WO2021179630A1 - 一种并发症风险预测***、方法、装置、设备及介质 - Google Patents
一种并发症风险预测***、方法、装置、设备及介质 Download PDFInfo
- Publication number
- WO2021179630A1 WO2021179630A1 PCT/CN2020/124611 CN2020124611W WO2021179630A1 WO 2021179630 A1 WO2021179630 A1 WO 2021179630A1 CN 2020124611 W CN2020124611 W CN 2020124611W WO 2021179630 A1 WO2021179630 A1 WO 2021179630A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- target
- risk
- weight
- complication
- risk factors
- 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
- 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
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
Definitions
- This application relates to the field of artificial intelligence technology, and in particular to a complication risk prediction system, method, device, equipment, and medium.
- the inventor realizes that the incidence of various diseases is increasing year by year, and patients are often accompanied by at least one complication, and these complications are often not detected in time, which increases the difficulty of treatment. Therefore, how to achieve reliable prediction of complications has become an urgent problem to be solved.
- the embodiments of the present application provide a complication risk prediction system, method, device, equipment, and medium, which help improve the reliability of complication prediction.
- an embodiment of the present application provides a complication risk prediction system, including: a risk prediction device and a storage device; wherein the storage device is used to store diagnosis and treatment data of a user;
- the risk prediction device is used to perform the following steps:
- Multiple single-task learning models are used to process the diagnosis and treatment data to obtain the first risk factor of each target complication among multiple target complications;
- the target complication is a complication under the target disease type ,
- the target complication corresponds to the single-task learning model one-to-one;
- the complication risk information of the target user for the target disease type is determined.
- the embodiments of the present application provide a method for predicting the risk of complications, including:
- Multiple single-task learning models are used to process the diagnosis and treatment data to obtain the first risk factor of each target complication among multiple target complications;
- the target complication is a complication under the target disease type ,
- the target complication corresponds to the single-task learning model one-to-one;
- the complication risk information of the target user for the target disease type is determined.
- an embodiment of the present application provides a complication risk prediction device, including:
- the obtaining module is used to obtain the diagnosis and treatment data corresponding to the target disease type of the target user;
- the processing module is used to process the diagnosis and treatment data by using multiple single-task learning models to obtain the first risk factor of each target complication among multiple target complications;
- the target complication is the target disease Complications under types, and the target complication corresponds to the single-task learning model one-to-one;
- the processing module is further configured to use a multi-task learning model to process the first risk factors corresponding to the multiple target complications to obtain multiple second risk factors, and determine the first risk factor of each second risk factor. Weight, and determine the second weight corresponding to each target complication;
- the prediction module is configured to determine the risk information of complications of the target user for the target disease type according to the first weight of each second risk factor and the second weight corresponding to each target complication.
- an embodiment of the present application provides a risk prediction device.
- the risk prediction device may include a processor and a memory, and the processor and the memory are connected to each other.
- the memory is used to store a computer program that supports the terminal device to execute the above-mentioned methods or steps, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the following methods:
- Multiple single-task learning models are used to process the diagnosis and treatment data to obtain the first risk factor of each target complication among multiple target complications;
- the target complication is a complication under the target disease type ,
- the target complication corresponds to the single-task learning model one-to-one;
- the complication risk information of the target user for the target disease type is determined.
- embodiments of the present application provide a computer-readable storage medium that stores a computer program, and the computer program includes program instructions that, when executed by a processor, cause all The processor executes the following methods:
- Multiple single-task learning models are used to process the diagnosis and treatment data to obtain the first risk factor of each target complication among multiple target complications;
- the target complication is a complication under the target disease type ,
- the target complication corresponds to the single-task learning model one-to-one;
- the complication risk information of the target user for the target disease type is determined.
- the computer-readable storage medium may be non-volatile or volatile.
- the embodiments of the present application can combine a single-task learning model and a multi-task learning model to realize complication prediction, and help improve the reliability of complication prediction.
- Figure 1 is a schematic structural diagram of a complication risk prediction system provided by an embodiment of the present application
- FIG. 2 is a schematic flowchart of a method for predicting a risk of complications according to an embodiment of the present application
- FIG. 3 is a schematic flowchart of another complication risk prediction method provided by an embodiment of the present application.
- FIG. 4 is a schematic structural diagram of a complication risk prediction device provided by an embodiment of the present application.
- Fig. 5 is a schematic structural diagram of a risk prediction device provided by an embodiment of the present application.
- the technical solution of the present application can be applied to the fields of artificial intelligence, digital medical care, smart city, blockchain and/or big data technology, for example, it can specifically involve machine learning technology to realize smart medical care.
- the data involved in this application such as diagnosis and treatment data, can be stored in a database, or can be stored in a blockchain, which is not limited in this application.
- the technical solution of the present application can be applied to a risk prediction system, and can be specifically applied to a risk prediction device (risk prediction device) for realizing the prediction of the risk of complications.
- the risk prediction device may be a terminal, a server, or a data platform or other devices.
- the terminal may include a mobile phone, a tablet computer, a computer, etc., which is not limited in this application. It can be understood that, in other embodiments, the terminal may also be called other names, such as terminal equipment, smart terminal, user equipment, user terminal, etc., which are not listed here.
- This application can combine multi-task learning to achieve risk prediction for multiple complications; moreover, compared to disease-focused risk factor analysis or risk prediction only for a single complication, this application is based on risk prediction for multiple complications, It helps to improve the reliability of complication risk prediction to achieve comprehensive targeted treatment.
- the technical solution of this application can be applied to the fields of artificial intelligence, smart city, blockchain and/or big data technology, and the data involved can be stored through blockchain nodes or can be stored in a database, which is not limited by this application.
- the embodiments of the present application provide a complication risk prediction system, method, device, equipment, medium, etc., so as to help improve the reliability of complication risk prediction. Detailed descriptions are given below.
- FIG. 1 is a schematic structural diagram of a complication risk prediction system provided by an embodiment of the present application.
- the complication risk prediction system may include a risk prediction device (risk prediction device) 101 and a storage device (storage device) 102. in,
- the storage device 102 can be used to store the user's diagnosis and treatment data
- the risk prediction device 101 can be used to obtain the diagnosis and treatment data corresponding to the target disease type of the target user from the storage device 102; and use multiple single-task learning models to process the diagnosis and treatment data to obtain each of the multiple target complications.
- the first risk factor of the target complication use the multi-task learning model to process the first risk factors corresponding to the multiple target complication to obtain multiple second risk factors; determine the first weight of each second risk factor , And determine the second weight corresponding to each target complication; determine the target user’s complication risk for the target disease type according to the first weight of each second risk factor and the second weight corresponding to each target complication information.
- the target complication is a complication of the target disease type, and the target complication can correspond to a single-task learning model one-to-one.
- the storage device and the risk prediction device may be independent devices, that is, independently deployed, or the storage device and the risk prediction device may also be deployed in the same device, which is not limited in this application, and FIG. 1 only shows Independent deployment scenario.
- the storage device and the risk prediction device may be deployed in a server, or in other words, the storage device may be deployed in a risk prediction device.
- the storage device may be a blockchain node, and the diagnosis and treatment data may be obtained from the blockchain. That is, the diagnosis and treatment data of each patient can be stored in the blockchain in advance. By obtaining the user's diagnosis and treatment data from the blockchain node, the reliability of the obtained diagnosis and treatment data can be improved, which in turn helps to improve the reliability of the risk of complications determined based on the diagnosis and treatment data.
- diagnosis and treatment data may include physical sign data, examination and inspection data, etc.
- the collected diagnosis and treatment data may be determined according to the target disease type, or the diagnosis and treatment data may be all the diagnosis and treatment data of the target user, or may be a preset time period (Such as the most recent year) diagnosis and treatment data.
- the data can be extracted from the monitoring system, and the storage device is a storage device in the monitoring system, or the data can be stored in the storage device after being extracted by the monitoring system, which is not limited in this application.
- the diagnosis and treatment data may be obtained by processing the collected raw medical data, and the processing includes sampling, filling in missing values, and so on.
- the patient's original medical data can be obtained, including the patient's historical baseline data.
- the historical base station data can include multiple visit records, and each visit record can include various diagnoses, tests, examinations, medications, and surgical items.
- the historical baseline data can be preprocessed.
- the physical sign data can be obtained by sampling the collected raw physical sign data in a preset time unit (for example, in 1h unit), and the raw physical sign data can be continuous data.
- diagnosis and treatment data can be text data, or vectors, such as binary features, or called two-dimensional feature vectors, and so on.
- the complication risk information can be used to indicate the complication risk level, the risk index, the probability of occurrence of the target complication, etc., which is not limited in this application.
- the risk prediction model can be used to process the first weight and the second weight to obtain the risk information of the target user's complication for the target disease type.
- the loss function of the risk prediction model may be determined based on the weights of risk factors corresponding to the target disease type and the weight of the target complications obtained based on multiple diagnosis and treatment sample data, or in other words, the risk prediction model may be based on multiple diagnosis and treatment samples.
- the risk factor weights corresponding to the target disease types obtained from the sample data and the weights of the target complications are trained.
- the risk prediction device 101 may train to obtain a risk prediction model in the following manner: obtain diagnosis and treatment sample data of multiple patients, and use multiple single-task learning models to process the diagnosis and treatment sample data to obtain Multiple first risk factors corresponding to multiple target complications; use a multi-task learning model to process the multiple first risk factors to determine multiple second risk factors from the multiple first risk factors, and Determine the weight of each second risk factor; determine the weight corresponding to each target complication; train the risk prediction model according to the weight of each second risk factor and the weight corresponding to each target complication.
- the loss function of the risk prediction model can be determined based on the weight of each second risk factor and the weight corresponding to each target complication, so as to train the risk prediction model, or in other words, it can be based on the weight of each second risk factor.
- the weight and the weight corresponding to each target complication determine the objective function of the risk prediction model to train the risk prediction model.
- the risk prediction model can be used to predict the user's complication risk information for the target disease type.
- the weight corresponding to the target complication may be determined according to the weight of the first risk factor and/or the second risk factor.
- the weight corresponding to the target complication may refer to the weight of the loss function corresponding to the target complication; for another example, the weight corresponding to the target complication may be the weight of the target complication determined by other means, such as the weight corresponding to the target complication
- the number of the first risk factor or the second risk factor is set, and the larger the number, the larger the weight corresponding to the target complication can be set.
- the diagnosis and treatment data of a plurality of patients can be acquired as the diagnosis and treatment sample data (sample data).
- the patient may be a patient suffering from a target complication corresponding to the target disease type.
- the risk factors (features) and weights of multiple complications corresponding to the target disease type can be determined according to the diagnosis and treatment sample data.
- feature engineering processing can be performed on the diagnosis and treatment data variables to obtain the risk factors and weights of complications corresponding to the target disease type.
- the loss function may be a least squares loss function.
- the risk prediction device 101 uses a multi-task learning model to process the multiple first risk factors to determine multiple second risk factors from the multiple first risk factors, and determine each For the weight of the second risk factor, L1 regularization can be used to process the multiple first risk factors to determine the weight matrix corresponding to the multiple first risk factors, and then according to the weight matrix corresponding to the multiple first risk factors Perform feature selection to determine the weight matrix corresponding to the multiple second risk factors, so that the multiple second risk factors and the weight of each second risk factor can be determined according to the weight matrix corresponding to the multiple second risk factors .
- L1 regularization can be used to process the multiple first risk factors to determine the weight matrix corresponding to the multiple first risk factors, and then according to the weight matrix corresponding to the multiple first risk factors Perform feature selection to determine the weight matrix corresponding to the multiple second risk factors, so that the multiple second risk factors and the weight of each second risk factor can be determined according to the weight matrix corresponding to the multiple second risk factors .
- the weight corresponding to the target complication may refer to the weight of the loss function of the target complication; when determining the weight corresponding to each target complication, the risk prediction device 101 may use the maximum Gaussian likelihood estimation The algorithm processes the second risk factor and the weight of each second risk factor to obtain the weight of the loss function of each target complication.
- the storage device may be a blockchain node.
- the risk prediction device 101 may also be used to receive a risk prediction request sent by a target user terminal, and the risk prediction request may carry an identifier of the target user.
- the risk prediction device 101 may be specifically configured to obtain the diagnosis and treatment data from a storage device such as the blockchain node according to the target user's identifier.
- the risk prediction device 101 may also be used to send a prompt message to the target user terminal according to the complication risk information.
- the prompt message may include information and treatment plan for indicating the target complication that is at risk, and so on.
- the information used to indicate a risky target complication may include one or more of the name, identification, risk score, risk level, probability, and other information of one or more target complication that is risky.
- the treatment plan may be a treatment plan corresponding to the user group to which the target user belongs.
- the user group to which the target user belongs may be the group with the largest net benefit under the treatment plan.
- user grouping can be realized based on the net benefit of each treatment plan corresponding to the target disease (type), and the user group with the largest net benefit under each treatment plan can be obtained respectively.
- the risk prediction system can obtain the diagnosis and treatment data corresponding to the user's target disease type, and use multiple single-task learning models to process the diagnosis and treatment data to obtain each of the multiple target complications.
- the first risk factor of the complication and then use the multi-task learning model to process the first risk factor corresponding to the multiple target complication to obtain multiple second risk factors, and according to the weight of each second risk factor and
- the weight corresponding to each target complication determines the complication risk information of the target user for the target disease type, which can combine single-task learning model and multi-task learning model to achieve complication prediction and help improve complication prediction Reliability.
- FIG. 2 is a schematic flowchart of a method for predicting complication risk according to an embodiment of the present application.
- the method may be executed by the above-mentioned risk prediction device.
- the complication risk prediction method may include the following steps:
- the diagnosis and treatment data may include physical sign data, examination and inspection data, and so on.
- the diagnosis and treatment data may be obtained by processing the collected raw medical data, which will not be repeated here.
- the target complication is a complication of the target disease type, and the target complication can correspond to the single-task learning model one-to-one.
- the target complication can be complication 1 and complication 2.
- the single task learning model 1 corresponding to complication 1 can be used to process the diagnosis and treatment data to obtain the risk factors corresponding to complication 1
- the single task learning model 2 corresponding to complication 2 can be used to process the diagnosis and treatment data to obtain the complication. 2 Corresponding risk factors.
- the risk factors screened out for each target complication can be input into the multitask learning model (multitask learning, MTL), and the risk factors of the multiple target complication Further screening is performed to obtain the risk factor of the complications corresponding to the target disease type, that is, the second risk factor.
- MTL multitask learning
- the weight of the second risk factor (ie, the first weight) may be obtained by performing characteristic processing on the first risk factor based on L1 regularization.
- risk factors corresponding to multiple complications of multiple disease types can be pre-trained, and the features can be further processed through L1 regularization to obtain a weight matrix to obtain each risk factor.
- the weight of the target complication (ie, the second weight) may refer to the weight of the loss function of the target complication, and the weight of the loss function of the target complication may be based on the Gaussian likelihood estimation of the uncertainty of the maximum task decision, Find the target complication loss function to determine the weight of each target complication.
- the first weight and the second weight may be determined based on historical patient diagnosis and treatment sample data, for example, based on historical patient diagnosis and treatment sample data combined with a single-task learning model and a multi-task learning model to determine each second risk
- the weight of the factor and the weight corresponding to each complication, and each second risk factor and its weight can be stored, as well as the weight corresponding to each complication.
- the weight of the second risk factor can be directly searched, and the weight corresponding to each target complication can be further searched, so as to achieve multiple target concurrent goals for the target user Risk prediction of disease.
- the second risk factor corresponding to the target user can also be determined in other ways, for example, according to the second risk factor set determined during the model training process, the target user's corresponding risk factor can be selected from the target user's diagnosis and treatment data. The second risk factor.
- the risk factors involved in this application may be vectors, such as binary features, so as to facilitate feature screening by the task learning model.
- the risk factor can also be called a feature, a feature vector or other names, which is not limited in this application.
- the complication risk information can be used to indicate the risk level, risk index, and probability of target complication for multiple complication, which is not limited in this application.
- the complication risk information may be determined based on the aforementioned first weight, second weight, and loss function such as a least squares loss function.
- the risk prediction device obtains the complication risk information of the target user based on the above-mentioned first weight, second weight, and risk prediction model.
- a risk prediction model can be used to obtain the first weight and the second weight.
- Processing is performed to obtain the risk information of the target user's complication for the target disease type.
- the loss function of the risk prediction model is determined based on the weight of the risk factor and the weight of the target complication corresponding to the target disease type obtained from multiple diagnosis and treatment sample data.
- the risk prediction device can obtain the diagnosis and treatment sample data of multiple patients, and use multiple single-task learning models to process the diagnosis and treatment sample data to obtain multiple first risk factors corresponding to multiple target complications; Use a multi-task learning model to process the multiple first risk factors to determine multiple second risk factors from the multiple first risk factors, and determine the weight of each second risk factor; determine each target The weight corresponding to the complication; according to the weight of each second risk factor and the weight corresponding to each target complication, the loss function is determined to train the risk prediction model.
- the risk prediction model is used to predict the user's complication risk information for the target disease type.
- the risk prediction device may use L1 regularization to process the multiple first risk factors to obtain the weight matrix corresponding to the multiple first risk factors; Furthermore, feature selection may be performed according to the weight matrix corresponding to the plurality of first risk factors to determine the weight matrix corresponding to the plurality of second risk factors; and the plurality of first risk factors are determined according to the weight matrix corresponding to the plurality of second risk factors. Two risk factors and the weight of each second risk factor.
- the weight corresponding to the target complication may be the weight of the loss function of the target complication, or may be a weight determined in other ways, which will not be repeated here.
- the maximum Gaussian likelihood estimation algorithm may be used to process the second risk factor and the weight of each second risk factor to obtain the weight of the loss function of each target complication.
- the risk prediction device may also receive a risk prediction request sent by the target user terminal, and the risk prediction request carries the target user's identifier.
- the diagnosis and treatment data can be obtained according to the identification of the target user, for example, the diagnosis and treatment data can be obtained from a storage device such as a blockchain node.
- the risk prediction device may also send a prompt message to the target user terminal according to the complication risk information, including information and treatment plan for indicating the target complication that is at risk, etc., which will not be repeated here.
- the risk prediction device may obtain the user's diagnosis and treatment data, and process the diagnosis and treatment data by combining the single-task learning model and the multi-task learning model to obtain multiple risk factors corresponding to multiple target disease types, and then According to the weight of the multiple risk factors and the weight corresponding to each target complication, the risk prediction of multiple complications of the user can be realized, which is helpful to improve the reliability of the prediction of the complication and comprehensive treatment.
- Fig. 3 is a schematic flowchart of another complication risk prediction method provided by an embodiment of the present application. As shown in the figure, the complication risk prediction method may include the following steps:
- the patient may be a patient suffering from a target complication corresponding to the target disease type.
- the diagnosis and treatment data may include physical sign data, examination data, and so on.
- the diagnosis and treatment data may be obtained by processing collected raw medical data.
- the patient's original medical data can be obtained, including the patient's historical baseline data and outcome data.
- the historical base station data can include records of multiple visits, and each visit record can include various diagnoses, tests, examinations, medications, surgical items, etc.
- Outcome data can be discharge diagnosis data corresponding to each patient's visit record, etc.
- the historical baseline data can be preprocessed, for example, multiple interpolations are used to fill in missing values in the original medical data, so as to obtain preprocessed diagnosis and treatment data.
- Multiple imputation is a method of processing missing values based on repeated simulations. It can generate a complete set of diagnosis and treatment data from a data set of electronic medical record data containing missing values. Missing data in two data sets. By interpolating the original diagnosis and treatment data multiple times, it is helpful to improve the reliability of the determined risk factors, thereby improving the reliability of the risk prediction of complications.
- the target complication may be a complication of the target disease type, and the target complication may correspond to the single-task learning model one-to-one.
- the risk factors (features) of the complications corresponding to the target disease type and their weights can be determined according to the diagnosis and treatment data. For example, feature engineering processing can be performed on the diagnosis and treatment data variables to obtain the risk factors and weights of the complications corresponding to the target disease type.
- the risk factor of the target complication is the first risk factor.
- the risk factor may be part of the data of the user's diagnosis and treatment data, or may be the data after the diagnosis and treatment data is processed.
- the risk factors screened out for each target complication can be input into the multi-task learning model, and the risk factors screened out by the single-task learning model can be processed to obtain the further screened risk factor, that is, the second risk factor, and determine the first risk factor.
- the weight of risk factors For example, the first risk factor can be processed through L1 regularization to form a sparse matrix to obtain the second risk factor. Among them, only a few features (that is, the second risk factor) contribute to this model, and most of the features have no or little contribution.
- the obtained weight matrix corresponding to the second risk factor can be as follows:
- orthogonal approach can be used to adjust the relative weight of each task (complication), for example, by determining the weight of the loss function of each task, that is, the second weight To determine the risk of complications.
- the second weight can be determined in the following way:
- f W (x) can refer to the weight result of the variable (x) after L1 regularization, that is, the first weight;
- ⁇ can refer to the variance of the data, that is, the weight of the loss function of the target complication, that is, the second weight .
- the target disease type is disease 1
- the target complication is complication 1 and complication 2
- complication risk information indicating the risk of complication 1y 1 and complication 2y 2 for disease 1 patients can be obtained.
- these two tasks obey Gaussian distribution:
- Gaussian likelihood estimation based on uncertainty of maximizing task decision
- ⁇ 1 and ⁇ 2 are obtained, which are the weights of the loss function of complication 1 and complication 2, respectively.
- the weight of the second risk factor and the weight corresponding to the target complication can be stored, so as to facilitate the subsequent rapid determination of the risk factor Weight and complication weight.
- the risk factor such as the weight of the second risk factor and the weight of the complication, such as the weight of the target complication, can be stored in the blockchain to improve storage security, thereby increasing the weight and concurrency of subsequent acquisition of risk factors The safety and reliability of symptom weight.
- each second risk factor and the weight corresponding to each target complication train to obtain a risk prediction model. For example, based on the weight of each second risk factor and the weight corresponding to each target complication, the loss function is determined to train the risk prediction model.
- the risk prediction model can be used to predict the user's complication risk information for the target disease type.
- the objective function may be made such as the target weighted sum Reach the minimum to determine the loss function (Alternatively called determining f(x)).
- ⁇ i is the loss of function of target complications weight, i.e. a second weight
- w i is the weight of risk factors of weight, i.e., a first weight, i.e., feature weights
- X i is a second risk factor (characteristic)
- Y i is the end, That is, the patient’s complication risk information
- the risk prediction model can be trained for multiple diseases (disease types). For example, model training is performed for each disease type to obtain a risk prediction model that can identify the risk of complications in multiple diseases (disease types), which will not be repeated here.
- the patient's diagnosis and treatment data can be subsequently obtained, based on the risk factor weights and complication (loss function) weights corresponding to the diagnosis and treatment data, to determine the patient's complication risk (outcome).
- the target user can be any user who performs risk prediction of complications.
- the acquisition operation of the diagnosis and treatment data may be triggered by a trigger condition.
- the trigger condition may be receiving a request for predicting the risk of complications for the target user, that is, when receiving a risk prediction for predicting the risk of complications for the target user, the diagnosis and treatment data of the target user can be obtained.
- the trigger condition may be receiving an admission request, so that when the target user's admission request is received, the target user's diagnosis and treatment data can be triggered, so that high-accuracy correlations can be obtained at the beginning of the patient's admission. Complication risk prediction, so as to provide doctors with accurate prognostic possibilities for patients, and achieve personalized treatment and disease management. It is also possible to trigger the acquisition of user diagnosis and treatment data based on other trigger conditions, and this application does not limit the trigger conditions for obtaining the diagnosis and treatment data.
- the first weight of each second risk factor can be determined, and the second weight corresponding to each target complication can be determined.
- the first weight may be determined based on the weight of the corresponding second risk factor determined in the model training stage. If the weight of the same risk factor is the same, the second weight may be determined based on the corresponding complication weight determined in the model training stage. If the weight of the same complication (such as the weight of the complication loss function) is the same.
- the risk prediction model obtained by training can be used to determine the complication risk information corresponding to the target user based on the above objective function , Such as ending Y.
- the diagnosis and treatment sample data of multiple patients can be acquired, and the single-task learning model and the multi-task learning model can be combined to process the diagnosis and treatment sample data to obtain multiple risk factors corresponding to multiple target disease types. Then train the risk prediction model based on the weights of the multiple risk factors and the weights corresponding to each target complication, so that the subsequent diagnosis and treatment data of the user can be obtained to determine the multiple corresponding diagnosis and treatment data under multiple target disease types.
- the weight of the risk factor and the weight corresponding to each target complication, and the use of a risk prediction model to realize the risk prediction of multiple complications for the user which helps to improve the reliability of the complication prediction.
- a highly accurate risk prediction of related complications can be obtained at the beginning of the patient’s admission, providing doctors with accurate prognostic possibilities for the patient’s prognostic outcome, achieving the purpose of personalized treatment and disease management, and helping to improve the reliability of prediction and implementation Comprehensive treatment.
- the embodiment of the present application also provides a complication risk prediction device.
- the device may include a module for executing the method described in FIG. 2 or FIG. 3 above.
- FIG. 4 is a schematic structural diagram of a complication risk prediction device provided by an embodiment of the present application.
- the complication risk prediction device described in this embodiment may be configured in a risk prediction device.
- the complication risk prediction device 400 of this embodiment may include: an acquisition module 401, a processing module 402, and a prediction module 403. in,
- the obtaining module 401 is used to obtain the diagnosis and treatment data corresponding to the target disease type of the target user;
- the processing module 402 is configured to process the diagnosis and treatment data by using multiple single-task learning models to obtain the first risk factor of each target complication among multiple target complications;
- the target complication is the target Complications under disease types, and the target complication corresponds to the single-task learning model one-to-one;
- the processing module 402 is further configured to use a multi-task learning model to process the first risk factors corresponding to the multiple target complications to obtain multiple second risk factors, and determine the first risk factor of each second risk factor A weight, and a second weight corresponding to each target complication;
- the prediction module 403 is configured to determine the risk information of complications of the target user for the target disease type according to the first weight of each second risk factor and the second weight corresponding to each target complication.
- the prediction module 403 may be specifically configured to use a risk prediction model to process the first weight and the second weight to obtain the target user’s complication for the target disease type Risk information;
- the loss function of the risk prediction model is determined based on the weight of the risk factor corresponding to the target disease type and the weight of the target complication obtained from a plurality of diagnosis and treatment sample data.
- the acquisition module 401 can also be used to acquire diagnosis and treatment sample data of multiple patients, and use multiple single-task learning models to process the diagnosis and treatment sample data to obtain multiple target complications. Of multiple first risk factors;
- the processing module 402 can also be used to process the multiple first risk factors by using a multi-task learning model, so as to determine multiple second risk factors from the multiple first risk factors, and determine each The weight of the second risk factor; determine the weight corresponding to each target complication; determine the loss function according to the weight of each second risk factor and the weight corresponding to each target complication, so as to train to obtain the A risk prediction model, which is used to predict the user's complication risk information for the target disease type.
- the processing module 402 is using a multi-task learning model to process the plurality of first risk factors to determine a plurality of second risk factors from the plurality of first risk factors, and When determining the weight of each second risk factor, it can be specifically used to:
- the multiple second risk factors and the weight of each second risk factor are determined according to the weight matrix corresponding to the multiple second risk factors.
- the weight corresponding to the target complication is the weight of the loss function of the target complication; when the processing module 402 determines the weight corresponding to each target complication, it can be specifically used to:
- the maximum Gaussian likelihood estimation algorithm is used to process the second risk factor and the weight of each second risk factor to obtain the weight of the loss function of each target complication.
- the obtaining module 401 may also be configured to receive a risk prediction request sent by a target user terminal, where the risk prediction request carries the target user's identifier;
- the obtaining module 401 may be specifically configured to obtain the diagnosis and treatment data from the storage device such as a blockchain node according to the target user's identifier;
- the prediction module 403 may also be used to send a prompt message to the target user terminal according to the complication risk information; wherein, the prompt message includes information and a treatment plan for indicating the target complication that is at risk.
- FIG. 5 is a schematic structural diagram of a risk prediction device provided by an embodiment of the present application.
- the risk prediction device may include: a processor 501 and a memory 502.
- the risk prediction device may further include a communication interface 503.
- the above-mentioned processor 501, memory 502, and communication interface 503 may be connected through a bus or in other ways.
- the connection through a bus is taken as an example.
- the communication interface 503 can be controlled by the processor to send and receive messages
- the memory 502 can be used to store a computer program
- the computer program includes program instructions
- the processor 501 is used to execute the program instructions stored in the memory 502.
- the processor 501 is configured to call the program instructions to execute the following steps:
- Multiple single-task learning models are used to process the diagnosis and treatment data to obtain the first risk factor of each target complication among multiple target complications;
- the target complication is a complication under the target disease type ,
- the target complication corresponds to the single-task learning model one-to-one;
- the complication risk information of the target user for the target disease type is determined.
- the processor 501 is executing the first weight of each second risk factor and the second weight corresponding to each target complication to determine the target user’s complication for the target disease type.
- the following steps can be specifically performed:
- the loss function of the risk prediction model is determined based on the weight of the risk factor corresponding to the target disease type and the weight of the target complication obtained from a plurality of diagnosis and treatment sample data.
- the processor 501 may also be configured to perform the following steps:
- the loss function is determined to train to obtain the risk prediction model, and the risk prediction model is used to predict that the user will target the target Information on the risk of complications for the disease type.
- the processor 501 is performing the processing of the plurality of first risk factors using the multi-task learning model, so as to determine the plurality of second risk factors from the plurality of first risk factors,
- the following steps can be specifically performed:
- the multiple second risk factors and the weight of each second risk factor are determined according to the weight matrix corresponding to the multiple second risk factors.
- the weight corresponding to the target complication is the weight of the loss function of the target complication; when determining the weight corresponding to each target complication, the following steps may be specifically performed:
- the maximum Gaussian likelihood estimation algorithm is used to process the second risk factor and the weight of each second risk factor to obtain the weight of the loss function of each target complication.
- the processor 501 may also perform the following steps:
- the processor 501 When the processor 501 obtains the diagnosis and treatment data corresponding to the target disease type of the target user, it may be specifically configured to execute the following steps:
- processor 501 may also perform the following steps:
- the target user terminal sends a prompt message to the target user terminal according to the complication risk information; wherein the prompt message includes information and a treatment plan for indicating the target complication that is at risk.
- the so-called processor 501 may be a central processing unit (Central Processing Unit, CPU), and the processor 501 may also be other general-purpose processors, digital signal processors (Digital Signal Processors, DSPs). ), 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 502 may include a read-only memory and a random access memory, and provides instructions and data to the processor 501. A part of the memory 502 may also include a non-volatile random access memory. For example, the memory 502 may also store diagnosis and treatment data of the user.
- the communication interface 503 may include an input device and/or an output device.
- the input device may be a control panel, a microphone, a receiver, etc.
- the output device may be a display screen, a transmitter, etc., which are not listed here.
- the processor 501, the memory 502, and the communication interface 503 described in the embodiment of the present application can execute the implementation described in the method embodiment described in FIG. 2 or FIG.
- the implementation of the complication risk prediction device described in the embodiment of the present application will not be repeated here.
- An embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, the computer program includes program instructions, and the program instructions can execute the aforementioned complications when executed by a processor.
- the storage medium involved in this application such as a computer-readable storage medium, may be non-volatile or volatile.
- An embodiment of the present application also provides a computer program product, the computer program product includes computer program code, when the computer program code is run on a computer, the computer is caused to execute the above-mentioned complication risk prediction device method embodiment. step.
- the computer-readable storage medium may mainly include a storage program area and a storage data area, where the storage program area may store an operating system, an application program required by at least one function, etc.; the storage data area may store Data created based on the use of blockchain nodes, etc.
- Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
- the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
- the program can be stored in a computer readable storage medium. During execution, it may include the procedures of the above-mentioned method embodiments.
- the storage medium may be a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.
Landscapes
- Medical Informatics (AREA)
- Engineering & Computer Science (AREA)
- Public Health (AREA)
- Health & Medical Sciences (AREA)
- Pathology (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Biomedical Technology (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Medical Treatment And Welfare Office Work (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
一种并发症风险预测***、方法、装置、设备及介质,应用于医疗技术领域,其中,该并发症风险预测***包括:风险预测设备(101)和存储设备(102);其中,该存储设备(102)用于存储用户的诊疗数据;该风险预测设备(101),用于执行以下步骤:获取目标用户的目标疾病类型对应的诊疗数据;结合多个单任务学习模型和多任务学习模型确定多个目标并发症对应的第二风险因子的权重,以及确定每个目标并发症对应的权重,以基于每个第二风险因子的权重和每个目标并发症对应的权重,确定所述目标用户的并发症风险信息。该方法有助于提升并发症预测的可靠性。
Description
本申请要求于2020年9月27日提交中国专利局、申请号为202011034019.X,发明名称为“一种并发症风险预测***、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。
本申请涉及人工智能技术领域,尤其涉及一种并发症风险预测***、方法、装置、设备及介质。
发明人意识到,各种疾病的发病率逐年升高,且患者常常伴有至少一种并发症,而这些并发症往往无法及时发现,导致增加了治疗难度。因此,如何实现并发症的可靠预测成为亟需解决的问题。
发明内容
本申请实施例提供了一种并发症风险预测***、方法、装置、设备及介质,有助于提升并发症预测的可靠性。
第一方面,本申请实施例提供了一种并发症风险预测***,包括:风险预测设备和存储设备;其中,所述存储设备用于存储用户的诊疗数据;
所述风险预测设备,用于执行以下步骤:
从所述存储设备获取目标用户的目标疾病类型对应的诊疗数据;
分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;
利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;
确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;
根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
第二方面,本申请实施例提供了一种并发症风险预测方法,包括:
获取目标用户的目标疾病类型对应的诊疗数据;
分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;
利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;
确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;
根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
第三方面,本申请实施例提供了一种并发症风险预测装置,包括:
获取模块,用于获取目标用户的目标疾病类型对应的诊疗数据;
处理模块,用于分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;
所述处理模块,还用于利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子,并确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;
预测模块,用于根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
第四方面,本申请实施例提供了一种风险预测设备,该风险预测设备可包括处理器和存储器,所述处理器和存储器相互连接。其中,所述存储器用于存储支持终端设备执行上述方法或步骤的计算机程序,所述计算机程序包括程序指令,所述处理器被配置用于调用所述程序指令,执行以下方法:
获取目标用户的目标疾病类型对应的诊疗数据;
分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;
利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;
确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;
根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
第五方面,本申请实施例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存储有计算机程序,所述计算机程序包括程序指令,所述程序指令当被处理器执行时使所述处理器执行以下方法:
获取目标用户的目标疾病类型对应的诊疗数据;
分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;
利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;
确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;
根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
可选的,该计算机可读存储介质可以是非易失性的,也可以是易失性的。
本申请实施例能够结合单任务学习模型和多任务学习模型实现并发症预测,且有助于提升并发症预测的可靠性。
为了更清楚地说明本申请实施例技术方案,下面将对实施例描述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的前提下,还可以根据这些附图获得其他的附图。
图1是本申请实施例提供的一种并发症风险预测***的结构示意图;
图2是本申请实施例提供的一种并发症风险预测方法的流程示意图;
图3是本申请实施例提供的另一种并发症风险预测方法的流程示意图;
图4是本申请实施例提供的一种并发症风险预测装置的结构示意图;
图5是本申请实施例提供的一种风险预测设备的结构示意图。
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本申请保护的范围。
本申请的技术方案可应用于人工智能、数字医疗、智慧城市、区块链和/或大数据技术领域,如具体可涉及机器学习技术,以实现智慧医疗。可选的,本申请涉及的数据如诊疗数据等可存储于数据库中,或者可以存储于区块链中,本申请不做限定。
本申请的技术方案可应用于风险预测***,并可具体应用于风险预测设备(风险预测装置)中,用于实现对并发症风险的预测。可选的,该风险预测设备可以是终端,也可以是服务器,还可以为数据平台或其他设备。该终端可包括手机、平板电脑、计算机等等,本申请不做限定。可以理解,在其他实施例中,该终端还可叫做其余名称,比如叫做终端设备、智能终端、用户设备、用户终端等等,此处不一一列举。
目前,疾病的发病率越来越高,本身治疗难度就很大,而且患者常常伴有并发症。疾病和并发症之间会互相影响,增加治疗难度,从而进入恶性循环,而并发症往往无法及时发现。本申请能够结合多任务学习,实现针对多个并发症的风险预测;而且,相比于聚焦疾病的风险因素分析或者仅针对单个并发症的风险预测,本申请基于多个并发症的风险预测,有助于提升并发症风险预测的可靠性,以实现行全面的针对性治疗。
本申请的技术方案可应用于人工智能、智慧城市、区块链和/或大数据技术领域,涉及的数据可通过区块链节点存储,或者可存储于数据库,本申请不做限定。
本申请实施例提供了一种并发症风险预测***、方法、装置、设备和介质等,使得有助于提升并发症风险预测的可靠性。以下分别详细说明。
请参见图1,是本申请实施例提供的一种并发症风险预测***的结构示意图。如图1所示,该并发症风险预测***可包括风险预测设备(风险预测装置)101和存储设备(存储装置)102。其中,
存储设备102,可用于存储用户的诊疗数据;
风险预测设备101,可用于从该存储设备102中获取目标用户的目标疾病类型对应的诊疗数据;分别利用多个单任务学习模型对该诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;利用多任务学习模型对该多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定该目标用户针对该目标疾病类型的并发症风险信息。
其中,该目标并发症为该目标疾病类型下的并发症,且该目标并发症可以与单任务学习模型一一对应。
可以理解,该存储设备和风险预测设备可以分别为独立的设备,即独立部署,或者,该存储设备和风险预测设备也可以部署于同一设备中,本申请不做限定,图1仅示出了独立部署的场景。例如,在一些实施例中,该存储设备和风险预测设备可部署于服务器中,或者说,该存储设备可以部署于风险预测设备中。
在一些实施例中,该存储设备可以为区块链节点,该诊疗数据可以从区块链获取。也即,各患者的诊疗数据可以预先存储于区块链中。通过从区块链节点中获取用户的诊疗数据,可以提升获取的诊疗数据的可靠性,进而有助于提升基于该诊疗数据确定出的并发症风险的可靠性。
可选的,该诊疗数据可以包括体征数据、检查检验数据等等,可以根据目标疾病类型确定采集的诊疗数据,或者该诊疗数据可以为该目标用户的所有诊疗数据,或者可以为预设时间段内(如最近一年内)的诊疗数据。进一步可选的,该数据可以由监护***中提取,该存储设备为监护***中的存储设备,或者该数据可以由监护***提取后存储于该存储设备,本申请不做限定。
可选的,该诊疗数据可以是对采集的原始医疗数据进行处理得到,该处理包括采样、填充缺失值等等。例如,可获取患者的原始医疗数据,包括患者的历史基线数据,该历史 基站数据可以包括多次就诊记录,每次就诊记录可包括各种诊断、检验、检查、药物、手术项目等。进一步的,可以对该历史基线数据进行预处理,例如,该体征数据可通过对采集的原始体征数据以预设时间单位(如以1h为单位)进行采样得到,该原始体征数据可以为连续数据;又如,可对检查检验数据,如脑利钠肽前体、乳酸等,可使用多次插补(多重插补)填充缺失值。从而得到预处理后的诊疗数据。进一步可选的,该诊疗数据可以为文本数据,也可以向量,如二元特征,或者称为二维特征向量,等等。
可选的,该并发症风险信息可用于指示并发症风险等级、风险指数、发生目标并发症的概率等等,本申请不做限定。
在一些实施例中,风险预测设备101在根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定该目标用户针对该目标疾病类型的并发症风险信息时,可以利用风险预测模型对该第一权重和该第二权重进行处理,以得到该目标用户针对该目标疾病类型的并发症风险信息。其中,该风险预测模型的损失函数可以是基于多个诊疗样本数据得到的目标疾病类型对应的风险因子权重和目标并发症的权重确定出的,或者说,该风险预测模型可以是基于多个诊疗样本数据得到的目标疾病类型对应的风险因子权重和目标并发症的权重训练得到的。
例如,在一些实施例中,风险预测设备101可通过以下方式训练得到风险预测模型:获取多个患者的诊疗样本数据,并分别利用多个单任务学习模型对该诊疗样本数据进行处理,以得到多个目标并发症对应的多个第一风险因子;利用多任务学习模型对该多个第一风险因子进行处理,以从该多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重;确定每个目标并发症对应的权重;根据该每个第二风险因子的权重和每个目标并发症对应的权重,训练得到该风险预测模型。比如可基于每个第二风险因子的权重和每个目标并发症对应的权重,确定出风险预测模型的损失函数,以训练得到该风险预测模型,或者说,可基于每个第二风险因子的权重和每个目标并发症对应的权重,确定出风险预测模型的目标函数,以训练得到该风险预测模型。其中,该风险预测模型可用于预测用户针对该目标疾病类型的并发症风险信息。
可选的,目标并发症对应的权重可以根据该第一风险因子和/或第二风险因子的权重确定出。例如,目标并发症对应的权重可以指目标并发症对应的损失函数的权重;又如,目标并发症对应的权重可以为通过其他方式确定出的目标并发症的权重,如基于目标并发症对应的第一风险因子或第二风险因子的数目设置得到,数目越多,该目标并发症对应的权重可以设置为越大。
也就是说,可以获取多个患者的诊疗数据作为诊疗样本数据(样本数据)。其中,该患者可以是患有目标疾病类型对应的目标并发症的患者。进而可根据该诊疗样本数据确定目标疾病类型对应的多个并发症的风险因子(特征)及其权重。例如,可对诊疗数据变量进行特征工程处理,以得到该目标疾病类型对应的并发症的风险因子及其权重。可选的,该损失函数可以是最小二乘法损失函数。
在一些实施例中,风险预测设备101在利用多任务学习模型对该多个第一风险因子进行处理,以从该多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重时,可以利用L1正则化对该多个第一风险因子进行处理,确定该多个第一风险因子对应的权重矩阵,进而根据该多个第一风险因子对应的权重矩阵进行特征选择,确定该多个第二风险因子对应的权重矩阵,由此可根据该多个第二风险因子对应的权重矩阵,确定该多个第二风险因子以及每个第二风险因子的权重。
在一些实施例中,该目标并发症对应的权重可以是指该目标并发症的损失函数的权重;风险预测设备101在确定每个目标并发症对应的权重时,可以利用最大化高斯似然估计算法对该第二风险因子和每个第二风险因子的权重进行处理,以得到每个目标并发症的损失 函数的权重。
在一些实施例中,该存储设备可以为区块链节点。可选的,该风险预测设备101,还可用于接收目标用户终端发送的风险预测请求,该风险预测请求中可携带该目标用户的标识。进而风险预测设备101可具体用于根据该目标用户的标识从存储设备如该区块链节点获取该诊疗数据。进一步可选的,该风险预测设备101,还可用于根据该并发症风险信息向该目标用户终端发送提示消息。其中,该提示消息可包括用于指示存在风险的目标并发症的信息和治疗方案等等。
可选的,用于指示存在风险的目标并发症的信息可以包括存在风险的一个或多个目标并发症的名称、标识、风险评分、风险等级、概率等信息中的一项或多项。可选的,该治疗方案可以为该目标用户所属的用户分群对应的治疗方案。例如,该目标用户所属的用户分群可以是该治疗方案下净效益最大的分群。例如可基于目标疾病(类型)对应的各治疗方案的净效益实现用户分群,分别得到各治疗方案下净效益最大的用户群。由此在向用户推荐治疗方案时,可以结合净效益进行推送,比如向目标用户推荐目标用户所属的用户分群对应的净效益最大的治疗方案。从而实现为用户提供符合卫生经济学成本效益最优的治疗方案推荐,使得在提供有效治疗的前提下,为患者选择最符合成本效益的治疗方式,有助于减轻患者经济,减轻医保负担。
在本申请实施例中,风险预测***可通过获取用户的目标疾病类型对应的诊疗数据,并分别利用多个单任务学习模型对该诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子,进而利用多任务学习模型对该多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子,并根据每个第二风险因子的权重和每个目标并发症对应的权重,确定该目标用户针对该目标疾病类型的并发症风险信息,由此能够结合单任务学习模型和多任务学习模型实现并发症预测,且有助于提升并发症预测的可靠性。
参见图2,图2是本申请实施例提供的一种并发症风险预测方法的流程示意图。该方法可以由上述的风险预测设备执行,如图2所示,该并发症风险预测方法可包括以下步骤:
201、获取目标用户的目标疾病类型对应的诊疗数据。
其中,该诊疗数据可以包括体征数据、检查检验数据等等。可选的,该诊疗数据可以是对采集的原始医疗数据进行处理得到,此处不赘述。
202、分别利用多个单任务学习模型对该诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子。
其中,该目标并发症为该目标疾病类型下的并发症,且该目标并发症可以与该单任务学习模型一一对应。
以目标疾病为疾病1为例,目标并发症可以为并发症1和并发症2。即可利用并发症1对应的单任务学习模型1对诊疗数据进行处理,以得到并发症1对应的风险因子,以及利用并发症2对应的单任务学习模型2对诊疗数据进行处理,得到并发症2对应的风险因子。
203、利用多任务学习模型对该多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子。
在确定出每个目标并发症的风险因子之后,即可将每个目标并发症筛选出的风险因子输入到多任务学习模型(multitask learning,MTL),对该多个目标并发症的风险因子进行进一步筛选,以得到该目标疾病类型对应的并发症的风险因子,即第二风险因子。
可选的,该第二风险因子的权重(即第一权重)可以是基于L1正则化对第一风险因子进行特征处理得到的。比如可预先训练得到多种疾病类型的多个并发症对应的风险因子,并通过L1正则化进一步对特征进行处理,得到权重矩阵,以得到各个风险因子。
204、确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重。
可选的,目标并发症的权重(即第二权重)可以指该目标并发症的损失函数权重,该 目标并发症的损失函数权重可以基于最大化任务决定的不确定性的高斯似然估计,求导目标并发症损失函数,来确定每个目标并发症的权重。
在一些实施例中,该第一权重和第二权重可以基于历史患者的诊疗样本数据确定出,比如基于历史患者的诊疗样本数据并结合单任务学习模型和多任务学习模型确定出各个第二风险因子的权重以及各并发症对应的权重,并可存储各个第二风险因子及其权重,以及存储各并发症对应的权重。进而在确定出该目标用户对应的第二风险因子之后,即可直接查找该第二风险因子的权重,并可进一步查找各目标并发症对应的权重,以实现对该目标用户的多个目标并发症的风险预测。
在一些实施例中,还可通过其他方式确定该目标用户对应的第二风险因子,比如根据模型训练过程中确定出的第二风险因子集合从目标用户的诊疗数据中筛选出该目标用户对应的第二风险因子。
可选的,本申请涉及的风险因子如第一风险因子、第二风险因子等可以是向量,如二元特征,以便于任务学习模型进行特征筛选。该风险因子还可称为特征、特征向量或其余名称,本申请不做限定。
205、根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定该目标用户针对该目标疾病类型的并发症风险信息。
其中,该并发症风险信息可用于指示针对多个并发症的风险等级、风险指数、发生目标并发症的概率等等,本申请不做限定。
可选的,该并发症风险信息可以是基于上述的第一权重、第二权重和损失函数如最小二乘法损失函数确定出的。
在一些实施例中,风险预测设备基于上述的第一权重、第二权重和风险预测模型,得到该目标用户的并发症风险信息,如可利用风险预测模型对该第一权重和该第二权重进行处理,以得到该目标用户针对该目标疾病类型的并发症风险信息。可选的,该风险预测模型的损失函数是基于多个诊疗样本数据得到的目标疾病类型对应的风险因子权重和目标并发症的权重确定出的。
例如,风险预测设备可获取多个患者的诊疗样本数据,并分别利用多个单任务学习模型对该诊疗样本数据进行处理,以得到多个目标并发症对应的多个第一风险因子;进而可利用多任务学习模型对该多个第一风险因子进行处理,以从该多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重;确定每个目标并发症对应的权重;根据该每个第二风险因子的权重和每个目标并发症对应的权重,确定出该损失函数,以训练得到该风险预测模型。其中,该风险预测模型用于预测用户针对该目标疾病类型的并发症风险信息。
在一些实施例中,风险预测设备在确定第二风险因子及其权重时,可以利用L1正则化对该多个第一风险因子进行处理,以得到该多个第一风险因子对应的权重矩阵;进而可根据该多个第一风险因子对应的权重矩阵进行特征选择,以确定该多个第二风险因子对应的权重矩阵;根据该多个第二风险因子对应的权重矩阵,确定该多个第二风险因子以及每个第二风险因子的权重。
可选的,该目标并发症对应的权重可以为该目标并发症的损失函数的权重,或者可以为其他方式确定出的权重,此处不赘述。例如,在一些实施例中,可以利用最大化高斯似然估计算法对该第二风险因子和每个第二风险因子的权重进行处理,以得到每个目标并发症的损失函数的权重。
在一些实施例中,风险预测设备还可接收目标用户终端发送的风险预测请求,该风险预测请求中携带该目标用户的标识。进而可根据该目标用户的标识从获取该诊疗数据,比如从存储设备如区块链节点获取诊疗数据。可选的,风险预测设备还可根据该并发症风险 信息向该目标用户终端发送提示消息,包括用于指示存在风险的目标并发症的信息和治疗方案等等,此处不赘述。
在本申请实施例中,风险预测设备可通过获取用户的诊疗数据,结合单任务学习模型和多任务学习模型对该诊疗数据进行处理,以得到多个目标疾病类型对应的多个风险因子,进而能够根据该多个风险因子的权重和每个目标并发症对应的权重,实现对用户的多个并发症的风险预测,有助于提升并发症预测的可靠性及全面治疗。
参见图3,图3是本申请实施例提供的另一种并发症风险预测方法的流程示意图,如图所示,该并发症风险预测方法可包括以下步骤:
301、获取多个患者的诊疗样本数据,并分别利用多个单任务学习模型对该诊疗样本数据进行处理,以得到多个目标并发症对应的多个第一风险因子。
可选的,该患者可以是患有目标疾病类型对应的目标并发症的患者。该诊疗数据可以包括体征数据、检查检验数据等等。可选的,该诊疗数据可以是对采集的原始医疗数据进行处理得到。例如,可获取患者的原始医疗数据,包括患者的历史基线数据和结局数据,该历史基站数据可以包括多次就诊记录,每次就诊记录可包括各种诊断、检验、检查、药物、手术项目等,结局数据可以是患者每次就诊记录对应的出院诊断数据等等。进一步的,可以对该历史基线数据进行预处理,比如采用多次插补的方式填充原始医疗数据中的缺失值,从而得到预处理后的诊疗数据。
对于某些指标的检验并不是每个患者都会进行检验,由此可以通过多次插补对诊疗数据中的缺失值进行填充。多次插补是一种基于重复模拟来处理缺失值,其能够从一个包含缺失值的电子病历数据中数据集中生成一组完整的诊疗数据的数据集,如利用用蒙特卡洛方法来填补每个数据集中的缺失数据。通过对原始诊疗数据进行多次插补,有助于提升确定出的风险因子的可靠性,进而提升并发症风险预测的可靠性。
其中,该目标并发症可以为该目标疾病类型下的并发症,且该目标并发症可以与该单任务学习模型一一对应。
302、利用多任务学习模型对该多个第一风险因子进行处理,以从该多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重。
在获取得到诊疗数据之后,即可根据该诊疗数据确定目标疾病类型对应的并发症的风险因子(特征)及其权重。例如,可对诊疗数据变量进行特征工程处理,以得到该目标疾病类型对应的并发症的风险因子及其权重,比如利用目标并发症的单任务学习模型(xgboost)对诊疗数据进行处理,筛选出该目标并发症的风险因子,即第一风险因子。其中,该风险因子可以为该用户诊疗数据的部分数据,也可以为对该诊疗数据进行处理后的数据。
进而可将每个目标并发症筛选出的风险因子输入到多任务学习模型,对单任务学习模型筛选出的风险因子进行处理,得到进一步筛选的风险因子,即第二风险因子,并确定该第二风险因子的权重。比如可通过L1正则化对第一风险因子进行处理,形成稀疏矩阵,得到第二风险因子。其中,只有少数特征(即第二风险因子)对这个模型是有贡献的,绝大部分特征是没有贡献或者贡献微小。例如,得到的第二风险因子对应的权重矩阵可以如下所示:
其中,W=[w
1,w
2,…,w
k]
n×k可以是多任务学习下的特征权重矩阵,即第一风险因子对应的权重矩阵;w
i=[W
i,1,W
i,2,…,W
i,k],相当于对特征矩阵进行了一次按行稀疏化,也就是按行进行特征选择,以得到第二风险因子对应的权重矩阵。由此可以确定出第二风险因子以及第二风险因子的权重,即第一权重。
303、确定每个目标并发症对应的权重。比如每个目标并发症的损失函数的权重(或称为系数)。
考虑每个并发症任务的不确定性,可以应用正交方法(orthogonal approach)来调整每个任务(并发症)的相对权重,比如通过确定出每个任务的损失函数的权重,即第二权重,以确定并发症风险信息。例如,该第二权重可通过以下方式确定出:
其中,f
W(x)可以是指L1正则化之后对于变量(x)的权重结果,即第一权重;σ可以是指数据的方差,也即目标并发症的损失函数权重,即第二权重。
进行基于最大化任务决定的不确定性的高斯似然估计:
例如,以目标疾病类型为疾病1,目标并发症为并发症1和并发症2为例,可获取指示疾病1患者发生并发症1y
1和并发症2y
2的风险的并发症风险信息。其中,这两个任务都服从高斯分布:
基于最大化任务决定的不确定性的高斯似然估计:
求导使其最小,即可得到σ
1和σ
2,分别为并发症1和并发症2的损失函数权重。
可选的,在确定出各第二风险因子的权重和目标并发症对应的权重之后,可以将该第二风险因子的权重和目标并发症对应的权重进行存储,以便于后续快速确定风险因子的权重和并发症权重。进一步可选的,该风险因子如第二风险因子的权重和并发症权重如目标并发症对应的权重可存储于区块链中,以提升存储安全性,进而提升后续获取风险因子的权重和并发症权重的安全性和可靠性。
304、根据该每个第二风险因子的权重和每个目标并发症对应的权重,训练得到风险预测模型。如基于每个第二风险因子的权重和每个目标并发症对应的权重,确定出损失函数,以训练得到风险预测模型。
其中,该风险预测模型可用于预测用户针对该目标疾病类型的并发症风险信息。
其中,
为最小二乘法损失函数,
σ
i是目标并发症的损失函数权重,即第二权重;w
i是风险因子的权重,即第一权重,也即特征权重;X
i是第二风险因子(特征);Y
i是结局,即患者的并发症风险信息;
是第二风险因子对应的权重矩阵。
在一些实施例中,可以针对多种疾病(疾病类型)训练得到该风险预测模型。比如分别针对每一种疾病类型进行模型训练,以得到能够识别多种疾病(疾病类型)下的并发症风险的风险预测模型,此处不赘述。
由此,后续可通过获取患者的诊疗数据,基于诊疗数据对应的风险因子权重和并发症 (损失函数)权重,判断得到患者的并发症风险(结局)。
305、获取目标用户的目标疾病类型对应的诊疗数据。
其中,该目标用户可以为任一进行并发症风险预测的用户。
可选的,该诊疗数据的获取操作可以是通过触发条件触发的。例如,该触发条件可以是接收到针对目标用户的并发症风险预测的请求,即可在接收到针对目标用户的并发症预测的风险请求时,获取该目标用户的诊疗数据。又如,该触发条件可以是接收到入院请求,由此可在接收到目标用户的入院请求时,触发获取该目标用户的诊疗数据,从而可在患者入院之初即可得到高准确度的相关并发症风险预测,以便于为医生提供患者的预后结局的精准可能,实现个性化治疗和疾病管理。还可基于其他触发条件触发获取用户诊疗数据,对于获取该诊疗数据触发条件,本申请不做限定。
306、分别利用多个单任务学习模型对该诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子。
307、利用多任务学习模型对该多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子。
其中,该步骤305-307可参照上述实施例的相关描述,此处不赘述。
308、利用风险预测模型对各第二风险因子的第一权重和各目标并发症对应的第二权重处理,以得到该目标用户针对该目标疾病类型的并发症风险信息。
在确定出该目标用户的诊疗数据对应的第二风险因子之后,即可确定每个第二风险因子的第一权重,并可确定每个目标并发症对应的第二权重。该第一权重可以基于模型训练阶段确定出的相应的第二风险因子的权重确定出,如相同风险因子的权重相同,该第二权重可以基于模型训练阶段确定出的相应的并发症权重确定出,如相同并发症权重(如并发症损失函数权重)相同。
在确定出各第二风险因子的第一权重和各目标并发症对应的第二权重之后,即可利用训练得到的风险预测模型,基于上述的目标函数确定出该目标用户对应的并发症风险信息,如结局Y。
在本申请实施例中,可通过获取多个患者的诊疗样本数据,结合单任务学习模型和多任务学习模型对该诊疗样本数据进行处理,以得到多个目标疾病类型对应的多个风险因子,进而根据该多个风险因子的权重和每个目标并发症对应的权重训练得到风险预测模型,使得后续能够通过获取用户的诊疗数据,确定出该诊疗数据在多个目标疾病类型下对应的多个风险因子的权重以及每个目标并发症对应的权重,并利用风险预测模型实现对用户的多个并发症的风险预测,这就有助于提升并发症预测的可靠性。比如可在患者入院之初即可得到高准确度的相关并发症风险预测,为医生提供患者的预后结局的精准可能,达到个性化治疗和疾病管理的目的,有助于提升预测可靠性以及实现全面治疗。
可以理解,上述方法实施例都是对本申请的并发症预测方法或***的举例说明,对各个实施例的描述都各有侧重,某个实施例中没有详述的部分,可以参见其他实施例的相关描述。
本申请实施例还提供了一种并发症风险预测装置。该装置可包括用于执行前述图2或者图3所述的方法的模块。请参见图4,是本申请实施例提供的一种并发症风险预测装置的结构示意图。本实施例中所描述的并发症风险预测装置,可配置于风险预测设备中,如图4所示,本实施例的并发症风险预测装置400可以包括:获取模块401、处理模块402和预测模块403。其中,
获取模块401,用于获取目标用户的目标疾病类型对应的诊疗数据;
处理模块402,用于分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型 下的并发症,且所述目标并发症与所述单任务学习模型一一对应;
所述处理模块402,还用于利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子,并确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;
预测模块403,用于根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
在一些实施例中,所述预测模块403,可具体用于利用风险预测模型对所述第一权重和所述第二权重进行处理,以得到所述目标用户针对所述目标疾病类型的并发症风险信息;
其中,所述风险预测模型的损失函数是基于多个诊疗样本数据得到的目标疾病类型对应的风险因子权重和目标并发症的权重确定出的。
在一些实施例中,所述获取模块401,还可用于获取多个患者的诊疗样本数据,并分别利用多个单任务学习模型对所述诊疗样本数据进行处理,以得到多个目标并发症对应的多个第一风险因子;
所述处理模块402,还可用于利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重;确定每个目标并发症对应的权重;根据所述每个第二风险因子的权重和每个目标并发症对应的权重,确定出所述损失函数,以训练得到所述风险预测模型,所述风险预测模型用于预测用户针对所述目标疾病类型的并发症风险信息。
在一些实施例中,所述处理模块402在利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重时,可具体用于:
利用L1正则化对所述多个第一风险因子进行处理,确定所述多个第一风险因子对应的权重矩阵;
根据所述多个第一风险因子对应的权重矩阵进行特征选择,确定所述多个第二风险因子对应的权重矩阵;
根据所述多个第二风险因子对应的权重矩阵,确定所述多个第二风险因子以及每个第二风险因子的权重。
在一些实施例中,所述目标并发症对应的权重为该目标并发症的损失函数的权重;所述处理模块402在确定每个目标并发症对应的权重时,可具体用于:
利用最大化高斯似然估计算法对所述第二风险因子和每个第二风险因子的权重进行处理,以得到每个目标并发症的损失函数的权重。
在一些实施例中,获取模块401还可用于接收目标用户终端发送的风险预测请求,所述风险预测请求中携带所述目标用户的标识;
获取模块401,可具体用于根据所述目标用户的标识从所述存储设备如区块链节点获取所述诊疗数据;
预测模块403,还可用于根据所述并发症风险信息向所述目标用户终端发送提示消息;其中,所述提示消息包括用于指示存在风险的目标并发症的信息和治疗方案。
可以理解的是,本实施例的并发症预测装置的各功能模块可根据上述方法实施例图2或者图3中的方法具体实现,其具体实现过程可以参照上述方法实施例图2或者图3的相关描述,此处不再赘述。
请参见图5,图5是本申请实施例提供的一种风险预测设备的结构示意图。如图5所示,该风险预测设备可包括:处理器501和存储器502。可选的,该风险预测设备还可包括通信接口503。上述处理器501、存储器502和通信接口503可通过总线或其他方式连接,在本申请实施例所示图5中以通过总线连接为例。其中,通信接口503可受所述处理器的 控制用于收发消息,存储器502可用于存储计算机程序,所述计算机程序包括程序指令,处理器501用于执行存储器502存储的程序指令。其中,处理器501被配置用于调用所述程序指令执行以下步骤:
获取目标用户的目标疾病类型对应的诊疗数据;
分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;
利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;
确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;
根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
在一些实施例中,处理器501在执行所述根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息时,可具体执行以下步骤:
利用风险预测模型对所述第一权重和所述第二权重进行处理,以得到所述目标用户针对所述目标疾病类型的并发症风险信息;
其中,所述风险预测模型的损失函数是基于多个诊疗样本数据得到的目标疾病类型对应的风险因子权重和目标并发症的权重确定出的。
在一些实施例中,所述处理器501还可用于执行以下步骤:
获取多个患者的诊疗样本数据,并分别利用多个单任务学习模型对所述诊疗样本数据进行处理,以得到多个目标并发症对应的多个第一风险因子;
利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重;
确定每个目标并发症对应的权重;
根据所述每个第二风险因子的权重和每个目标并发症对应的权重,确定出所述损失函数,以训练得到所述风险预测模型,所述风险预测模型用于预测用户针对所述目标疾病类型的并发症风险信息。
在一些实施例中,处理器501在执行所述利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重时,可具体执行以下步骤:
利用L1正则化对所述多个第一风险因子进行处理,确定所述多个第一风险因子对应的权重矩阵;
根据所述多个第一风险因子对应的权重矩阵进行特征选择,确定所述多个第二风险因子对应的权重矩阵;
根据所述多个第二风险因子对应的权重矩阵,确定所述多个第二风险因子以及每个第二风险因子的权重。
在一些实施例中,所述目标并发症对应的权重为该目标并发症的损失函数的权重;所述确定每个目标并发症对应的权重时,可具体执行以下步骤:
利用最大化高斯似然估计算法对所述第二风险因子和每个第二风险因子的权重进行处理,以得到每个目标并发症的损失函数的权重。
在一些实施例中,处理器501还可执行以下步骤:
通过通信接口503接收目标用户终端发送的风险预测请求,所述风险预测请求中携带所述目标用户的标识;
处理器501在获取目标用户的目标疾病类型对应的诊疗数据时,可具体用于执行以下步骤:
根据所述目标用户的标识从所述存储设备如区块链节点获取所述诊疗数据;
可选的,处理器501还可执行以下步骤:
根据所述并发症风险信息向所述目标用户终端发送提示消息;其中,所述提示消息包括用于指示存在风险的目标并发症的信息和治疗方案。
应当理解,在本申请实施例中,所称处理器501可以是中央处理单元(Central Processing Unit,CPU),该处理器501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)、现成可编程门阵列(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。
该存储器502可以包括只读存储器和随机存取存储器,并向处理器501提供指令和数据。存储器502的一部分还可以包括非易失性随机存取存储器。例如,存储器502还可以存储用户的诊疗数据。
该通信接口503可以包括输入设备和/或输出设备,例如该输入设备是可以是控制面板、麦克风、接收器等,输出设备可以是显示屏、发送器等,此处不一一列举。
具体实现中,本申请实施例中所描述的处理器501、存储器502和通信接口503可执行本申请实施例提供的图2或者图3所述的方法实施例所描述的实现方式,也可执行本申请实施例所描述的并发症风险预测装置的实现方式,在此不再赘述。
本申请实施例中还提供一种计算机可读存储介质,所述计算机可读存储介质存储有计算机程序,所述计算机程序包括程序指令,所述程序指令被处理器执行时,可执行上述并发症风险预测方法实施例中所执行的部分或全部步骤,如风险预测设备执行的部分或全部步骤。可选的,本申请涉及的存储介质如计算机可读存储介质可以是非易失性的,也可以是易失性的。
本申请实施例还提供一种计算机程序产品,所述计算机程序产品包括计算机程序代码,当所述计算机程序代码在计算机上运行时,使得计算机执行上述并发症风险预测装置方法实施例中所执行的步骤。
在一些实施例中,所述的计算机可读存储介质可主要包括存储程序区和存储数据区,其中,存储程序区可存储操作***、至少一个功能所需的应用程序等;存储数据区可存储根据区块链节点的使用所创建的数据等。
其中,本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain),本质上是一个去中心化的数据库,是一串使用密码学方法相关联产生的数据块,每一个数据块中包含了一批次网络交易的信息,用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机程序来指令相关的硬件来完成,所述的程序可存储于一计算机可读取存储介质中,该程序在执行时,可包括如上述各方法的实施例的流程。其中,所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory,ROM)或随机存储记忆体(Random Access Memory,RAM)等。
以上所揭露的仅为本申请一种较佳实施例而已,当然不能以此来限定本申请之权利范围,本领域普通技术人员可以理解实现上述实施例的全部或部分流程,并依本申请权利要求所作的等同变化,仍属于发明所涵盖的范围。
Claims (20)
- 一种并发症风险预测***,包括:风险预测设备和存储设备;其中,所述存储设备用于存储用户的诊疗数据;所述风险预测设备,用于执行以下步骤:从所述存储设备获取目标用户的目标疾病类型对应的诊疗数据;分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
- 根据权利要求1所述的***,其中,所述根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息,包括:利用风险预测模型对所述第一权重和所述第二权重进行处理,以得到所述目标用户针对所述目标疾病类型的并发症风险信息;其中,所述风险预测模型的损失函数是基于多个诊疗样本数据得到的目标疾病类型对应的风险因子权重和目标并发症的权重确定出的。
- 根据权利要求2所述的***,其中,所述风险预测设备,还用于执行以下步骤:获取多个患者的诊疗样本数据,并分别利用多个单任务学习模型对所述诊疗样本数据进行处理,以得到多个目标并发症对应的多个第一风险因子;利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重;确定每个目标并发症对应的权重;根据所述每个第二风险因子的权重和每个目标并发症对应的权重,确定出所述损失函数,以训练得到所述风险预测模型,所述风险预测模型用于预测用户针对所述目标疾病类型的并发症风险信息。
- 根据权利要求3所述的***,其中,所述利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重,包括:利用L1正则化对所述多个第一风险因子进行处理,确定所述多个第一风险因子对应的权重矩阵;根据所述多个第一风险因子对应的权重矩阵进行特征选择,确定所述多个第二风险因子对应的权重矩阵;根据所述多个第二风险因子对应的权重矩阵,确定所述多个第二风险因子以及每个第二风险因子的权重。
- 根据权利要求3所述的***,其中,所述目标并发症对应的权重为该目标并发症的损失函数的权重;所述确定每个目标并发症对应的权重,包括:利用最大化高斯似然估计算法对所述第二风险因子和每个第二风险因子的权重进行处理,以得到每个目标并发症的损失函数的权重。
- 根据权利要求1-5任一项所述的***,其中,所述存储设备为区块链节点;所述风险预测设备,还用于接收目标用户终端发送的风险预测请求,所述风险预测请 求中携带所述目标用户的标识;所述风险预测设备,具体用于根据所述目标用户的标识从所述区块链节点获取所述诊疗数据;所述风险预测设备,还用于根据所述并发症风险信息向所述目标用户终端发送提示消息;其中,所述提示消息包括用于指示存在风险的目标并发症的信息和治疗方案。
- 一种并发症风险预测方法,包括:获取目标用户的目标疾病类型对应的诊疗数据;分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
- 根据权利要求7所述的方法,其中,所述根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息,包括:利用风险预测模型对所述第一权重和所述第二权重进行处理,以得到所述目标用户针对所述目标疾病类型的并发症风险信息;其中,所述风险预测模型的损失函数是基于多个诊疗样本数据得到的目标疾病类型对应的风险因子权重和目标并发症的权重确定出的。
- 根据权利要求8所述的方法,其中,所述方法还包括:获取多个患者的诊疗样本数据,并分别利用多个单任务学习模型对所述诊疗样本数据进行处理,以得到多个目标并发症对应的多个第一风险因子;利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重;确定每个目标并发症对应的权重;根据所述每个第二风险因子的权重和每个目标并发症对应的权重,确定出所述损失函数,以训练得到所述风险预测模型,所述风险预测模型用于预测用户针对所述目标疾病类型的并发症风险信息。
- 根据权利要求9所述的方法,其中,所述利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重,包括:利用L1正则化对所述多个第一风险因子进行处理,确定所述多个第一风险因子对应的权重矩阵;根据所述多个第一风险因子对应的权重矩阵进行特征选择,确定所述多个第二风险因子对应的权重矩阵;根据所述多个第二风险因子对应的权重矩阵,确定所述多个第二风险因子以及每个第二风险因子的权重。
- 根据权利要求9所述的方法,其中,所述目标并发症对应的权重为该目标并发症的损失函数的权重;所述确定每个目标并发症对应的权重,包括:利用最大化高斯似然估计算法对所述第二风险因子和每个第二风险因子的权重进行处理,以得到每个目标并发症的损失函数的权重。
- 一种并发症风险预测装置,包括:获取模块,用于获取目标用户的目标疾病类型对应的诊疗数据;处理模块,用于分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;所述处理模块,还用于利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子,并确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;预测模块,用于根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
- 一种风险预测设备,包括处理器和存储器,所述处理器和所述存储器相互连接,其中,所述存储器用于存储计算机程序,所述计算机程序包括程序指令,所述处理器被配置用于调用所述程序指令,执行以下方法:获取目标用户的目标疾病类型对应的诊疗数据;分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
- 根据权利要求13所述的风险预测设备,其中,所述根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息时,具体执行:利用风险预测模型对所述第一权重和所述第二权重进行处理,以得到所述目标用户针对所述目标疾病类型的并发症风险信息;其中,所述风险预测模型的损失函数是基于多个诊疗样本数据得到的目标疾病类型对应的风险因子权重和目标并发症的权重确定出的。
- 根据权利要求14所述的风险预测设备,其中,所述处理器还用于执行:获取多个患者的诊疗样本数据,并分别利用多个单任务学习模型对所述诊疗样本数据进行处理,以得到多个目标并发症对应的多个第一风险因子;利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重;确定每个目标并发症对应的权重;根据所述每个第二风险因子的权重和每个目标并发症对应的权重,确定出所述损失函数,以训练得到所述风险预测模型,所述风险预测模型用于预测用户针对所述目标疾病类型的并发症风险信息。
- 根据权利要求15所述的风险预测设备,其中,所述利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重时,具体执行:利用L1正则化对所述多个第一风险因子进行处理,确定所述多个第一风险因子对应的权重矩阵;根据所述多个第一风险因子对应的权重矩阵进行特征选择,确定所述多个第二风险因 子对应的权重矩阵;根据所述多个第二风险因子对应的权重矩阵,确定所述多个第二风险因子以及每个第二风险因子的权重。
- 一种计算机可读存储介质,所述计算机可读存储介质存储有计算机程序,所述计算机程序包括程序指令,所述程序指令当被处理器执行时使所述处理器执行以下方法:获取目标用户的目标疾病类型对应的诊疗数据;分别利用多个单任务学习模型对所述诊疗数据进行处理,以得到多个目标并发症中每个目标并发症的第一风险因子;所述目标并发症为所述目标疾病类型下的并发症,且所述目标并发症与所述单任务学习模型一一对应;利用多任务学习模型对所述多个目标并发症对应的第一风险因子进行处理,以得到多个第二风险因子;确定每个第二风险因子的第一权重,以及确定每个目标并发症对应的第二权重;根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息。
- 根据权利要求17所述的计算机可读存储介质,其中,所述根据每个第二风险因子的第一权重和每个目标并发症对应的第二权重,确定所述目标用户针对所述目标疾病类型的并发症风险信息时,具体执行:利用风险预测模型对所述第一权重和所述第二权重进行处理,以得到所述目标用户针对所述目标疾病类型的并发症风险信息;其中,所述风险预测模型的损失函数是基于多个诊疗样本数据得到的目标疾病类型对应的风险因子权重和目标并发症的权重确定出的。
- 根据权利要求18所述的计算机可读存储介质,其中,所述程序指令当被处理器执行时还用于使所述处理器执行:获取多个患者的诊疗样本数据,并分别利用多个单任务学习模型对所述诊疗样本数据进行处理,以得到多个目标并发症对应的多个第一风险因子;利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重;确定每个目标并发症对应的权重;根据所述每个第二风险因子的权重和每个目标并发症对应的权重,确定出所述损失函数,以训练得到所述风险预测模型,所述风险预测模型用于预测用户针对所述目标疾病类型的并发症风险信息。
- 根据权利要求19所述的计算机可读存储介质,其中,所述利用多任务学习模型对所述多个第一风险因子进行处理,以从所述多个第一风险因子中确定出多个第二风险因子,并确定每个第二风险因子的权重时,具体执行:利用L1正则化对所述多个第一风险因子进行处理,确定所述多个第一风险因子对应的权重矩阵;根据所述多个第一风险因子对应的权重矩阵进行特征选择,确定所述多个第二风险因子对应的权重矩阵;根据所述多个第二风险因子对应的权重矩阵,确定所述多个第二风险因子以及每个第二风险因子的权重。
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011034019.XA CN111933284B (zh) | 2020-09-27 | 2020-09-27 | 一种并发症风险预测***、方法、装置、设备及介质 |
CN202011034019.X | 2020-09-27 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021179630A1 true WO2021179630A1 (zh) | 2021-09-16 |
Family
ID=73334241
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/124611 WO2021179630A1 (zh) | 2020-09-27 | 2020-10-29 | 一种并发症风险预测***、方法、装置、设备及介质 |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN111933284B (zh) |
WO (1) | WO2021179630A1 (zh) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114999659A (zh) * | 2022-04-26 | 2022-09-02 | 北京市农林科学院信息技术研究中心 | 并发症风险预警方法、装置、电子设备及存储介质 |
CN115798711A (zh) * | 2022-12-22 | 2023-03-14 | 之江实验室 | 基于反事实对比学习的慢性肾病诊疗决策支持*** |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113782216B (zh) * | 2021-09-15 | 2023-10-24 | 平安科技(深圳)有限公司 | 一种失能权重确定方法及装置、电子设备、存储介质 |
CN114255872A (zh) * | 2021-12-14 | 2022-03-29 | 安徽科大讯飞医疗信息技术有限公司 | 一种风险预测方法、装置、设备及存储介质 |
CN114913982B (zh) | 2022-07-18 | 2022-10-11 | 之江实验室 | 基于对比学习的终末期肾病并发症风险预测*** |
CN116434968B (zh) * | 2023-06-14 | 2023-08-25 | 之江实验室 | 一种层级聚类约束的多标签细粒度术后并发症预测装置 |
CN116825356B (zh) * | 2023-07-12 | 2024-02-06 | 中国医学科学院基础医学研究所 | 多关联手术并发症风险评估方法、***及计算设备 |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110202486A1 (en) * | 2009-07-21 | 2011-08-18 | Glenn Fung | Healthcare Information Technology System for Predicting Development of Cardiovascular Conditions |
US20120116799A1 (en) * | 2009-05-08 | 2012-05-10 | Sven Lindskog | System for assessing risk for progression or development of periodontitis for a patent |
CN106682434A (zh) * | 2016-12-30 | 2017-05-17 | 深圳中科金证科技有限公司 | 患病风险评估方法及装置 |
CN109657696A (zh) * | 2018-11-05 | 2019-04-19 | 阿里巴巴集团控股有限公司 | 多任务监督学习模型训练、预测方法和装置 |
CN111710420A (zh) * | 2020-05-15 | 2020-09-25 | 深圳先进技术研究院 | 一种基于电子病历大数据的并发症发病风险预测方法、***、终端以及存储介质 |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2013012990A1 (en) * | 2011-07-20 | 2013-01-24 | Siemens Corporation | Multi-task learning for bayesian matrix factorization |
CN109785971B (zh) * | 2019-01-30 | 2023-05-23 | 华侨大学 | 一种基于先验医学知识的疾病风险预测方法 |
CN111063442B (zh) * | 2019-11-28 | 2022-07-15 | 南京邮电大学 | 基于弱监督多任务矩阵补全的脑疾病进程预测方法及*** |
CN111260209B (zh) * | 2020-01-14 | 2022-03-11 | 山东大学 | 电子病历与医学影像结合的心血管疾病风险预测评估*** |
CN111276242A (zh) * | 2020-01-20 | 2020-06-12 | 吉林大学 | 一种针对医院重症监护室患者疾病诊断与病情状态评估建模方法 |
-
2020
- 2020-09-27 CN CN202011034019.XA patent/CN111933284B/zh active Active
- 2020-10-29 WO PCT/CN2020/124611 patent/WO2021179630A1/zh active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120116799A1 (en) * | 2009-05-08 | 2012-05-10 | Sven Lindskog | System for assessing risk for progression or development of periodontitis for a patent |
US20110202486A1 (en) * | 2009-07-21 | 2011-08-18 | Glenn Fung | Healthcare Information Technology System for Predicting Development of Cardiovascular Conditions |
CN106682434A (zh) * | 2016-12-30 | 2017-05-17 | 深圳中科金证科技有限公司 | 患病风险评估方法及装置 |
CN109657696A (zh) * | 2018-11-05 | 2019-04-19 | 阿里巴巴集团控股有限公司 | 多任务监督学习模型训练、预测方法和装置 |
CN111710420A (zh) * | 2020-05-15 | 2020-09-25 | 深圳先进技术研究院 | 一种基于电子病历大数据的并发症发病风险预测方法、***、终端以及存储介质 |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114999659A (zh) * | 2022-04-26 | 2022-09-02 | 北京市农林科学院信息技术研究中心 | 并发症风险预警方法、装置、电子设备及存储介质 |
CN115798711A (zh) * | 2022-12-22 | 2023-03-14 | 之江实验室 | 基于反事实对比学习的慢性肾病诊疗决策支持*** |
CN115798711B (zh) * | 2022-12-22 | 2023-08-29 | 之江实验室 | 基于反事实对比学习的慢性肾病诊疗决策支持*** |
Also Published As
Publication number | Publication date |
---|---|
CN111933284B (zh) | 2021-01-05 |
CN111933284A (zh) | 2020-11-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021179630A1 (zh) | 一种并发症风险预测***、方法、装置、设备及介质 | |
WO2021180244A1 (zh) | 一种疾病风险预测***、方法、装置、设备及介质 | |
CN112017789B (zh) | 分诊数据处理方法、装置、设备及介质 | |
CN112216361A (zh) | 基于人工智能的随访计划单生成方法、装置、终端及介质 | |
CN112102950B (zh) | 一种数据处理***、方法、装置及存储介质 | |
EP3826027A1 (en) | Event data modelling | |
US11152123B1 (en) | Processing brain data using autoencoder neural networks | |
CN111814146B (zh) | 针对对象的关联关系建立方法、装置、服务器及存储介质 | |
CN112397195B (zh) | 用于生成体格检查模型的方法、装置、电子设备和介质 | |
CN113724878B (zh) | 基于机器学习的医疗风险信息推送方法及装置 | |
JP2024523080A (ja) | 卵巣刺激を最適化するための機械学習 | |
US20210335491A1 (en) | Predictive adaptive intelligent diagnostics and treatment | |
Hammed et al. | Unlocking the potential of autism detection: integrating traditional feature selection and machine learning techniques | |
CN112017788B (zh) | 基于强化学习模型的疾病排序方法、装置、设备及介质 | |
CN112017784B (zh) | 一种基于多模态数据的冠心病风险预测方法及相关设备 | |
CN110415779A (zh) | 保温措施有效性检测方法、装置、设备及存储介质 | |
CN113724824B (zh) | 慢性病患者随访方法、装置、计算机设备及可读存储介质 | |
CN113066531B (zh) | 风险预测方法、装置、计算机设备及存储介质 | |
WO2021151330A1 (zh) | 用户分群方法、装置、设备和计算机可读存储介质 | |
CN113609295A (zh) | 文本分类方法、装置及相关设备 | |
CN113760367A (zh) | 一种数据处理方法、装置、电子设备以及存储介质 | |
CN112394924A (zh) | 用于生成提问模型的方法、装置、电子设备和介质 | |
Pivithuru et al. | E-patient Card: An Integrated Electronic Health Recording System for Patient | |
WO2022227164A1 (zh) | 基于人工智能的数据处理方法、装置、设备和介质 | |
US11842801B2 (en) | Systems and methods for guiding traversal through logic series for event chains |
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: 20924925 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: 20924925 Country of ref document: EP Kind code of ref document: A1 |