Clozapine dosage recommendation device, model training method, equipment, medium and product
Through the deep learning-based clozapine dose recommendation device, personalized dose recommendation is performed using multi-dimensional information and DANETs models, which solves the problem of poor clozapine dose recommendation in the prior art, and achieves more efficient and safe therapeutic effects.
Patent Information
- Application Number
- CN202411901887.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-23
- Publication Date
- 2025-05-13
AI Technical Summary
The prior art is difficult to effectively recommend clozapine doses, and cannot adapt to individualized characteristics, resulting in poor treatment effects and high side effects.
The clozapine dose recommendation device based on deep learning is used to obtain multi-dimensional information through the data reading device, and the recommended model trained by the preset DANETs model is analyzed to output personalized clozapine dose recommendation.
Personalized clozapine dose recommendations have been achieved, which improves the treatment effect, reduces the side effects of drugs, reduces the workload of doctors and pharmacists, and improves the satisfaction and compliance of patients with treatment plans.
Smart Images
Figure CN119993369A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to a clozapine dosage recommendation device, a model training method, equipment, a medium and a product. Background Art
[0002] Clozapine has unique advantages in improving severe schizophrenia, but its toxicity risk is very high. Due to the narrow therapeutic window of clozapine and the large individual differences, therapeutic drug monitoring (TDM) is required to keep the blood drug concentration within the therapeutic window. Therefore, different users have different needs for clozapine.
[0003] In the related technology, drug dosage recommendations are usually made through data statistical analysis, but this method often cannot adapt to individual characteristics. Therefore, how to make clozapine dosage recommendations more effectively has become an urgent problem to be solved in the industry. Summary of the invention
[0004] The present invention provides a clozapine dosage recommendation device, a model training method, equipment, a medium and a product, which are used to solve the problem of how to more effectively recommend clozapine dosage in the prior art, which has become an urgent problem to be solved in the industry.
[0005] The present invention provides a clozapine dosage recommendation device based on deep learning, comprising: a data reading device and an analysis device; The data reading device is used to obtain the multi-dimensional information of the target object and transmit the multi-dimensional information to the analysis device; wherein the multi-dimensional information of the object includes at least one of the following: basic information of the object, treatment dimension information, diagnosis dimension information, adverse reaction dimension information and clozapine blood concentration information; The analysis device is loaded with a recommendation model for clozapine dosage recommendation, and after receiving the multi-dimensional information of the object, the analysis device outputs the clozapine dosage recommendation information corresponding to the target object; The recommendation model is obtained by training a preset DANETs model based on multi-dimensional sample information of subjects carrying clozapine dosage labels.
[0006] According to a clozapine dosage recommendation device based on deep learning provided by the present invention, the device further includes: a pre-training device; the pre-training device is specifically used for: Acquiring multi-dimensional sample information of the screened object, wherein the multi-dimensional sample information of the object carries a clozapine dosage label; For any of the multi-dimensional sample information of the object carrying the clozapine dosage label, recursively perform feature abstraction on the multi-dimensional sample information of the object on the main path of any basic block of the preset DANETs model to obtain the main path output feature; wherein the main path includes multiple ABSTLAYs; Extracting shortcut path input features directly from the multi-dimensional sample information of the object through the shortcut path of the basic block; Fusing the main path output features and the shortcut path input features to obtain a fused feature vector; Input the fused feature vector into the next basic block until the current basic block is the network end of the preset DANETs model, and output the fused feature vector as the model of the output layer in the DANETs model; Calculating the loss value based on the clozapine dosage label corresponding to the multi-dimensional sample information of the object output by the model, and optimizing the preset DANETs model according to the loss value; When the preset training conditions are met, the model training is stopped, a recommended model is obtained, and the recommended model is transmitted to the analysis device.
[0007] According to a clozapine dosage recommendation device based on deep learning provided by the present invention, the pre-training device is also used for: Selecting a target feature subset from the multi-dimensional sample information of the object using a learnable sparse mask; The target feature subset is converted into a higher-level target feature representation through a fully connected layer and an attention mechanism; By summing up each element, each of the target feature representations is fused into the main path output feature.
[0008] According to a clozapine dosage recommendation device based on deep learning provided by the present invention, the device also includes: a data sample library construction device; the data sample library construction device is specifically used to: Obtain multi-dimensional sample information of multiple original objects; After performing missing value screening and statistical method preliminary screening on the original object multi-dimensional sample information, the screened original object multi-dimensional sample information is obtained; Performing a secondary screening on the screened original object multi-dimensional sample information according to a correlation analysis method to obtain the secondary screened original object multi-dimensional sample information; Based on feature engineering analysis, a variable final screening is performed on the multi-dimensional sample information of the original object after the secondary screening to obtain the multi-dimensional sample information of the object after the final screening; A data sample library is constructed based on the multi-dimensional sample information of each of the objects after final screening.
[0009] The present invention also provides a recommendation model training method, comprising: Acquire multi-dimensional sample information of a plurality of objects after screening, each of the multi-dimensional sample information of the object carries a corresponding clozapine dosage label; For any of the multi-dimensional sample information of the object carrying the clozapine dosage label, on the main path of any basic block of the preset DANETs model, recursively perform feature abstraction on the multi-dimensional sample information of the object to obtain the main path output feature; wherein the main path includes multiple ABSTLAYs; Extracting shortcut path input features directly from the multi-dimensional sample information of the object through the shortcut path of the basic block; Fusing the main path output features and the shortcut path input features to obtain a fused feature vector; Input the fused feature vector into the next basic block until the current basic block is the network end of the preset DANETs model, and output the fused feature vector as the model of the output layer in the DANETs model; Calculating the loss value based on the clozapine dosage label corresponding to the multi-dimensional sample information of the object output by the model, and optimizing the preset DANETs model according to the loss value; When the preset training conditions are met, stop model training and obtain the recommended model; The recommendation model is used to output clozapine dosage recommendation information corresponding to the target object based on the input multi-dimensional information of the target object.
[0010] According to a recommendation model training method provided by the present invention, the method for obtaining the main path output feature specifically includes: Selecting a target feature subset from the multi-dimensional sample information of the object using a learnable sparse mask; The target feature subset is converted into a higher-level target feature representation through a fully connected layer and an attention mechanism; By summing up each element, each of the target feature representations is fused into the main path output feature.
[0011] According to a recommendation model training method provided by the present invention, the step of obtaining multi-dimensional sample information of multiple objects after screening includes: Obtain multi-dimensional sample information of multiple original objects; After performing missing value screening and statistical method preliminary screening on the original object multi-dimensional sample information, the screened original object multi-dimensional sample information is obtained; Performing a secondary screening on the screened original object multi-dimensional sample information according to a correlation analysis method to obtain the secondary screened original object multi-dimensional sample information; Based on feature engineering analysis, the multi-dimensional sample information of the original object after the secondary screening is subjected to final variable screening to obtain the multi-dimensional sample information of the object after final screening.
[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the recommendation model training method described above is implemented.
[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the recommendation model training methods described above.
[0014] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the recommendation model training method described in any one of the above is implemented.
[0015] The clozapine dosage recommendation device, model training method, equipment, medium and product provided by the present invention, the analysis device is the core of the device, and it includes a deep learning model, namely a recommendation model. This model is responsible for processing the multi-dimensional information transmitted from the data reading device, and outputs clozapine dosage recommendations for specific patients based on this information. By analyzing the multi-dimensional information of the patient, the device can provide customized clozapine dosage recommendations for each patient, improve the treatment effect and reduce side effects. Automated data processing and dosage recommendations reduce the workload of doctors and pharmacists, allowing them to focus more on other clinical tasks. Individualized dosage recommendations help improve patient satisfaction and compliance with the treatment plan, thereby improving the overall treatment effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0017] Figure 1 It is a structural schematic diagram of a clozapine dosage recommendation device based on deep learning provided by the present invention; Figure 2 The recommendation model training method provided by the present invention; Figure 3 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION
[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0019] Figure 1 is a schematic diagram of the structure of a clozapine dosage recommendation device based on deep learning provided by the present invention, such as Figure 1 As shown, it includes: a data reading device 11 and an analysis device 12; The data reading device 11 is used to obtain the multi-dimensional information of the target object and transmit the multi-dimensional information to the analysis device; wherein the multi-dimensional information of the object includes at least one of the following: basic information of the object, treatment dimension information, diagnosis dimension information, adverse reaction dimension information and clozapine blood concentration information; The analysis device 12 is loaded with a recommendation model for clozapine dosage recommendation. After receiving the multi-dimensional information of the object, the analysis device outputs the clozapine dosage recommendation information corresponding to the target object. The recommendation model is obtained by training a preset DANETs model based on multi-dimensional sample information of subjects carrying clozapine dosage labels.
[0020] In the present invention, the data reading device is responsible for collecting multi-dimensional information of the target object, which may include basic information of the patient (such as age, gender, weight, etc.), treatment dimension information (such as past medication history, current treatment plan, etc.), diagnosis dimension information (such as disease diagnosis results, severity of the disease, etc.), adverse reaction dimension information (such as the patient's adverse reaction record to the drug) and clozapine blood concentration information (such as the concentration level of clozapine in the blood). This information is crucial for precision medicine and individualized treatment.
[0021] In the present invention, the analysis device includes a deep learning model, namely, a recommendation model. This model is responsible for processing the multi-dimensional information transmitted from the data reading device, and outputting a clozapine dosage recommendation for a specific patient based on this information. The construction and training of the recommendation model relies on a large amount of sample information, which includes clozapine dosage labels, i.e., known appropriate dosage data, which is used to train the model to learn how to predict the appropriate clozapine dosage based on the multi-dimensional information of the patient.
[0022] In the present invention, the recommendation model is trained based on the DANETs (Deep Abstract Networks) architecture. DANETs is a deep learning model that is particularly suitable for processing tabular data and image data, and can adaptively integrate local features and global dependencies through an attention mechanism. In this device, the recommendation model takes advantage of the DANETs architecture and learns how to predict the most appropriate clozapine dose based on the patient's multi-dimensional information, thereby achieving individualized treatment.
[0023] In the present invention, the analysis device is the core of the device, which contains a deep learning model, namely the recommendation model. This model is responsible for processing the multi-dimensional information transmitted from the data reading device, and outputs clozapine dosage recommendations for specific patients based on this information. By analyzing the multi-dimensional information of the patient, the device can provide customized clozapine dosage recommendations for each patient, improve the treatment effect and reduce side effects. Automated data processing and dosage recommendations reduce the workload of doctors and pharmacists, allowing them to focus more on other clinical tasks. Individualized dosage recommendations help improve patient satisfaction and compliance with the treatment plan, thereby improving the overall treatment effect.
[0024] Optionally, the apparatus further comprises: a pre-training device; the pre-training device is specifically used for: Acquiring multi-dimensional sample information of the screened object, wherein the multi-dimensional sample information of the object carries a clozapine dosage label; For any of the multi-dimensional sample information of the object carrying the clozapine dosage label, recursively perform feature abstraction on the multi-dimensional sample information of the object on the main path of any basic block of the preset DANETs model to obtain the main path output feature; wherein the main path includes multiple ABSTLAYs; Extracting shortcut path input features directly from the multi-dimensional sample information of the object through the shortcut path of the basic block; Fusing the main path output features and the shortcut path input features to obtain a fused feature vector; Input the fused feature vector into the next basic block until the current basic block is the network end of the preset DANETs model, and output the fused feature vector as the model of the output layer in the DANETs model; Calculating the loss value based on the clozapine dosage label corresponding to the multi-dimensional sample information of the object output by the model, and optimizing the preset DANETs model according to the loss value; When the preset training conditions are met, the model training is stopped, a recommended model is obtained, and the recommended model is transmitted to the analysis device.
[0025] In the present invention, the pre-training device first obtains multi-dimensional sample information of the screened objects, and the sample information carries clozapine dosage labels to provide the necessary data basis for model training.
[0026] In the present invention, the basic blocks in the preset DANETs model, especially the multiple ABSTLAYs on the main path, can be used to perform recursive feature abstraction on the multi-dimensional sample information of objects carrying clozapine dosage labels to extract deep feature representations.
[0027] The shortcut path of the basic block directly extracts the shortcut path input features from the multi-dimensional sample information of the object, retains the original information, and increases the model's ability to understand and express the data.
[0028] The features output by the main path and the features input by the shortcut path are fused to form a fused feature vector. This step enhances the model's ability to integrate information at different levels.
[0029] The fused feature vector is input to the next basic block until the network end of the preset DANETs model is reached to ensure the effective transmission and deep processing of features in the network. Based on the clozapine dosage label corresponding to the model output and sample information, the loss value is calculated, and the preset DANETs model is optimized according to the loss value to improve the prediction accuracy of the model.
[0030] When the preset training conditions are met, the model training is stopped, the recommended model is obtained, and the recommended model is transmitted to the analysis device for actual clozapine dosage recommendation.
[0031] In the present invention, through recursive feature abstraction and feature fusion, the pre-training device can extract deeper and richer feature representations, thereby improving the accuracy of clozapine dosage recommendations. The original information retained by the shortcut path and the abstract features extracted by the main path are used to enhance the generalization ability of the model for new samples. By calculating the loss value and optimizing the model according to the loss value, the pre-training device can more effectively adjust the model parameters, accelerate the convergence speed, and improve the training efficiency. The recommendation model obtained by training can provide personalized clozapine dosage recommendations based on the patient's specific multi-dimensional information, thereby realizing precision medicine.
[0032] Optionally, the pre-training device is also used for: Selecting a target feature subset from the multi-dimensional sample information of the object using a learnable sparse mask; The target feature subset is converted into a higher-level target feature representation through a fully connected layer and an attention mechanism; By summing up each element, each of the target feature representations is fused into the main path output feature.
[0033] In the present invention, the pre-training device uses a learnable sparse mask to select a target feature subset from the multi-dimensional sample information of the object. This step is achieved through ABSTLAY, where the sparse mask (defined by the weight vector Wmask) is responsible for selecting the features most relevant to the clozapine dose from a large number of input features.
[0034] The selected target feature subset is converted into a higher-level target feature representation through a fully connected layer and an attention mechanism. This process uses the attention mechanism to strengthen the model's focus on key features and improve the quality of feature representation. Each target feature representation is fused into the main path output feature by element-by-element summation. This fusion operation integrates information from different feature groups and enhances the model's ability to comprehensively understand the data.
[0035] In the present invention, through precise feature selection and high-level feature representation, the pre-training device can improve the accuracy of clozapine dosage recommendation, thereby optimizing the treatment plan; the main path output features that integrate multi-dimensional information enable the model to have better generalization ability and can adapt to the characteristics and needs of different patients.
[0036] Optionally, the apparatus further comprises: a data sample library construction device; the data sample library construction device is specifically used to: Obtain multi-dimensional sample information of multiple original objects; After performing missing value screening and statistical method preliminary screening on the original object multi-dimensional sample information, the screened original object multi-dimensional sample information is obtained; Performing a secondary screening on the screened original object multi-dimensional sample information according to a correlation analysis method to obtain the secondary screened original object multi-dimensional sample information; Based on feature engineering analysis, a variable final screening is performed on the multi-dimensional sample information of the original object after the secondary screening to obtain the multi-dimensional sample information of the object after the final screening; A data sample library is constructed based on the multi-dimensional sample information of each of the objects after final screening.
[0037] In the present invention, the data sample library construction device first obtains multi-dimensional sample information of multiple original objects, which may include basic information of patients, treatment history, diagnosis results, adverse reaction records, and clozapine blood concentration, etc.
[0038] The original sample information was screened for missing values, and variables with missing rates greater than the preset threshold (e.g., 50%) were removed to ensure the integrity and reliability of the data. Statistical methods were used to preliminarily screen the original sample information, including the Wilcoxon rank sum test, T test, and variance test for continuous independent variables, and the chi-square test and Fish test for categorical variables to determine the statistical significance between the variable and the target variable (clozapine dose).
[0039] The original sample information is screened again according to the correlation analysis method, and the Pearson correlation coefficient or other similarity measurement methods between the feature variables and the target variables are calculated to further streamline the variables and retain the features that are highly correlated with the target variables.
[0040] Based on feature engineering analysis, the multi-dimensional sample information of the objects after secondary screening is subjected to final variable screening, which may include using the GBDT algorithm to calculate the importance of the variables and select the variables with the highest importance to construct the final data sample library.
[0041] Based on the multi-dimensional sample information of the objects after final screening, a data sample library is constructed to provide high-quality training data for model training.
[0042] More specifically, feature engineering analysis can refer to the importance feature screening algorithm based on GBDT, which can specifically refer to: a. Taking "clozapine dosage" as the target variable and the variables obtained from the initial screening as covariates, a prediction model was constructed based on the XGBoost algorithm.
[0043] b. Perform 80% cross validation on the model and adjust parameters (parameters include loss function selection, iterations, learning_rate, l2_leaf_reg, depth, etc.) to optimize the model's evaluation indicator F1-Score.
[0044] c. Calculate the importance score of each variable and sort them in descending order. The higher the importance score, the greater the impact of the variable on the model.
[0045] d. Select the top 10 variables in terms of importance.
[0046] In the present invention, the data quality in the data sample library is ensured through missing value screening and statistical method preliminary screening, providing a reliable data basis for model training. The correlation analysis method and feature engineering final screening help to screen out the features most relevant to the target variable, enhancing the generalization ability and prediction accuracy of the model.
[0047] Figure 2 The recommendation model training method provided by the present invention is as follows: Figure 2 As shown, it includes: step 210, obtaining a plurality of object multi-dimensional sample information after screening, each of the object multi-dimensional sample information carries a corresponding clozapine dosage label; Step 220, for any of the multi-dimensional sample information of the object carrying the clozapine dosage label, recursively perform feature abstraction on the multi-dimensional sample information of the object on the main path of any basic block of the preset DANETs model to obtain the main path output feature; wherein the main path includes multiple ABSTLAYs; In the present invention, multi-dimensional sample information of an object carrying a clozapine dosage label is received, and the information may include the patient's basic information, treatment history, diagnosis results, drug reactions, etc.
[0048] In the DANETs model, the main path consists of multiple ABSTLAYs (Abstract Layers), each of which is responsible for abstracting and transforming the input features.
[0049] The multi-dimensional sample information first enters the first ABSTLAY of the basic block for feature selection and abstraction. ABSTLAY selects the feature subset most relevant to clozapine dosage through the learned sparse mask and converts it into a higher-level feature representation.
[0050] These high-level feature representations are then passed to the next ABSTLAY, where feature abstraction is performed again, and this process is recursively performed within the basic block. The output feature representations of each ABSTLAY are summed element by element and fused into a main path output feature vector. This fused feature vector combines the feature information extracted by multiple ABSTLAYs, providing a comprehensive feature representation for subsequent model layers.
[0051] After all ABSTLAY processing and fusion within the basic block, the final main path output feature vector contains rich, multi-level feature information that can better represent the original data and be used for clozapine dosage prediction.
[0052] Step 230, extracting shortcut path input features directly from the multi-dimensional sample information of the object through the shortcut path of the basic block; In the present invention, a shortcut path is a design in a deep learning network for extracting features directly from the input layer or early layers, bypassing some network layers, to preserve the original information and prevent the gradient vanishing problem.
[0053] The shortcut path of the basic block directly extracts features from the multi-dimensional sample information of the object. These features are not subjected to multiple abstractions and transformations on the main path, thus retaining more original data features.
[0054] The features extracted by the shortcut path are called shortcut path input features. These features are combined with the main path output features to increase the robustness of the model and provide a more comprehensive data representation. The shortcut path input features are usually fused with the main path output features at the end of the basic block or in subsequent layers of the network to form a more complete feature representation.
[0055] Through the shortcut path, the model can directly access the original features in the input data, which helps prevent information loss in deep networks, especially when dealing with high-dimensional and complex medical data. The shortcut path provides a mechanism that enables the model to simultaneously learn multiple levels of representation from raw data to abstract features, enhancing the model's ability to express data.
[0056] Step 240, fusing the main path output features and the shortcut path input features to obtain a fused feature vector; In the present invention, the purpose of feature fusion is to integrate feature information from different paths to obtain a more comprehensive and rich feature representation, which helps to improve the model's understanding and prediction capabilities of the data. The main path output features are deep feature representations obtained through recursive abstraction of multiple ABSTLAY layers, which contain key information and complex patterns in the data. The shortcut path input features are directly extracted from the original multi-dimensional sample information, retaining the original features and details of the data, which are useful for capturing direct and intuitive relationships in the data.
[0057] The main path output features and the shortcut path input features are fused, usually by element-by-element addition, but may also include other types of fusion strategies, such as weighted summation, splicing, etc.
[0058] The fused feature vector contains the direct features of the original data and the abstracted deep features, which enables the model to utilize both the intuitive information and abstract representation of the data.
[0059] Step 250, inputting the fused feature vector into the next basic block until the current basic block is the network end of the preset DANETs model, and outputting the fused feature vector as the model of the output layer in the DANETs model; In the present invention, the fused feature vector is input to the next basic block in the DANETs model. In each basic block, the fused feature vector will go through the process of feature abstraction and fusion again to further extract and integrate feature information.
[0060] This process is recursively performed in each basic block of the DANETs model, and each basic block further processes and abstracts the input fused feature vector, continuously deepening the level and richness of the features. Until the current basic block is the network end of the preset DANETs model, it means that the fused feature vector has been processed by all basic blocks in the model.
[0061] At the end of the network, the final fused feature vector is used as the input to the output layer in the DANETs model, thereby generating the final output of the model, which is the recommendation for clozapine dosage.
[0062] Step 260, calculating the loss value based on the clozapine dosage label corresponding to the multi-dimensional sample information of the object output by the model, and optimizing the preset DANETs model according to the loss value; In the present invention, in order to calculate the difference between the model output and the actual label, a suitable loss function needs to be selected. For regression problems, commonly used loss functions include mean square error (MSE) or mean absolute error (MAE). For classification problems, cross entropy loss may be used.
[0063] The loss value is a quantitative measure of the difference between the model output and the actual label. Through the loss function, the loss value of each sample can be calculated, and then the average loss value of the entire training batch is calculated. The calculation of the loss value is the starting point of the backpropagation algorithm. Through backpropagation, the gradient of the loss value is passed back to each parameter of the model. This process involves the chain rule to calculate the contribution of each parameter to the final loss.
[0064] Based on the calculated gradient, the model parameters are updated using an optimization algorithm (such as SGD, Adam, etc.). The purpose of this step is to adjust the model parameters to reduce the loss value, thereby improving the prediction accuracy of the model.
[0065] Step 270, when the preset training conditions are met, stop model training and obtain a recommended model; The recommendation model is used to output clozapine dosage recommendation information corresponding to the target object based on the input multi-dimensional information of the target object.
[0066] In the present invention, some termination conditions are set before the model training begins. These conditions may include: reaching a predetermined training cycle (epochs); the training loss value drops below a lower threshold, indicating that the model has fully learned the patterns in the data; the loss value on the validation set has not improved significantly over multiple consecutive cycles, which may indicate that the model has begun to overfit; reaching a predetermined time or resource consumption limit.
[0067] Once any of the above preset training conditions are met, the training process will be terminated to prevent overfitting or unnecessary waste of computing resources. After the training is terminated, the current state of the model will be saved as the recommendation model. This model has learned how to predict the clozapine dose based on the input features through the training data.
[0068] After obtaining the recommended model, a final evaluation is usually performed on an independent test set to verify the performance and generalization ability of the model.
[0069] In the present invention, the recommendation model can provide accurate clozapine dosage recommendations based on the patient's specific multi-dimensional information. Accurate dosage recommendations help improve treatment effects, reduce drug side effects, and improve the patient's quality of life.
[0070] Optionally, other model training processes may refer to the above embodiments and will not be described in detail here.
[0071] In an optional embodiment, the specific modeling method is as follows: (1) Feature Abstraction Layer The first step of ABSTLAY (Abstract Layer) involves selecting a most representative subset from the input features through a learnable sparse mask. This process is achieved by applying a weight vector Wmask to the input feature vector and combining it with Entmax sparse mapping to achieve efficient feature selection. Given a vector after feature selection , sent to a fully connected layer, combined with the attention mechanism, and extracted high-level features from specific features through abstract function learning. In order to reduce the complexity of ABSTLAY in the reasoning process, feature selection and feature abstraction are combined together. Feature abstraction function It consists of two linear transformations and an activation function. The abstract function is defined as: Where N is the number of feature sampling functions in ABSTLAY, and is the weight matrix of the kth feature abstraction function, and is the bias vector.
[0072] (2) Output fusion operation ABSTLAY can process multiple feature groups in parallel, and the output fusion operation adds the outputs of all feature abstraction functions by element-by-element summation to obtain the final output feature vector.
[0073] (3) Basic blocks Basic blocks are mainly constructed with ABSTLAYs, which can obtain information from the original features by introducing new shortcut paths and combining them with the output of ABSTLAY to increase the diversity at different levels. Defined as: in, is a shortcut path, including an ABSTLAY and Dropout layer, x is the original feature, There are multiple ABSTLAYs on the main path, whose input has the features generated by the previous basic block.
[0074] (4) Network architecture DANETs are built by stacking a series of carefully designed basic blocks, each of which has the ability to recursively abstract features. After learning sparse masks, feature interactions and transformations, residual connections, and normalization and activation functions, the last part adds a three-layer MLP with ReLU activation function to transform the output using the Softmax function. , the formula is as follows: Where C is the number of categories and z is the linear combination result of the three-layer MLP.
[0075] Figure 3 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 3 As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330 and a communication bus 340, wherein the processor 310, the communication interface 320 and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logic instructions in the memory 330 to execute the recommendation model training method, which includes: obtaining multi-dimensional sample information of multiple objects after screening, each of the multi-dimensional sample information of the object carries a corresponding clozapine dosage label; For any of the multi-dimensional sample information of the object carrying the clozapine dosage label, on the main path of any basic block of the preset DANETs model, recursively perform feature abstraction on the multi-dimensional sample information of the object to obtain the main path output feature; wherein the main path includes multiple ABSTLAYs; Extracting shortcut path input features directly from the multi-dimensional sample information of the object through the shortcut path of the basic block; Fusing the main path output features and the shortcut path input features to obtain a fused feature vector; Input the fused feature vector into the next basic block until the current basic block is the network end of the preset DANETs model, and output the fused feature vector as the model of the output layer in the DANETs model; Calculating the loss value based on the clozapine dosage label corresponding to the multi-dimensional sample information of the object output by the model, and optimizing the preset DANETs model according to the loss value; When the preset training conditions are met, stop model training and obtain the recommended model; The recommendation model is used to output clozapine dosage recommendation information corresponding to the target object based on the input multi-dimensional information of the target object.
[0076] In addition, the logic instructions in the above-mentioned memory 330 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.
[0077] On the other hand, the present invention further provides a computer program product, the computer program product includes a computer program, the computer program can be stored on a non-transitory computer-readable storage medium, when the computer program is executed by a processor, the computer can execute the recommendation model training method provided by the above methods, the method comprising: obtaining multi-dimensional sample information of a plurality of objects after screening, each of the multi-dimensional sample information of the object carries a corresponding clozapine dosage label; For any of the multi-dimensional sample information of the object carrying the clozapine dosage label, on the main path of any basic block of the preset DANETs model, recursively perform feature abstraction on the multi-dimensional sample information of the object to obtain the main path output feature; wherein the main path includes multiple ABSTLAYs; Extracting shortcut path input features directly from the multi-dimensional sample information of the object through the shortcut path of the basic block; Fusing the main path output features and the shortcut path input features to obtain a fused feature vector; Input the fused feature vector into the next basic block until the current basic block is the network end of the preset DANETs model, and output the fused feature vector as the model of the output layer in the DANETs model; Calculating the loss value based on the clozapine dosage label corresponding to the multi-dimensional sample information of the object output by the model, and optimizing the preset DANETs model according to the loss value; When the preset training conditions are met, stop model training and obtain the recommended model; The recommendation model is used to output clozapine dosage recommendation information corresponding to the target object based on the input multi-dimensional information of the target object.
[0078] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the recommendation model training method provided by the above methods, the method comprising: obtaining multi-dimensional sample information of a plurality of objects after screening, each of the multi-dimensional sample information of the object carrying a corresponding clozapine dosage label; For any of the multi-dimensional sample information of the object carrying the clozapine dosage label, on the main path of any basic block of the preset DANETs model, recursively perform feature abstraction on the multi-dimensional sample information of the object to obtain the main path output feature; wherein the main path includes multiple ABSTLAYs; Extracting shortcut path input features directly from the multi-dimensional sample information of the object through the shortcut path of the basic block; Fusing the main path output features and the shortcut path input features to obtain a fused feature vector; Input the fused feature vector into the next basic block until the current basic block is the network end of the preset DANETs model, and output the fused feature vector as the model of the output layer in the DANETs model; Calculating the loss value based on the clozapine dosage label corresponding to the multi-dimensional sample information of the object output by the model, and optimizing the preset DANETs model according to the loss value; When the preset training conditions are met, stop model training and obtain the recommended model; The recommendation model is used to output clozapine dosage recommendation information corresponding to the target object based on the input multi-dimensional information of the target object.
[0079] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.
[0080] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A clozapine dosage recommendation device based on deep learning, characterized in that: include: Data reading equipment and analysis equipment; The data reading device is used to obtain the multi-dimensional information of the target object and transmit the multi-dimensional information to the analysis device; wherein the multi-dimensional information of the object includes at least one of the following: basic information of the object, treatment dimension information, diagnosis dimension information, adverse reaction dimension information and clozapine blood concentration information; The analysis device is loaded with a recommendation model for clozapine dosage recommendation, and after receiving the multi-dimensional information of the object, the analysis device outputs the clozapine dosage recommendation information corresponding to the target object; The recommendation model is obtained by training a preset DANETs model based on multi-dimensional sample information of subjects carrying clozapine dosage labels.
2. The clozapine dosage recommendation device based on deep learning according to claim 1, characterized in that: The apparatus further comprises: a pre-training device; the pre-training device is specifically used for: Acquiring multi-dimensional sample information of the screened object, wherein the multi-dimensional sample information of the object carries a clozapine dosage label; For any of the multi-dimensional sample information of the object carrying the clozapine dosage label, recursively perform feature abstraction on the multi-dimensional sample information of the object on the main path of any basic block of the preset DANETs model to obtain the main path output feature; wherein the main path includes multiple ABSTLAYs; Extracting shortcut path input features directly from the multi-dimensional sample information of the object through the shortcut path of the basic block; Fusing the main path output features and the shortcut path input features to obtain a fused feature vector; Input the fused feature vector into the next basic block until the current basic block is the network end of the preset DANETs model, and output the fused feature vector as the model of the output layer in the DANETs model; Calculating the loss value based on the clozapine dosage label corresponding to the multi-dimensional sample information of the object output by the model, and optimizing the preset DANETs model according to the loss value; When the preset training conditions are met, the model training is stopped, a recommended model is obtained, and the recommended model is transmitted to the analysis device.
3. The clozapine dosage recommendation device based on deep learning according to claim 2, characterized in that: The pre-training device is also used for: Selecting a target feature subset from the multi-dimensional sample information of the object using a learnable sparse mask; The target feature subset is converted into a higher-level target feature representation through a fully connected layer and an attention mechanism; By summing up each element, each of the target feature representations is fused into the main path output feature.
4. The clozapine dosage recommendation device based on deep learning according to claim 1, characterized in that: The apparatus further includes: a data sample library construction device; the data sample library construction device is specifically used to: Obtain multi-dimensional sample information of multiple original objects; After performing missing value screening and statistical method preliminary screening on the original object multi-dimensional sample information, the screened original object multi-dimensional sample information is obtained; Performing a secondary screening on the screened original object multi-dimensional sample information according to a correlation analysis method to obtain the secondary screened original object multi-dimensional sample information; Based on feature engineering analysis, a variable final screening is performed on the multi-dimensional sample information of the original object after the secondary screening to obtain the multi-dimensional sample information of the object after the final screening; A data sample library is constructed based on the multi-dimensional sample information of each of the objects after final screening.
5. A recommendation model training method, characterized in that: include: Acquire multi-dimensional sample information of a plurality of objects after screening, each of the multi-dimensional sample information of the object carries a corresponding clozapine dosage label; For any of the multi-dimensional sample information of the object carrying the clozapine dosage label, on the main path of any basic block of the preset DANETs model, recursively perform feature abstraction on the multi-dimensional sample information of the object to obtain the main path output feature; wherein the main path includes multiple ABSTLAYs; Extracting shortcut path input features directly from the multi-dimensional sample information of the object through the shortcut path of the basic block; Fusing the main path output features and the shortcut path input features to obtain a fused feature vector; Input the fused feature vector into the next basic block until the current basic block is the network end of the preset DANETs model, and output the fused feature vector as the model of the output layer in the DANETs model; Calculating the loss value based on the clozapine dosage label corresponding to the multi-dimensional sample information of the object output by the model, and optimizing the preset DANETs model according to the loss value; When the preset training conditions are met, stop model training and obtain the recommended model; The recommendation model is used to output clozapine dosage recommendation information corresponding to the target object based on the input multi-dimensional information of the target object.
6. The recommendation model training method according to claim 5, characterized in that: The method for obtaining the main path output feature specifically includes: Selecting a target feature subset from the multi-dimensional sample information of the object using a learnable sparse mask; The target feature subset is converted into a higher-level target feature representation through a fully connected layer and an attention mechanism; By summing up each element, each of the target feature representations is fused into the main path output feature.
7. The recommendation model training method according to claim 5, characterized in that: The obtaining of multi-dimensional sample information of the plurality of objects after screening includes: Obtain multi-dimensional sample information of multiple original objects; After performing missing value screening and statistical method preliminary screening on the original object multi-dimensional sample information, the screened original object multi-dimensional sample information is obtained; Performing a secondary screening on the screened original object multi-dimensional sample information according to a correlation analysis method to obtain the secondary screened original object multi-dimensional sample information; Based on feature engineering analysis, the multi-dimensional sample information of the original object after the secondary screening is subjected to final variable screening to obtain the multi-dimensional sample information of the object after final screening.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, it implements the recommendation model training method as described in any one of claims 5 to 7.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the recommendation model training method as described in any one of claims 5 to 7.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the recommendation model training method as described in any one of claims 5 to 7.