Coronary heart disease prescription generation equipment for executing multi-task deep learning algorithm
Through multi-task deep learning algorithm combined with single-task integrated learning model, the problem of rationality and overfitting of traditional Chinese medicine is solved, high-precision coronary heart disease prescription generation is achieved, and the modernization and intelligent development of traditional Chinese medicine is promoted.
Patent Information
- Application Number
- CN202510384899.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-28
- Publication Date
- 2025-07-11
AI Technical Summary
In the prior art, single-task machine learning models ignore the synergistic effects of traditional Chinese medicine, resulting in a decrease in the rationality of the compatibility. Traditional deep learning models are prone to overfitting in the case of high-dimensional small samples of traditional Chinese medicine data, making it difficult to effectively generate standardized and personalized traditional Chinese medicine prescriptions for coronary heart disease.
A multi-task deep learning algorithm is adopted, combining a single-task integrated learning model and a multi-task deep learning model, and a coronary heart disease prescription is generated through feature fusion and parameter optimization, taking into account the synergy of traditional Chinese medicine to prevent overfitting.
It improves the accuracy of Chinese medicine dosage prediction, ensures reasonable compatibility, prevents overfitting, provides efficient and standardized traditional Chinese medicine prescriptions for coronary heart disease, and promotes the intelligent application of traditional Chinese medicine.
Smart Images
Figure CN120299636A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer technology. More specifically, it relates to a coronary heart disease prescription generation device that executes a multi-task deep learning algorithm. Background Art
[0002] Traditional Chinese medicine diagnosis and treatment highly relies on the experience of physicians. Renowned traditional Chinese medicine doctors have unique diagnosis and treatment ideas and methods through long-term clinical practice, and their diagnosis and treatment experience has high clinical value and guiding significance. And using modern technology to inherit the diagnosis and treatment experience of renowned traditional Chinese medicine doctors is one of the directions for the development of modern traditional Chinese medicine. However, in the related art, the method of using a single-task machine learning model to independently predict each drug dosage ignores the synergistic effect of traditional Chinese medicines and there is a risk of reducing the rationality of compatibility; in addition, when directly training with a traditional deep learning model end-to-end, overfitting problems are prone to occur in traditional Chinese medicine data with the characteristics of high dimension and small samples. Summary of the Invention
[0003] An object of the present invention is to provide a coronary heart disease prescription generation device that executes a multi-task deep learning algorithm to solve at least one of the problems existing in the prior art.
[0004] To achieve the above object, the present invention adopts the following technical solutions:
[0005] The present invention provides a coronary heart disease prescription generation device that executes a multi-task deep learning algorithm, including a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor is configured to execute the following steps when loading the computer program:
[0006] Obtain coronary heart disease medical record data, and process the coronary heart disease medical record data to obtain a first feature vector;
[0007] Input the first feature vector into multiple pre-trained single-task ensemble learning models, and output the prediction results of each traditional Chinese medicine dosage as a second feature vector;
[0008] Perform normalization processing on the first feature vector, and perform feature fusion on the normalized first feature vector and the second feature vector to obtain a third feature vector;
[0009] Input the third feature vector into a pre-trained multi-task deep learning model to obtain each traditional Chinese medicine dosage, and obtain a coronary heart disease prescription according to each traditional Chinese medicine dosage.
[0010] Optionally, the processing the coronary heart disease medical record data to obtain a first feature vector includes:
[0011] Extract data from the coronary heart disease medical record data to obtain a data extraction result;
[0012] Perform data cleaning on the data extraction result to obtain the first feature vector.
[0013] Optionally, before inputting the first feature vector into multiple pre-trained single-task ensemble learning models, the method further includes:
[0014] Obtain historical medical records data of coronary heart disease;
[0015] Perform data extraction and data cleaning on the historical medical records data of coronary heart disease to obtain a fourth feature vector, where the fourth feature vector includes symptoms, signs, auxiliary examination results, currently taken medications, and prescribed traditional Chinese medicines;
[0016] Use the traditional Chinese medicines and their dosages in the prescribed traditional Chinese medicines as the dependent variables to establish multiple single-task class boosting regression models;
[0017] Use the fourth feature vector as a training set to train the multiple single-task class boosting regression models;
[0018] Use a parameter optimization algorithm to optimize the parameters of the trained multiple single-task class boosting regression models to obtain multiple pre-trained single-task ensemble learning models.
[0019] Optionally, after using the parameter optimization algorithm to optimize the parameters of the trained multiple single-task class boosting regression models, the method further includes:
[0020] Use the fourth feature vector as a test set to test the multiple single-task class boosting regression models with optimized parameters;
[0021] Determine whether the coefficient of determination in the test result is less than or equal to zero;
[0022] If so, add an attention mechanism to train and optimize the parameters of the multiple single-task class boosting regression models.
[0023] Optionally, the determination of whether the coefficient of determination in the test result is less than or equal to zero further includes:
[0024] If not, the multiple single-task class boosting regression models with optimized parameters are multiple pre-trained single-task ensemble learning models.
[0025] Optionally, after obtaining the multiple pre-trained single-task ensemble learning models, the method further includes:
[0026] Input the fourth feature vector into the multiple pre-trained single-task ensemble learning models, and output the predicted results of each traditional Chinese medicine dosage as a fifth feature vector;
[0027] Normalize the fourth feature vector, and perform feature fusion on the normalized fourth feature vector and the fifth feature vector to obtain a sixth feature vector.
[0028] Optionally, before inputting the third feature vector into a pre-trained multi-task deep learning model to obtain the doses of various traditional Chinese medicines, the method further includes:
[0029] Build a multi-task deep learning model, and use the sixth feature vector as a training set to train the multi-task deep learning model;
[0030] Optimize the parameters of the trained multi-task deep learning model to obtain the input feature dimension, the number of tasks, the number of neurons in each layer, the dropout rate in each layer, and the learning rate;
[0031] Construct a hybrid architecture for the multi-task deep learning model with optimized parameters to obtain a pre-trained multi-task deep learning model, where the pre-trained multi-task deep learning model includes a feature cross layer, a compatibility relationship learning layer, a dose synergy prediction layer, and a multi-task output layer.
[0032] Optionally, use a dynamic early stopping mechanism and step learning rate decay to train the multi-task deep learning model.
[0033] Optionally, the compatibility relationship learning layer includes a compatibility taboo screening module, and the compatibility taboo screening module is used to detect the compatibility taboos of traditional Chinese medicines in the prescription.
[0034] Optionally, the multi-task output layer includes multiple parallel linear units.
[0035] The beneficial effects of the present invention are as follows:
[0036] The technical solution of the present invention can improve the prediction accuracy of the doses of various traditional Chinese medicines by using a single-task ensemble learning model and a multi-task deep learning model; can fully consider the synergistic effect of traditional Chinese medicines and avoid the risk of reduced compatibility rationality; can prevent the problem of overfitting; can combine the advantages of traditional Chinese medicine compatibility rules and intelligent decision-making, provide efficient and standardized tertiary expert-level prescription support for primary medical care; effectively assist primary doctors in generating standardized and personalized traditional Chinese medicine treatment plans for coronary heart disease; can transform traditional Chinese medicine expert experience into a quantifiable and compliant artificial intelligence diagnosis and treatment model, which not only promotes the inheritance and modern transformation of traditional Chinese medicine knowledge, but also promotes the standardized development of the intelligent application of traditional Chinese medicine in the global medical system. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] The following further elaborates on the specific implementation manners of the present invention in conjunction with the accompanying drawings.
[0038] Figure 1The flowchart of the coronary heart disease prescription generation method based on multi-task deep learning provided by the embodiments of the present invention is shown.
[0039] Figure 2 The structural schematic diagram of the computer device provided by the embodiments of the present invention is shown. Detailed implementation manners
[0040] To illustrate the present invention more clearly, the present invention will be further described below in conjunction with embodiments and drawings. Similar components in the drawings are denoted by the same reference numerals. Those skilled in the art should understand that the content specifically described below is illustrative rather than restrictive, and should not be used to limit the protection scope of the present invention.
[0041] Traditional Chinese medicine diagnosis and treatment highly rely on the experience of physicians. Renowned traditional Chinese medicine doctors have unique diagnosis and treatment ideas and methods through long-term clinical practice. Using modern technology to inherit the diagnosis and treatment experience of renowned traditional Chinese medicine doctors is one of the problems that need to be solved in the development of modern traditional Chinese medicine. In related technologies, using a single-task machine learning model to independently predict each drug dosage ignores the synergistic effect of traditional Chinese medicines and there is a risk of reducing the compatibility rationality; while traditional deep learning models are also affected by the characteristics of high-dimensional data and small samples, and are prone to overfitting risks.
[0042] In view of this, as Figure 1 shown, an embodiment of the present invention provides a coronary heart disease prescription generation method based on multi-task deep learning. This method is executed by a processor loading a computer program stored in a memory, and includes the following steps: obtaining coronary heart disease medical record data, and processing the coronary heart disease medical record data to obtain a first feature vector; inputting the first feature vector into multiple pre-trained single-task ensemble learning models, and outputting each traditional Chinese medicine dosage prediction result as a second feature vector; performing normalization processing on the first feature vector, and performing feature fusion on the normalized first feature vector and the second feature vector to obtain a third feature vector; inputting the third feature vector into a pre-trained multi-task deep learning model to obtain each traditional Chinese medicine dosage, and obtaining a coronary heart disease prescription according to each traditional Chinese medicine dosage.
[0043] In a specific example, this method includes performing structured processing on coronary heart disease medical record data to generate a first feature vector including symptoms and signs; inputting the first feature vector into a group of pre-trained ensemble learning models (i.e., multiple pre-trained single-task ensemble learning models), and outputting each traditional Chinese medicine single-task dosage prediction result as a second feature vector; performing normalization processing on the first feature vector, performing feature fusion with the second feature vector, and generating a third feature vector for enhancing features; inputting the third feature vector into a cascaded deep neural network (i.e., a multi-task deep learning model) to output the traditional Chinese medicine dosage in the prescription.
[0044] Furthermore, using the medical record data of Professor Wang Jie in treating coronary heart disease, after data cleaning and assignment, it is organized into the first feature vector; the first feature vector is input into a category boosting (CatBoost) regression model group (i.e., multiple pre-trained single-task ensemble learning models) to predict the doses of traditional Chinese medicines in a single-task prescription, and the output of the model prediction result is the second feature vector; the first feature vector and the second feature vector are fused using feature enhancement to be organized into the third feature vector; the third feature vector is input into a TensorFlow deep learning model (i.e., a multi-task deep learning model) for multi-task learning optimization prediction.
[0045] Furthermore, 580 medical records of Professor Wang Jie in treating coronary heart disease are collected, data extraction and cleaning are performed on the medical record data, and the symptoms, signs, auxiliary examination results, and current medication data after data cleaning are organized into the first feature vector; the traditional Chinese medicines and their doses after data cleaning are organized into the dependent variable, and a single-task CatBoost regression model group is constructed; the first feature vector is standardized and randomly divided into a training set and a test set with a ratio of 8:2; the parameters of each single-task CatBoost regression model are optimized; the single-task CatBoost regression model is trained with the optimal parameters to construct a single-task CatBoost regression model; the test set is used to evaluate the single-task model, and according to the model evaluation results, an attention mechanism is added to improve the performance of the single-task CatBoost regression model.
[0046] Furthermore, the output of the single-task CatBoost regression model is the prediction results of the doses of each traditional Chinese medicine in a single task, and the prediction results of the doses of each traditional Chinese medicine in a single task are used as the second feature vector.
[0047] Furthermore, the standardized first feature vector and the second feature vector generated by the prediction results of the CatBoost regression model are fused to generate the third feature vector with enhanced features.
[0048] Furthermore, a multi-task deep learning model is constructed.
[0049] Furthermore, the third feature vector is randomly divided into a training set and a test set with a ratio of 8:2.
[0050] Furthermore, the parameters of the multi-task deep learning model are optimized, and the input feature dimension, the number of model tasks, the number of neurons in each layer of the model, the dropout rate of each layer of the model, and the learning rate of the model are configured.
[0051] Furthermore, a hybrid architecture is constructed for the multi-task deep learning model, the port layer is defined, the input features are passed into the multi-task deep learning model, and the feature cross layer, the compatibility relationship learning layer, the dose collaborative prediction layer, and the multi-task output layer are set.
[0052] Further, in the hybrid architecture of the multi-task deep learning model, a fully connected layer (Dense) is used for feature crossing, the rectified linear unit (ReLU) activation function is applied, batch normalization is performed using batch normalization (Batch Normalization), and a dropout layer is used to prevent overfitting.
[0053] Further, in the multi-task deep learning model, the fit method is used to train the model, and callbacks for early stopping and learning rate adjustment are applied.
[0054] Further, in the multi-task deep learning model, the third feature vector test set is used to evaluate the model, and the mean absolute error is used as the evaluation metric for each task.
[0055] In this embodiment, by using a single-task ensemble learning model and a multi-task deep learning model, the prediction accuracy of each traditional Chinese medicine dosage can be improved; the synergistic effect of traditional Chinese medicine can be fully considered, and the risk of reduced compatibility rationality can be avoided; the problem of overfitting can be prevented; the advantages of traditional Chinese medicine compatibility rules and intelligent decision-making can be combined to provide efficient and standardized tertiary expert-level prescription support for primary care; effectively assist primary care doctors in generating standardized and personalized traditional Chinese medicine treatment plans for coronary heart disease; and the experience of traditional Chinese medicine experts can be transformed into a quantifiable and compliant artificial intelligence diagnosis and treatment model, which not only promotes the inheritance and modern transformation of traditional Chinese medicine knowledge, but also promotes the standardized development of the intelligent application of traditional Chinese medicine in the global medical system.
[0056] In a possible implementation manner, the processing of the coronary heart disease medical record data to obtain the first feature vector includes: performing data extraction on the coronary heart disease medical record data to obtain a data extraction result, where the data extraction result includes symptoms, signs, auxiliary examination results, currently taking medications, and prescribed traditional Chinese medicines; and performing data cleaning on the data extraction result to obtain the first feature vector, where the first feature vector includes the symptoms, signs, auxiliary examination results, and currently taking medications after data cleaning.
[0057] In a specific example, 580 medical records of Professor Wang Jie's treatment of coronary heart disease are collected, and data extraction and data cleaning are performed on the medical record data. The data extraction content mainly includes the symptoms, signs, auxiliary examination results, currently taking medications, and prescribed traditional Chinese medicines of the patients.
[0058] Further, the names of traditional Chinese medicines in the prescriptions are uniformly standardized with reference to the Pharmacopoeia of the People's Republic of China.
[0059] Further, data cleaning includes merging symptoms and signs with small quantities and the same pathogenesis, merging currently taking medications with the same pharmacological mechanism, deleting prescriptions with unclear records, and checking and verifying missing values.
[0060] Further, the symptoms, signs, auxiliary examination results, and current medication data after data cleaning are organized into a first feature vector.
[0061] In a possible implementation, before inputting the first feature vector into multiple pre-trained single-task ensemble learning models, the method further includes: obtaining historical coronary heart disease medical record data; performing data extraction and data cleaning on the historical coronary heart disease medical record data to obtain a fourth feature vector, where the fourth feature vector includes symptoms, signs, auxiliary examination results, current medications, and traditional Chinese medicines in prescriptions; using the traditional Chinese medicines and their dosages in the prescriptions as dependent variables to establish multiple single-task class boosting regression models; using the fourth feature vector as a training set to train the multiple single-task class boosting regression models; using a parameter optimization algorithm to optimize the parameters of the trained multiple single-task class boosting regression models to obtain multiple pre-trained single-task ensemble learning models.
[0062] In a specific example, the traditional Chinese medicines and their dosages after data cleaning are organized as dependent variables to construct a single-task CatBoost regression model group.
[0063] Further, the fourth feature vector is subjected to data standardization processing and randomly divided into a training set and a test set with a ratio of 8:2.
[0064] Further, parameter optimization is performed on each single-task CatBoost regression model. For example, Optuna is used for hyperparameter search, and the optimization direction is to minimize the root mean square error (RMSE). The parameter ranges include: the range of the number of iterations (iterations) is from 500 to 2000, the range of the tree depth (depth) is from 2 to 16, the range of the learning rate (learning_rate) is from 0.001 to 0.05, and the range of the random perturbation strength (random_strength) is from 0.5 to 10.
[0065] Further, the single-task CatBoost regression model is trained using the optimal parameters to construct a single-task CatBoost regression model.
[0066] Further, the test set is used to evaluate the single-task model, and the evaluation metrics include: the coefficient of determination R 2 , mean squared error, explained variance, and mean absolute error (MAE).
[0067] Further, if the model evaluation result is poor, such as the coefficient of determination R 2 ≤0, then an attention mechanism is added to improve the performance of the single-task CatBoost regression model.
[0068] The R of the single-task CatBoost regression prediction model in this embodiment2 The value is 0.1758, the mean squared error value is 20.3569, the explained variance is 0.1802, and the mean absolute error value is 4.1726.
[0069] In a specific example, the ensemble learning model group is a CatBoost regression model group, and the number N of single-task models independently trained is N≥180; each single-task CatBoost regression model performs its own feature engineering and generates its own optimal parameters through dynamic parameter optimization; an attention mechanism is added to improve the performance of the single-task models.
[0070] Furthermore, the output result of the CatBoost regression model group is used as the second feature vector.
[0071] In a possible implementation, after parameter optimization of multiple trained single-task CatBoost regression models using the parameter optimization algorithm, the method further includes: using the fourth feature vector as a test set to test the parameter-optimized multiple single-task CatBoost regression models; determining whether the coefficient of determination in the test result is less than or equal to zero; if so, adding an attention mechanism to train and optimize the parameters of the multiple single-task CatBoost regression models; if not, the parameter-optimized multiple single-task CatBoost regression models are multiple pre-trained single-task ensemble learning models.
[0072] This embodiment can achieve single-task prediction for each traditional Chinese medicine and output the predicted dosage. The feature engineering and the selected optimal parameters among each single-task model are the best choices for their respective models.
[0073] This embodiment combines a single-task CatBoost regression model optimized by an attention mechanism, which can effectively improve the prediction accuracy of the dosage of a single traditional Chinese medicine.
[0074] In a possible implementation, after obtaining multiple pre-trained single-task ensemble learning models, the method further includes: inputting the fourth feature vector into the multiple pre-trained single-task ensemble learning models and outputting the predicted results of the dosages of each traditional Chinese medicine as the fifth feature vector; performing standardization processing on the fourth feature vector, and performing feature fusion on the standardized fourth feature vector and the fifth feature vector to obtain a sixth feature vector.
[0075] In a specific example, Z-score standardization processing is performed on the fourth feature vector; the Z-score standardized fourth feature vector and the fifth feature vector are subjected to feature fusion to achieve feature enhancement and generate a sixth feature vector.
[0076] In a possible implementation, before inputting the third feature vector into the pre-trained multi-task deep learning model to obtain each traditional Chinese medicine dosage, the method further includes: establishing a multi-task deep learning model, training the multi-task deep learning model using the sixth feature vector as a training set; optimizing the parameters of the trained multi-task deep learning model to obtain the input feature dimension, the number of tasks, the number of neurons in each layer, the dropout rate in each layer, and the learning rate; constructing a hybrid architecture for the multi-task deep learning model with optimized parameters to obtain the pre-trained multi-task deep learning model, where the pre-trained multi-task deep learning model includes a feature cross layer, a compatibility relationship learning layer, a dosage collaboration prediction layer, and a multi-task output layer.
[0077] In a specific example, the prediction result of the single-task CatBoost regression model is used as the fifth feature vector.
[0078] Furthermore, the fourth feature vector after Z-score normalization and the fifth feature vector generated from the prediction result of the CatBoost regression model are fused to generate the sixth feature vector with enhanced features. In this embodiment, a feature enhancement technique is adopted to fuse the standardized clinical features and the dosage prediction results into an enhanced feature set, which can effectively improve the prediction accuracy.
[0079] Furthermore, a multi-task deep learning model is constructed.
[0080] Furthermore, the sixth feature vector is randomly divided into a training set and a test set at a ratio of 8:2.
[0081] Furthermore, the parameters of the multi-task deep learning model are optimized, and the input feature dimension, the number of model tasks, the number of neurons in each layer of the model, the dropout rate in each layer of the model, and the learning rate of the model are configured.
[0082] Furthermore, a hybrid architecture is constructed for the multi-task deep learning model. A port layer is defined, the input features are passed into the model, and the feature cross layer, the compatibility relationship learning layer, the dosage collaboration prediction layer, and the multi-task output layer are set.
[0083] Furthermore, the multi-task deep learning hybrid architecture uses the Dense layer for feature crossing, applies the ReLU activation function, uses Batch Normalization for batch normalization, and uses the Dropout layer to prevent overfitting.
[0084] Furthermore, the multi-task deep learning model uses the fit method to train the model and applies callbacks for early stopping and learning rate adjustment.
[0085] Furthermore, for the multi-task deep learning model, the model is evaluated using the sixth eigenvector test set, and the mean absolute error is used as the evaluation metric for each task.
[0086] In this embodiment, a hybrid architecture model is used to achieve feature crossing, compatibility relationship learning, and multi-task collaborative prediction, improving the rationality of prescriptions and the accuracy of the dosages and combinations of traditional Chinese medicines in the prescriptions.
[0087] Furthermore, the feature crossing layer is the first fully connected layer, including 512 nodes, using ReLU activation, followed by BatchNormalization and Dropout 0.3; the compatibility relationship learning layer is the second fully connected layer, including 256 nodes, using the rectified linear unit ReLU activation, followed by batch normalization Batch Normalization and dropout layer Dropout 0.2; the dosage collaborative prediction layer is the third fully connected layer, including 128 nodes, using the rectified linear unit ReLU activation, followed by batch normalization Batch Normalization and dropout layer Dropout 0.1.
[0088] In this embodiment, a hybrid architecture model is used to achieve feature crossing, compatibility relationship learning, and multi-task collaborative prediction, improving the rationality of prescriptions and accurately determining the dosages and combinations of traditional Chinese medicines in the prescriptions.
[0089] In one possible implementation, the method further includes: training the multi-task deep learning model using a dynamic early stopping mechanism and step learning rate decay.
[0090] In a specific example, a dynamic early stopping mechanism and step learning rate decay are used for model training.
[0091] Furthermore, the dynamic early stopping mechanism terminates if the validation loss has not improved for 10 consecutive training epochs.
[0092] Furthermore, the decay coefficient of the step learning rate decay is 0.5, and the patience value is 5 epochs.
[0093] In one possible implementation, the compatibility relationship learning layer includes a compatibility taboo screening module, which is used to detect the compatibility taboos of traditional Chinese medicines in the prescription.
[0094] In a specific example, the compatibility taboo screening module is used to detect compatibility taboos such as the eighteen incompatible medicaments and the nineteen medicaments with mutual inhibition in the prescription.
[0095] In one possible implementation, the multi-task output layer includes multiple parallel linear units.
[0096] In a specific example, the output layer includes N parallel linear units, where N ≥ 180.
[0097] In this embodiment, by inputting clinical data such as patient symptoms and signs, an intelligent prescription for synergistic optimization of traditional Chinese medicine doses with a three-stage architecture is output. The first stage includes using CatBoost to process categorical features to generate a dose baseline; the second stage includes constructing an enhanced feature space by stacking standardized features and predicted values across dimensions; the third stage includes using a deep multi-task network to synchronously output all traditional Chinese medicine doses, and combining dynamic early stopping and learning rate decay to improve the generalization ability of the model. Therefore, the model architecture combining single-task machine learning and deep learning is an effective algorithm model for generating coronary heart disease prescriptions for high-dimensional and small-sample data sets.
[0098] As Figure 2 shown, a computer system suitable for implementing the computer device provided in this embodiment includes a central processing module (CPU), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage part into a random access memory (RAM). In the RAM, various programs and data required for the operation of the computer system are also stored. The CPU, ROM, and RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.
[0099] The following components are connected to the I / O interface: an input part including a keyboard, a mouse, etc.; an output part including a liquid crystal display (LCD), a speaker, etc.; a storage part including a hard disk, etc.; and a communication part including a network interface card such as a LAN card, a modem, etc. The communication part performs communication processing via a network such as the Internet. A drive is also connected to the I / O interface as needed. A removable medium, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive as needed so that a computer program read from it can be installed into the storage part as needed.
[0100] Specifically, according to this embodiment, the process described in the above flowchart can be implemented as a computer software program. For example, this embodiment includes a computer program product, which includes a computer program tangibly contained on a computer-readable medium, and the above computer program includes program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part and / or installed from a removable medium.
[0101] The flowcharts and schematic diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the system, method, and computer program product of this embodiment. In this regard, each block in the flowchart or schematic diagram may represent a module, a segment of a program, or a part of code, and the above-mentioned module, segment of a program, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the schematic diagram and / or flowchart, as well as the combination of blocks in the schematic and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.
[0102] The method of the present invention can be stored as a program in a medium. Because, on the other hand, the present invention provides a non-volatile computer storage medium, which may be the non-volatile computer storage medium included in the above-mentioned device in the above-mentioned embodiment, or may exist separately and be a non-volatile computer storage medium not assembled into the terminal. The above-mentioned non-volatile computer storage medium stores one or more programs, and when the above-mentioned one or more programs are executed by a device, the device is caused to execute the method for generating a coronary heart disease prescription based on multi-task deep learning disclosed by the present invention.
[0103] It should be noted that in the description of the present invention, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or device including the said element.
[0104] Obviously, the above-mentioned embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation manners here. Any obvious changes or modifications derived from the technical solutions of the present invention still fall within the protection scope of the present invention.
Claims
1. A coronary heart disease prescription generation device for executing a multi-task deep learning algorithm, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor is configured to perform the following steps when loading the computer program: Obtain coronary heart disease medical record data, and process the coronary heart disease medical record data to obtain a first feature vector; Input the first feature vector into multiple pre-trained single-task ensemble learning models, and output each traditional Chinese medicine dosage prediction result as a second feature vector; Perform normalization processing on the first feature vector, and perform feature fusion on the normalized first feature vector and the second feature vector to obtain a third feature vector; Input the third feature vector into a pre-trained multi-task deep learning model to obtain each traditional Chinese medicine dosage, and obtain a coronary heart disease prescription according to each traditional Chinese medicine dosage.
2. The coronary heart disease prescription generation device for executing a multi-task deep learning algorithm according to claim 1, wherein The processing the coronary heart disease medical record data to obtain a first feature vector includes: Perform data extraction on the coronary heart disease medical record data to obtain a data extraction result; Perform data cleaning on the data extraction result to obtain the first feature vector.
3. The coronary heart disease prescription generation device for executing a multi-task deep learning algorithm according to claim 2, wherein Before inputting the first feature vector into multiple pre-trained single-task ensemble learning models, the method further includes: Obtain historical coronary heart disease medical record data; Perform data extraction and data cleaning on the historical coronary heart disease medical record data to obtain a fourth feature vector, where the fourth feature vector includes symptoms, signs, auxiliary examination results, current medications, and traditional Chinese medicines in the prescription; Use the traditional Chinese medicines and their dosages in the prescription as dependent variables to establish multiple single-task class boosting regression models; Use the fourth feature vector as a training set to train the multiple single-task class boosting regression models; Use a parameter optimization algorithm to optimize the parameters of the trained multiple single-task class boosting regression models to obtain multiple pre-trained single-task ensemble learning models.
4. The coronary heart disease prescription generation device for executing a multi-task deep learning algorithm according to claim 3, wherein After using the parameter optimization algorithm to optimize the parameters of the trained multiple single-task class boosting regression models, the method further includes: Use the fourth feature vector as a test set to test the multiple single-task class boosting regression models with optimized parameters; Determine whether the coefficient of determination in the test result is less than or equal to zero; If so, add an attention mechanism to train and optimize the parameters of the multiple single-task class boosting regression models.
5. The coronary heart disease prescription generation device for executing a multi-task deep learning algorithm according to claim 4, wherein The determining whether the coefficient of determination in the test result is less than or equal to zero further includes: If not, the multiple single-task class boosting regression models with optimized parameters are multiple pre-trained single-task ensemble learning models.
6. The coronary heart disease prescription generation device for executing a multi-task deep learning algorithm according to claim 5, wherein After obtaining the multiple pre-trained single-task ensemble learning models, the method further includes: Input the fourth feature vector into multiple pre-trained single-task ensemble learning models, and output the predicted results of each traditional Chinese medicine dosage as the fifth feature vector; Perform normalization processing on the fourth feature vector, and perform feature fusion on the normalized fourth feature vector and the fifth feature vector to obtain a sixth feature vector.
7. An equipment for generating coronary heart disease prescriptions by executing a multi-task deep learning algorithm according to claim 6, characterized in that, Before inputting the third feature vector into the pre-trained multi-task deep learning model to obtain each traditional Chinese medicine dosage, the method further includes: Establish a multi-task deep learning model, and use the sixth feature vector as a training set to train the multi-task deep learning model; Optimize the parameters of the trained multi-task deep learning model to obtain the input feature dimension, the number of tasks, the number of neurons in each layer, the dropout rate in each layer, and the learning rate; Construct a hybrid architecture for the multi-task deep learning model with optimized parameters to obtain a pre-trained multi-task deep learning model. The pre-trained multi-task deep learning model includes a feature crossing layer, a compatibility relationship learning layer, a dosage collaborative prediction layer, and a multi-task output layer.
8. The coronary heart disease prescription generation device for executing a multi-task deep learning algorithm according to claim 7, wherein, The method further includes: Train the multi-task deep learning model using a dynamic early stopping mechanism and step learning rate decay.
9. An equipment for generating coronary heart disease prescriptions by executing a multi-task deep learning algorithm according to claim 8, characterized in that, The compatibility relationship learning layer includes a compatibility taboo screening module, and the compatibility taboo screening module is used to detect the compatibility taboos of traditional Chinese medicines in the prescription.
10. An equipment for generating coronary heart disease prescriptions by executing a multi-task deep learning algorithm according to claim 9, characterized in that, The multi-task output layer includes multiple parallel linear units.