据库。例如,相似病患B的用药信息{6123457,D120,D130},那么第二药品推荐结果为D120及D130。
融合单元70,用于将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果。
在一种实施方式中,第一药品推荐结果先与第二药品推荐结果进行合并,然后在合并后的多个药品中删除重复的药品,得到个性化药品推荐结果。
例如,第一药品推荐结果为(x1,x2),第二药品推荐结果为(x2,y1,z1),融合得到(x1,x2,x2,y1,z1),删除一个重复的x2,最后个性化药品推荐结果为(x1,x2,y1,z1)。
在本方案中,通过将同一疾病的病患的病历数据进行合并处理,筛选出与目标病患的当前病情特征信息相似的病患,并进一步根据相似病患的用药记录得到第二药品推荐结果,最后将第一药品推荐结果与第二药品推荐结果进行融合,相比于现有的仅利用第一药品推荐系结果来给病患提供用药推荐,更加综合全面的考虑病患病情的动态变化过程,提供用药的精准度,满足病患的个性化用药需求。
可选地,处理单元40包括转换子单元、处理子单元、压缩子单元、池化子单元及合并子单元。
转变子单元,用于根据预设的映射表将识别到的每个病患的用药信息转变为对应的数值型数据。可以理解地,将药品映射为一个数值,方便后续计算处理,例如将“硝苯地平”映射为“D130”,将“苯磺酸左旋氨氯地平”映射为“D131”等等。
处理子单元,用于将每个病患的病历数据中的结构化数据根据时间顺序形成稀疏矩阵。具体地,利用变分自编码器(Variational Autoencoder,VAE)来处理结构化数据,使其按照时序信息压缩成稀疏矩阵。VAE是一种自编码器,VAE的编码结果可以将高维数据进行降维。每个病患的结构化数据例如Xvae={x1,x2,…xi,…xn},X中的每一个变量代表一个输入向量,向量的元素是与病患相关的因子,如心率、血压、实验检验肌酐、血糖、尿素,X表示一种疾病。
压缩子单元,用于利用变分自编码器对数值型数据、稀疏矩阵进行压缩编码处理,得到病患的第一编码及第二编码。第一编码包括来源于文本数据的第一特征信息及时间维度信息,同样地,第二编码包括来源于结构化数据的第二特征信息及时间维度信息。
进一步地,这里时间维度信息不仅仅是治疗日期,而是用于表示病程的时间,例如初 诊时间,第一次复诊时间,第二次复诊时间,从而判断病患的病情所处的阶段。
池化子单元,用于利用预设的卷积神经网络将影像数据进行池化处理,得到病患的第三编码。
池化处理时,先获取影像数据中的每个像素点的三原色值;再利用预设的卷积神经网络并根据所有像素点的三原色值提取特征部分,形成病患的第三编码。
具体地,通过卷积神经网络将其池化,突出其中的重点信息,例如肺炎的区域,卷积神经网络输出第三编码,第三编码包括来源于影像数据的第三特征信息及时间维度信息。例如,病患a于2019-9-30的肺部X光片中的右下肺斑片状模糊面积为1cm*2cm;病患a于2019-10-02的肺部X光片中的右下肺斑片状模糊面积为0.5cm*0.5cm。病患的第三编码包括来源于影像数据的第三特征信息及时间维度信息。
合并子单元,用于将第一编码、第二编码及第三编码进行合并处理,得到每个病患的病情特征信息。具体地,以时间维度信息为基准将每个病患的第一特征信息、第二特征信息及第三特征信息合并处理,得到每个病患的病情特征信息,病情特征信息是一条多维记录数据。
可选地,第二筛选单元50包括预处理子单元、第一计算子单元、确认子单元及筛选子单元。
预处理子单元,用于将每个病患的病情特征信息输入预设的词向量表示模型,得到每个病患的特征信息向量。在本实施方式中,预设的词向量表示模型例如可以是Word2vea模型,在其他实施方式中,也可以采用其他的词向量表示模型,使得多维记录数据转变为词向量,方便后续的相似度计算。
第一计算子单元,用于计算每个历史病患的特征信息向量与目标病患的特征信息向量之间的欧式距离。可以理解地,通过计算各个病患的病情相似度,从而找到与目标病患的当前病情相似的病患,因为每个人的病程变化、体质等都不相同,因此在给予用药推荐时,应该充分考虑个体的差异性。
确认子单元,用于将欧式距离确认为历史病患与目标病患的病情相似度。
在其他实施方式中,也可以采取其他的相似度计算方法来计算历史病患与目标病患之间病情的相似度,例如余弦距离、编辑距离等,在此不做限定。
筛选子单元,用于根据病情相似度从多个历史病患中筛选出至少一个相似病患,其中,相似病患与目标病患的病情相似度大于预设阈值。
例如预设阈值为90%,如果历史病患贾某与目标病患的病情相似度为92%,那么就可以将历史病患贾某确认为是目标病患的相似病患。
可选地,生成单元60包括查找子单元、获取子单元、第一生成子单元。
查找子单元,用于从预设的疾病-药品的有向连接图中查找疾病的关联疾病;获取子单元,用于根据有向连接图获取关联疾病及疾病的用药信息;第一生成子单元,用于根据关联疾病、疾病的用药信息及相似病患的用药信息生成第二药品推荐结果。
在本实施方式中,疾病-药品的有向连接图,是通过NER挖掘标注出疾病与对应药物制剂的关联(如他汀可用于降低血脂),形成点和边,和疾病VS疾病图网络进行融合,最终形成以疾病、对应药物为顶点,之间关联为边的有向连接图,以记录疾病之间的关联,例 如并发症等。
可选地,融合单元70包括融合子单元、比对子单元、调整子单元、第二生成子单元。
融合子单元,用于将第一药品推荐结果与第二药品推荐结果进行融合处理,得到融合药品推荐结果;比对子单元,用于将融合药品推荐结果与预设的互斥药品组进行比对,以判断融合药品推荐结果中是否存在互斥药品组;调整子单元,用于如存在,采用同药性药品替换策略调整融合药品推荐结果,以消除互斥药品组;第二生成子单元,用于根据调整后的融合药品推荐结果生成目标病患的个性化药品推荐结果。
例如抗生素中的“氯霉素”与磺脲类降血糖药,合用会引起低血糖。因此是互斥药品组,不能同时服用;阿司匹林与消炎痛也是互斥药品组。
融合单元70还包括第三生成单元,用于如不存在,根据所述融合药品推荐结果生成所述目标病患的个性化药品推荐结果。
在本方案中,通过将同一疾病的病患的病历数据进行合并处理,筛选出与目标病患的当前病情特征信息相似的病患,并进一步根据相似病患的用药记录得到第二药品推荐结果,最后将第一药品推荐结果与第二药品推荐结果进行融合,相比于现有的仅利用第一药品推荐系结果来给病患提供用药推荐,更加综合全面的考虑病患病情的动态变化过程,提供用药的精准度,满足病患的个性化用药需求。
本申请实施例提供了一种存储介质,存储介质包括存储的程序。涉及的存储介质可以是计算机可读存储介质,该存储介质如计算机可读存储介质可以是非易失性的(如计算机非易失性存储介质),也可以是易失性的(如计算机易失性存储介质)。
其中,在程序运行时控制存储介质所在设备执行以下步骤:
获取患同一疾病的多个病患的病历数据,病历数据包括结构化数据、文本数据及影像数据,其中,病患包括历史病患及当前需要被推荐用药的目标病患;利用命名实体识别算法从每个病患的文本数据中得到病患的用药信息;基于药品的协同过滤算法从多个历史病患的用药信息中筛选得到目标病患的第一药品推荐结果;将病患的病历数据进行合并处理,得到病患的病情特征信息;基于病患的协同过滤算法从多个历史病患中筛选出与目标病患的当前病情特征信息相似的至少一个相似病患;根据相似病患的用药信息生成第二药品推荐结果;将第一药品推荐结果及第二药品推荐结果进行融合处理,得到目标病患的个性化药品推荐结果。
可选地,在程序运行时控制存储介质所在设备执行将病患的病历数据进行合并处理,得到病患的病情特征信息的步骤,包括:
根据预设的映射表将识别到的每个病患的用药信息转变为对应的数值型数据;将每个病患的病历数据中的结构化数据根据时间顺序形成稀疏矩阵;利用变分自编码器对数值型数据、稀疏矩阵进行压缩编码处理,得到病患的第一编码及第二编码;利用预设的卷积神经网络将影像数据进行池化处理,得到病患的第三编码;将第一编码、第二编码及第三编码进行合并处理,得到每个病患的病情特征信息。
可选地,第一编码包括来源于文本数据的第一特征信息及时间维度信息,第二编码包括来源于结构化数据的第二特征信息及时间维度信息,第三编码包括来源于影像数据的第三特征信息及时间维度信息;在程序运行时控制存储介质所在设备执行将第一编码、第二编码及第三编码进行合并处理,得到每个病患的病情特征信息的步骤,包括:
以时间维度信息为基准将每个病患的第一特征信息、第二特征信息及第三特征信息合并处理,得到每个病患的病情特征信息,病情特征信息是一条多维记录数据。
可选地,在程序运行时控制存储介质所在设备执行基于病患的协同过滤算法从多个历史病患中筛选出与目标病患的当前病情特征信息相似的至少一个相似病患的步骤,包括:
将每个病患的病情特征信息输入预设的词向量表示模型,得到每个病患的特征信息向 量;计算每个历史病患的特征信息向量与目标病患的特征信息向量之间的欧式距离;将欧式距离确认为历史病患与目标病患的病情相似度;根据病情相似度从多个历史病患中筛选出至少一个相似病患,其中,相似病患与目标病患的病情相似度大于预设阈值。
可选地,在程序运行时控制存储介质所在设备执行根据相似病患的用药信息生成第二药品推荐结果的步骤,包括:
从预设的疾病-药品的有向连接图中查找疾病的关联疾病;根据有向连接图获取关联疾病及疾病的用药信息;根据关联疾病、疾病的用药信息及相似病患的用药信息生成第二药品推荐结果。
图3是本申请实施例提供的一种计算机设备的示意图。如图3所示,该实施例的计算机设备100包括:处理器101、存储器102以及存储在存储器102中并可在处理器101上运行的计算机程序103,处理器101执行计算机程序103时实现实施例中的个性化精准用药推荐方法,为避免重复,此处不一一赘述。或者,该计算机程序被处理器101执行时实现实施例中个性化精准用药推荐装置中各模型/单元的功能,为避免重复,此处不一一赘述。
计算机设备100可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。计算机设备可包括,但不仅限于,处理器101、存储器102。本领域技术人员可以理解,图3仅仅是计算机设备100的示例,并不构成对计算机设备100的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件,例如计算机设备还可以包括输入输出设备、网络接入设备、总线等。
所称处理器101可以是中央处理单元(Central Processing Unit,CPU),还可以是其他通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现场可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
存储器102可以是计算机设备100的内部存储单元,例如计算机设备100的硬盘或内存。存储器102也可以是计算机设备100的外部存储设备,例如计算机设备100上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,存储器102还可以既包括计算机设备100的内部存储单元也包括外部存储设备。存储器102用于存储计算机程序以及计算机设备所需的其他程序和数据。存储器102还可以用于暂时地存储已经输出或者将要输出的数据。
所属领域的技术人员可以清楚地了解到,为描述的方便和简洁,上述描述的系统,装置和单元的具体工作过程,可以参考前述方法实施例中的对应过程,在此不再赘述。
在本申请所提供的几个实施例中,应该理解到,所揭露的系统,装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述单元的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式,例如,多个单元或组件可以结合或者可以集成到另一个系统,或一些特征可以忽略,或不执行。另一点,所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口,装置或单元的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外,在本申请各个实施例中的各功能单元可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用硬件加软件功能单元的形式实现。
上述以软件功能单元的形式实现的集成的单元,可以存储在一个计算机可读取存储介 质中。上述软件功能单元存储在一个存储介质中,包括若干指令用以使得一台计算机装置(可以是个人计算机,服务器,或者网络装置等)或处理器(Processor)执行本申请各个实施例所述方法的部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(Read-Only Memory,ROM)、随机存取存储器(Random Access Memory,RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述仅为本申请的较佳实施例而已,并不用以限制本申请,凡在本申请的精神和原则之内,所做的任何修改、等同替换、改进等,均应包含在本申请保护的范围之内。
Claims (20)
- 一种个性化精准用药推荐方法,其中,所述方法包括:获取患同一疾病的多个病患的病历数据,所述病历数据包括结构化数据、文本数据及影像数据,其中,所述病患包括历史病患及当前需要被推荐用药的目标病患;利用命名实体识别算法从每个所述病患的文本数据中得到所述病患的用药信息;基于药品的协同过滤算法从多个所述历史病患的用药信息中筛选得到所述目标病患的第一药品推荐结果;将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息;基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患;根据所述相似病患的用药信息生成第二药品推荐结果;将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果。
- 根据权利要求1所述的方法,其中,所述将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息的步骤,包括:根据预设的映射表将识别到的每个所述病患的用药信息转变为对应的数值型数据;将每个所述病患的病历数据中的结构化数据根据时间顺序形成稀疏矩阵;利用变分自编码器对所述数值型数据、所述稀疏矩阵进行压缩编码处理,得到所述病患的第一编码及第二编码;利用预设的卷积神经网络将所述影像数据进行池化处理,得到所述病患的第三编码;将所述第一编码、所述第二编码及所述第三编码进行合并处理,得到每个所述病患的病情特征信息。
- 根据权利要求2所述的方法,其中,所述第一编码包括来源于所述文本数据的第一特征信息及时间维度信息,所述第二编码包括来源于所述结构化数据的第二特征信息及所述时间维度信息,所述第三编码包括来源于所述影像数据的第三特征信息及所述时间维度信息;所述将所述第一编码、所述第二编码及所述第三编码进行合并处理,得到每个所述病患的病情特征信息的步骤,包括:以所述时间维度信息为基准将每个所述病患的所述第一特征信息、所述第二特征信息及所述第三特征信息合并处理,得到每个所述病患的病情特征信息,所述病情特征信息是一条多维记录数据。
- 根据权利要求1所述的方法,其中,所述基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患的步骤,包括:将每个所述病患的病情特征信息输入预设的词向量表示模型,得到每个所述病患的特征信息向量;计算每个所述历史病患的特征信息向量与所述目标病患的特征信息向量之间的欧式距离;将所述欧式距离确认为所述历史病患与所述目标病患的病情相似度;根据所述病情相似度从多个所述历史病患中筛选出至少一个相似病患,其中,所述相似病患与所述目标病患的病情相似度大于预设阈值。
- 根据权利要求1所述的方法,其中,所述根据所述相似病患的用药信息生成第二药品推荐结果的步骤,包括:从预设的疾病-药品的有向连接图中查找所述疾病的关联疾病;根据所述有向连接图获取所述关联疾病及所述疾病的用药信息;根据所述关联疾病、所述疾病的用药信息及所述相似病患的用药信息生成第二药品推 荐结果。
- 根据权利要求1所述的方法,其中,所述将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果的步骤,包括:将所述第一药品推荐结果与所述第二药品推荐结果进行融合处理,得到融合药品推荐结果;将所述融合药品推荐结果与预设的互斥药品组进行比对,以判断所述融合药品推荐结果中是否存在互斥药品组;如存在,采用同药性药品替换策略调整所述融合药品推荐结果,以消除所述互斥药品组;根据调整后的所述融合药品推荐结果生成所述目标病患的个性化药品推荐结果。
- 根据权利要求2中任意一项所述的方法,其中,所述利用预设的卷积神经网络将所述影像数据进行池化处理,得到所述病患的第三编码的步骤,包括:获取所述影像数据中的每个像素点的三原色值;利用预设的卷积神经网络并根据所有所述像素点的三原色值提取特征部分,形成所述病患的第三编码。
- 一种个性化精准用药推荐装置,其中,所述装置包括:获取单元,用于获取患同一疾病的多个病患的病历数据,所述病历数据包括结构化数据、文本数据及影像数据,其中,所述病患包括历史病患及当前需要被推荐用药的目标病患;识别单元,用于利用命名实体识别算法从每个所述病患的文本数据中得到所述病患的用药信息;第一筛选单元,用于基于药品的协同过滤算法从多个所述历史病患的用药信息中筛选得到所述目标病患的第一药品推荐结果;处理单元,用于将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息;第二筛选单元,用于基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患;生成单元,用于根据所述相似病患的用药信息生成第二药品推荐结果;融合单元,用于将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果。
- 一种存储介质,所述存储介质包括存储的程序,其中,在所述程序运行时控制所述存储介质所在设备执行以下步骤:获取患同一疾病的多个病患的病历数据,所述病历数据包括结构化数据、文本数据及影像数据,其中,所述病患包括历史病患及当前需要被推荐用药的目标病患;利用命名实体识别算法从每个所述病患的文本数据中得到所述病患的用药信息;基于药品的协同过滤算法从多个所述历史病患的用药信息中筛选得到所述目标病患的第一药品推荐结果;将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息;基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患;根据所述相似病患的用药信息生成第二药品推荐结果;将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果。
- 根据权利要求9所述的存储介质,其中,所述将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息的步骤,包括:根据预设的映射表将识别到的每个所述病患的用药信息转变为对应的数值型数据;将每个所述病患的病历数据中的结构化数据根据时间顺序形成稀疏矩阵;利用变分自编码器对所述数值型数据、所述稀疏矩阵进行压缩编码处理,得到所述病患的第一编码及第二编码;利用预设的卷积神经网络将所述影像数据进行池化处理,得到所述病患的第三编码;将所述第一编码、所述第二编码及所述第三编码进行合并处理,得到每个所述病患的病情特征信息。
- 根据权利要求10所述的存储介质,其中,所述第一编码包括来源于所述文本数据的第一特征信息及时间维度信息,所述第二编码包括来源于所述结构化数据的第二特征信息及所述时间维度信息,所述第三编码包括来源于所述影像数据的第三特征信息及所述时间维度信息;所述将所述第一编码、所述第二编码及所述第三编码进行合并处理,得到每个所述病患的病情特征信息的步骤,包括:以所述时间维度信息为基准将每个所述病患的所述第一特征信息、所述第二特征信息及所述第三特征信息合并处理,得到每个所述病患的病情特征信息,所述病情特征信息是一条多维记录数据。
- 根据权利要求9所述的存储介质,其中,所述基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患的步骤,包括:将每个所述病患的病情特征信息输入预设的词向量表示模型,得到每个所述病患的特征信息向量;计算每个所述历史病患的特征信息向量与所述目标病患的特征信息向量之间的欧式距离;将所述欧式距离确认为所述历史病患与所述目标病患的病情相似度;根据所述病情相似度从多个所述历史病患中筛选出至少一个相似病患,其中,所述相似病患与所述目标病患的病情相似度大于预设阈值。
- 根据权利要求9所述的存储介质,其中,所述根据所述相似病患的用药信息生成第二药品推荐结果的步骤,包括:从预设的疾病-药品的有向连接图中查找所述疾病的关联疾病;根据所述有向连接图获取所述关联疾病及所述疾病的用药信息;根据所述关联疾病、所述疾病的用药信息及所述相似病患的用药信息生成第二药品推荐结果。
- 根据权利要求9所述的存储介质,其中,所述将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果的步骤,包括:将所述第一药品推荐结果与所述第二药品推荐结果进行融合处理,得到融合药品推荐结果;将所述融合药品推荐结果与预设的互斥药品组进行比对,以判断所述融合药品推荐结果中是否存在互斥药品组;如存在,采用同药性药品替换策略调整所述融合药品推荐结果,以消除所述互斥药品组;根据调整后的所述融合药品推荐结果生成所述目标病患的个性化药品推荐结果。
- 一种计算机设备,包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序,其中,所述处理器执行所述计算机程序时实现以下步骤:获取患同一疾病的多个病患的病历数据,所述病历数据包括结构化数据、文本数据及影像数据,其中,所述病患包括历史病患及当前需要被推荐用药的目标病患;利用命名实体识别算法从每个所述病患的文本数据中得到所述病患的用药信息;基于药品的协同过滤算法从多个所述历史病患的用药信息中筛选得到所述目标病患的第一药品推荐结果;将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息;基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患;根据所述相似病患的用药信息生成第二药品推荐结果;将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果。
- 根据权利要求15所述的计算机设备,其中,所述将所述病患的病历数据进行合并处理,得到所述病患的病情特征信息的步骤,包括:根据预设的映射表将识别到的每个所述病患的用药信息转变为对应的数值型数据;将每个所述病患的病历数据中的结构化数据根据时间顺序形成稀疏矩阵;利用变分自编码器对所述数值型数据、所述稀疏矩阵进行压缩编码处理,得到所述病患的第一编码及第二编码;利用预设的卷积神经网络将所述影像数据进行池化处理,得到所述病患的第三编码;将所述第一编码、所述第二编码及所述第三编码进行合并处理,得到每个所述病患的病情特征信息。
- 根据权利要求16所述的计算机设备,其中,所述第一编码包括来源于所述文本数据的第一特征信息及时间维度信息,所述第二编码包括来源于所述结构化数据的第二特征信息及所述时间维度信息,所述第三编码包括来源于所述影像数据的第三特征信息及所述时间维度信息;所述将所述第一编码、所述第二编码及所述第三编码进行合并处理,得到每个所述病患的病情特征信息的步骤,包括:以所述时间维度信息为基准将每个所述病患的所述第一特征信息、所述第二特征信息及所述第三特征信息合并处理,得到每个所述病患的病情特征信息,所述病情特征信息是一条多维记录数据。
- 根据权利要求15所述的计算机设备,其中,所述基于病患的协同过滤算法从多个所述历史病患中筛选出与所述目标病患的当前病情特征信息相似的至少一个相似病患的步骤,包括:将每个所述病患的病情特征信息输入预设的词向量表示模型,得到每个所述病患的特征信息向量;计算每个所述历史病患的特征信息向量与所述目标病患的特征信息向量之间的欧式距离;将所述欧式距离确认为所述历史病患与所述目标病患的病情相似度;根据所述病情相似度从多个所述历史病患中筛选出至少一个相似病患,其中,所述相似病患与所述目标病患的病情相似度大于预设阈值。
- 根据权利要求15所述的计算机设备,其中,所述根据所述相似病患的用药信息生成第二药品推荐结果的步骤,包括:从预设的疾病-药品的有向连接图中查找所述疾病的关联疾病;根据所述有向连接图获取所述关联疾病及所述疾病的用药信息;根据所述关联疾病、所述疾病的用药信息及所述相似病患的用药信息生成第二药品推荐结果。
- 根据权利要求15所述的计算机设备,其中,所述将所述第一药品推荐结果及所述第二药品推荐结果进行融合处理,得到所述目标病患的个性化药品推荐结果的步骤,包括:将所述第一药品推荐结果与所述第二药品推荐结果进行融合处理,得到融合药品推荐结果;将所述融合药品推荐结果与预设的互斥药品组进行比对,以判断所述融合药品推荐结果中是否存在互斥药品组;如存在,采用同药性药品替换策略调整所述融合药品推荐结果,以消除所述互斥药品组;根据调整后的所述融合药品推荐结果生成所述目标病患的个性化药品推荐结果。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910983855.3 | 2019-10-16 | ||
CN201910983855.3A CN110880361B (zh) | 2019-10-16 | 2019-10-16 | 一种个性化精准用药推荐方法及装置 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021073277A1 true WO2021073277A1 (zh) | 2021-04-22 |
Family
ID=69727978
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/112186 WO2021073277A1 (zh) | 2019-10-16 | 2020-08-28 | 一种个性化精准用药推荐方法及装置 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN110880361B (zh) |
WO (1) | WO2021073277A1 (zh) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113539411A (zh) * | 2021-06-29 | 2021-10-22 | 兰州大学第一医院 | 一种辅助用药效果预判分析系统及方法 |
CN113658659A (zh) * | 2021-08-20 | 2021-11-16 | 平安国际智慧城市科技股份有限公司 | 医疗信息的处理方法、装置、计算机设备和存储介质 |
CN113744867A (zh) * | 2021-08-30 | 2021-12-03 | 平安科技(深圳)有限公司 | 一种药物推荐循证支持方法、装置、设备及存储介质 |
CN115691741A (zh) * | 2023-01-03 | 2023-02-03 | 四川大学华西医院 | 一种医疗信息化信息传递及信息合并方法 |
CN115982472A (zh) * | 2023-03-21 | 2023-04-18 | 中化现代农业有限公司 | 农药用药推荐方法、装置和电子设备 |
CN116110533A (zh) * | 2023-02-27 | 2023-05-12 | 之江实验室 | 基于事件图谱的药物种类和用量推荐系统及方法 |
CN116612852A (zh) * | 2023-07-20 | 2023-08-18 | 青岛美迪康数字工程有限公司 | 一种实现药物推荐的方法、装置和计算机设备 |
CN118098488A (zh) * | 2024-02-02 | 2024-05-28 | 山东卓业医疗科技有限公司 | 基于药品知识图谱推理的用药决策支持系统的构建方法 |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN111402070A (zh) * | 2020-03-23 | 2020-07-10 | 平安医疗健康管理股份有限公司 | 医疗信息识别方法、装置、计算机设备及存储介质 |
CN111696675B (zh) * | 2020-05-22 | 2023-09-19 | 深圳赛安特技术服务有限公司 | 基于物联网数据的用户数据分类方法、装置及计算机设备 |
CN111627515B (zh) * | 2020-05-29 | 2023-07-18 | 上海商汤智能科技有限公司 | 药物推荐方法、装置、电子设备及介质 |
CN111863181A (zh) * | 2020-07-15 | 2020-10-30 | 至本医疗科技(上海)有限公司 | 药品推荐方法、装置、计算机设备和存储介质 |
CN111833984B (zh) * | 2020-07-16 | 2023-06-20 | 平安科技(深圳)有限公司 | 基于机器学习的药品质控分析方法、装置、设备及介质 |
CN112116978B (zh) * | 2020-09-17 | 2023-01-31 | 陕西师范大学 | 一种风湿免疫药物推荐方法、系统及装置 |
CN111933302B (zh) * | 2020-10-09 | 2021-01-05 | 平安科技(深圳)有限公司 | 药物推荐方法、装置、计算机设备及存储介质 |
CN113010783A (zh) * | 2021-03-17 | 2021-06-22 | 华南理工大学 | 基于多模态心血管疾病信息的医疗推荐方法、系统及介质 |
CN113284627B (zh) * | 2021-04-15 | 2024-05-17 | 北京交通大学 | 基于患者表征学习的用药推荐方法 |
CN113436746B (zh) * | 2021-06-30 | 2024-04-12 | 平安科技(深圳)有限公司 | 基于排序算法的用药推荐方法、装置、设备及存储介质 |
CN113593670A (zh) * | 2021-08-05 | 2021-11-02 | 江西省科学院应用物理研究所 | 一种家用直流电刺激医疗设备的处方生成方法及系统 |
CN116936021A (zh) * | 2023-09-18 | 2023-10-24 | 万链指数(青岛)信息科技有限公司 | 一种基于区块链的医疗电子病历信息管理方法及系统 |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106778042A (zh) * | 2017-01-26 | 2017-05-31 | 中电科软件信息服务有限公司 | 心脑血管患者相似性分析方法及系统 |
CN106845105A (zh) * | 2017-01-18 | 2017-06-13 | 腾讯科技(深圳)有限公司 | 问诊辅助方法及装置 |
CN109102855A (zh) * | 2018-07-03 | 2018-12-28 | 北京康夫子科技有限公司 | 药物推荐方法 |
US20190035506A1 (en) * | 2017-07-31 | 2019-01-31 | Hefei University Of Technology | Intelligent auxiliary diagnosis method, system and machine-readable medium thereof |
CN109785928A (zh) * | 2018-12-25 | 2019-05-21 | 平安科技(深圳)有限公司 | 诊疗方案推荐方法、装置及存储介质 |
CN110880361A (zh) * | 2019-10-16 | 2020-03-13 | 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 |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10789634B2 (en) * | 2014-12-08 | 2020-09-29 | Lg Cns Co., Ltd. | Personalized recommendation method and system, and computer-readable record medium |
CN108986879B (zh) * | 2018-05-31 | 2024-04-05 | 平安医疗科技有限公司 | 药品推荐方法、装置、计算机设备和存储介质 |
CN108899070B (zh) * | 2018-05-31 | 2024-02-02 | 平安医疗科技有限公司 | 处方推荐生成方法、装置、计算机设备和存储介质 |
CN110232960A (zh) * | 2019-05-16 | 2019-09-13 | 武汉市西贝智诚科技有限公司 | 一种基于大数据的智能荐药系统及其取药柜 |
CN110245217B (zh) * | 2019-06-17 | 2022-07-22 | 京东方科技集团股份有限公司 | 一种药品推荐方法、装置及电子设备 |
-
2019
- 2019-10-16 CN CN201910983855.3A patent/CN110880361B/zh active Active
-
2020
- 2020-08-28 WO PCT/CN2020/112186 patent/WO2021073277A1/zh active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106845105A (zh) * | 2017-01-18 | 2017-06-13 | 腾讯科技(深圳)有限公司 | 问诊辅助方法及装置 |
CN106778042A (zh) * | 2017-01-26 | 2017-05-31 | 中电科软件信息服务有限公司 | 心脑血管患者相似性分析方法及系统 |
US20190035506A1 (en) * | 2017-07-31 | 2019-01-31 | Hefei University Of Technology | Intelligent auxiliary diagnosis method, system and machine-readable medium thereof |
CN109102855A (zh) * | 2018-07-03 | 2018-12-28 | 北京康夫子科技有限公司 | 药物推荐方法 |
CN109785928A (zh) * | 2018-12-25 | 2019-05-21 | 平安科技(深圳)有限公司 | 诊疗方案推荐方法、装置及存储介质 |
CN110880361A (zh) * | 2019-10-16 | 2020-03-13 | 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 |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113539411A (zh) * | 2021-06-29 | 2021-10-22 | 兰州大学第一医院 | 一种辅助用药效果预判分析系统及方法 |
CN113658659B (zh) * | 2021-08-20 | 2023-07-21 | 深圳平安智慧医健科技有限公司 | 医疗信息的处理方法、装置、计算机设备和存储介质 |
CN113658659A (zh) * | 2021-08-20 | 2021-11-16 | 平安国际智慧城市科技股份有限公司 | 医疗信息的处理方法、装置、计算机设备和存储介质 |
CN113744867A (zh) * | 2021-08-30 | 2021-12-03 | 平安科技(深圳)有限公司 | 一种药物推荐循证支持方法、装置、设备及存储介质 |
CN113744867B (zh) * | 2021-08-30 | 2024-03-08 | 平安科技(深圳)有限公司 | 一种药物推荐循证支持方法、装置、设备及存储介质 |
CN115691741A (zh) * | 2023-01-03 | 2023-02-03 | 四川大学华西医院 | 一种医疗信息化信息传递及信息合并方法 |
CN116110533A (zh) * | 2023-02-27 | 2023-05-12 | 之江实验室 | 基于事件图谱的药物种类和用量推荐系统及方法 |
CN116110533B (zh) * | 2023-02-27 | 2023-09-01 | 之江实验室 | 基于事件图谱的药物种类和用量推荐系统及方法 |
CN115982472B (zh) * | 2023-03-21 | 2023-08-08 | 中化现代农业有限公司 | 农药用药推荐方法、装置和电子设备 |
CN115982472A (zh) * | 2023-03-21 | 2023-04-18 | 中化现代农业有限公司 | 农药用药推荐方法、装置和电子设备 |
CN116612852A (zh) * | 2023-07-20 | 2023-08-18 | 青岛美迪康数字工程有限公司 | 一种实现药物推荐的方法、装置和计算机设备 |
CN116612852B (zh) * | 2023-07-20 | 2023-10-31 | 青岛美迪康数字工程有限公司 | 一种实现药物推荐的方法、装置和计算机设备 |
CN118098488A (zh) * | 2024-02-02 | 2024-05-28 | 山东卓业医疗科技有限公司 | 基于药品知识图谱推理的用药决策支持系统的构建方法 |
Also Published As
Publication number | Publication date |
---|---|
CN110880361A (zh) | 2020-03-13 |
CN110880361B (zh) | 2023-02-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021073277A1 (zh) | 一种个性化精准用药推荐方法及装置 | |
CN108899070B (zh) | 处方推荐生成方法、装置、计算机设备和存储介质 | |
WO2021135429A1 (zh) | 基于知识图谱的健康信息推荐方法、装置、设备及介质 | |
CN111696675B (zh) | 基于物联网数据的用户数据分类方法、装置及计算机设备 | |
US9165116B2 (en) | Patient data mining | |
WO2022062353A1 (zh) | 医疗数据处理方法、装置、计算机设备和存储介质 | |
US20200034366A1 (en) | Identifying Missing Questions by Clustering and Outlier Detection | |
CN110032728B (zh) | 疾病名称标准化的转换方法和装置 | |
CN108766561B (zh) | 病症信息处理方法、装置、计算机设备和存储介质 | |
CN109036545B (zh) | 医疗信息处理方法、装置、计算机设备和存储介质 | |
US20140341449A1 (en) | Computer system and method for atlas-based consensual and consistent contouring of medical images | |
CN107016438A (zh) | 一种基于中医辨证人工神经网络算法模型的系统 | |
CN109036508B (zh) | 一种中医医疗信息处理方法、装置、计算机设备和存储介质 | |
Darapureddy et al. | Optimal weighted hybrid pattern for content based medical image retrieval using modified spider monkey optimization | |
CN111091906A (zh) | 一种基于真实世界数据的辅助医疗诊断方法及系统 | |
WO2023240837A1 (zh) | 基于病患数据的服务包生成方法、装置、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 | |
CN112037904A (zh) | 在线诊疗数据处理方法、装置、计算机设备和存储介质 | |
Herbadji et al. | Combining multiple biometric traits using asymmetric aggregation operators for improved person recognition | |
JP2022546192A (ja) | 医学データの検証方法 | |
CN113724859A (zh) | 基于人工智能的疾病提示设备、方法、装置及存储介质 | |
CN112447270A (zh) | 一种用药推荐方法、装置、设备及存储介质 | |
CN113436725A (zh) | 数据处理方法、系统、计算机设备及计算机可读存储介质 | |
KR102595904B1 (ko) | 의료상담 콘텐츠에 기초한 의료상담자 추천 방법 및 시스템 | |
WO2020132918A1 (zh) | 药品预测方法、装置、计算机设备及存储介质 |
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: 20876898 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20876898 Country of ref document: EP Kind code of ref document: A1 |