Dose optimization method and device of paliperidone, electronic equipment and storage medium

Abstract

The invention provides a paliperidone dosage optimization method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence, and the method comprises the steps: obtaining the state data of a patient, the state data comprising the plasma concentration of paliperidone, the heart rate variability (HRV) time domain index and the smoking amount; preprocessing the state data to obtain preprocessed state data, and performing feature extraction on the preprocessed state data to obtain a feature vector of the blood concentration, a feature vector of the HRV time domain index and a feature vector of the smoking amount; and inputting the feature vector of the blood concentration, the feature vector of the HRV time domain index and the feature vector of the smoking amount into a pre-constructed dose optimization model to obtain the optimal dose of paliperidone of the patient output by the dose optimization model. According to the method, the optimal dose of the paliperidone of the patient is predicted based on the dose optimization model, so that the accuracy of dose prediction is improved, a personalized treatment scheme is convenient to formulate, and the curative effect of the medicine is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a paliperidone dosage optimization method, device, electronic device and storage medium. Background Art

[0002] Paliperidone, as an atypical antipsychotic, is widely used in the treatment of schizophrenia and schizoaffective disorder. Its clinical efficacy has been verified by many studies, but it still faces major challenges in practical application. During the medication process, how to adjust the dose individually according to the patient's condition is a key issue that needs to be solved urgently. Existing dosing regimens are usually formulated based on group average data, considering only a single factor, and it is difficult to accurately match the individual needs of patients, resulting in unstable efficacy or increased risk of adverse reactions. Summary of the invention

[0003] The present invention provides a paliperidone dosage optimization method, device, electronic device and storage medium, which are used to solve the defect that the prior art formulates a paliperidone dosage regimen based on group average data, only considers a single factor, and is difficult to accurately match the individual needs of patients.

[0004] The present invention provides a paliperidone dosage optimization method, which is applied to a client, comprising: Acquiring status data of the patient, wherein the status data includes blood concentration of paliperidone, time domain index of heart rate variability (HRV), and smoking amount; Preprocessing the state data to obtain preprocessed state data, and extracting features from the preprocessed state data to obtain a feature vector of blood drug concentration, a feature vector of HRV time domain index, and a feature vector of smoking amount; Inputting the characteristic vector of the blood drug concentration, the characteristic vector of the HRV time-domain index, and the characteristic vector of the smoking amount into a pre-constructed dosage optimization model to obtain the optimal dosage of paliperidone for the patient output by the dosage optimization model; The dosage optimization model is trained based on historical status data of historical patients and historical doses of paliperidone of historical patients; the historical status data include historical blood drug concentration, historical HRV time domain index and historical smoking amount.

[0005] In some embodiments, the dose optimization model is a reinforcement learning model based on a long short-term memory (LSTM) network, and the dose optimization model includes an LSTM layer, and the LSTM layer includes a metabolic state gating layer, and the metabolic state gating layer is used to dynamically correct the metabolic rate parameters of the patient's paliperidone.

[0006] In some embodiments, inputting the feature vector of the blood drug concentration, the feature vector of the HRV time domain index, and the feature vector of the smoking amount into a pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model includes: Inputting the feature vector of the blood drug concentration, the feature vector of the HRV time domain index, and the feature vector of the smoking amount into a pre-constructed dose optimization model; Based on the dose optimization model, fusing the feature vector of the blood drug concentration, the feature vector of the HRV time domain index, and the feature vector of the smoking amount to obtain a fused feature vector, and calculating the reward values of multiple doses of paliperidone according to the fused feature vector, and determining the optimal dose from the multiple doses; Wherein, the reward value is calculated according to the efficacy reward data and the side effect penalty data.

[0007] In some embodiments, calculating the reward values of multiple doses of paliperidone according to the fused feature vector and determining the optimal dose from the multiple doses includes: Based on the fused feature vector, dynamically correcting the metabolic rate parameter of paliperidone for the patient, predicting the blood drug concentration after the patient ingests multiple doses of paliperidone, and obtaining the blood drug concentration prediction values corresponding to the multiple doses of paliperidone; According to the blood drug concentration prediction values corresponding to the multiple doses of paliperidone, calculating the reward values of the multiple doses of paliperidone, and determining the paliperidone dose with the highest reward value as the optimal dose.

[0008] In some embodiments, fusing the feature vector of the blood drug concentration, the feature vector of the HRV time domain index, and the feature vector of the smoking amount to obtain a fused feature vector includes: Calculating the respective attention weights of the feature vector of the blood drug concentration, the feature vector of the HRV time domain index, and the feature vector of the smoking amount; Based on the attention weights, performing weighted fusion on the feature vector of the blood drug concentration, the feature vector of the HRV time domain index, and the feature vector of the smoking amount to obtain the fused feature vector.

[0009] In some embodiments, the training process of the dose optimization model includes: Obtaining local clinical data, and constructing training data according to the local clinical data, where the training data includes the historical state data, historical action data, and historical reward data of historical patients; Among them, the historical state data includes: the historical blood drug concentration of paliperidone, the historical HRV time domain index, and the historical smoking amount; the historical action data is the historical dose of paliperidone; the historical reward data is calculated based on the historical efficacy data and the historical adverse reaction penalty data; Based on the training data, train the local initial dose optimization model. After the training is completed, obtain the local model parameters; Perform differential privacy encryption on the local model parameters to obtain the ciphertext of the local model parameters; Send the ciphertext of the local model parameters to the server side; the server side is used to perform weighted aggregation on the ciphertexts of the local model parameters of different clients to obtain the aggregated parameters, and based on the aggregated parameters, iteratively update the initial dose optimization model to obtain the dose optimization model, and send the dose optimization model to the client; Receive the dose optimization model sent by the server side.

[0010] In some embodiments, the initial dose optimization model includes an initial LSTM layer, and the initial LSTM layer includes a metabolic state gating layer. Training the local initial dose optimization model includes: Extract features from the historical blood drug concentration to obtain the feature vector of the historical blood drug concentration; Extract features from the historical HRV time domain index to obtain the feature vector of the historical HRV time domain index; Extract features from the historical smoking amount to obtain the feature vector of the historical smoking amount; Fuse the feature vector of the historical blood drug concentration, the feature vector of the historical HRV time domain index, and the feature vector of the historical smoking amount to obtain the historical fusion feature vector; Input the historical fusion feature vector and the historical dose into the initial LSTM layer to obtain the blood drug concentration prediction sequence corresponding to the historical dose of paliperidone output by the initial LSTM layer; Based on the historical blood drug concentration sequence and the blood drug concentration prediction sequence, calculate the weighted mean square error, and according to the weighted mean square error, iteratively optimize the parameters of the initial LSTM layer.

[0011] The present invention also provides a dose optimization device for paliperidone, which is applied to a client and includes: An acquisition unit, configured to acquire the state data of a patient, where the state data includes the blood drug concentration of paliperidone, the heart rate variability HRV time domain index, and the smoking amount; A data processing unit, configured to preprocess the status data to obtain preprocessed status data, and extract features from the preprocessed status data to obtain a feature vector of blood drug concentration, a feature vector of HRV time domain indexes, and a feature vector of smoking amount; A prediction unit, configured to input the feature vector of blood drug concentration, the feature vector of HRV time domain indexes, and the feature vector of smoking amount into a pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model; Wherein, the dose optimization model is trained based on the historical status data of historical patients and the historical doses of paliperidone of historical patients; the historical status data includes historical blood drug concentration, historical HRV time domain indexes, and historical smoking amount.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the dose optimization method of paliperidone as described in any one of the above is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the dose optimization method of paliperidone as described in any one of the above is implemented.

[0014] The dose optimization method, device, electronic device, and storage medium of paliperidone provided by the present invention obtain the status data of a patient, preprocess the status data, extract features from the preprocessed status data to obtain a feature vector of blood drug concentration, a feature vector of HRV time domain indexes, and a feature vector of smoking amount; input the feature vector of blood drug concentration, the feature vector of HRV time domain indexes, and the feature vector of smoking amount into a pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model, improving the accuracy of drug dose prediction, facilitating the formulation of personalized treatment plans, and enhancing drug efficacy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0016] Figure 1 It is a flowchart of the dose optimization method of paliperidone provided by the embodiment of the present invention.

[0017] Figure 2It is a schematic flowchart of the training process of the dose optimization model provided by an embodiment of the present invention.

[0018] Figure 3 It is a schematic structural diagram of the paliperidone dose optimization device provided by an embodiment of the present invention.

[0019] Figure 4 It is a schematic structural diagram of the electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts fall within the protection scope of the present invention.

[0021] Currently, the existing drug dose optimization methods have the following defects: (1) Inefficient multi-modal data fusion mechanism: Traditional systems fuse multi-modal data by simple splicing or linear weighting, resulting in poor cross-modal feature interaction ability and inability to dynamically allocate weights. The entropy of the fused feature information decreases, affecting the model prediction accuracy.

[0022] (2) Mismatch between model architecture and metabolic characteristics: Traditional LSTM models have not been improved for the dynamic time-varying characteristics of drug metabolism. For example, the weights of the forget gates are fixed and cannot respond to fluctuations in drug metabolism rates; the output layer only predicts the blood drug concentration at a single time point and does not cover the complete pharmacokinetics (PK) and pharmacodynamics (PD) curves.

[0023] (3) Difficulty in balancing privacy protection and model performance: Existing federated learning systems have high communication overhead due to full-gradient transmission, and static pseudonymization cannot meet the requirements of dynamic desensitization of personal information.

[0024] (4) Insufficiency in data preprocessing: Traditional systems have insufficient handling of the asynchrony problem of multi-modal data and only perform data preprocessing by mean filling or simple interpolation, resulting in distortion of temporal features.

[0025] To this end, an embodiment of the present invention provides a method, device, electronic device, and storage medium for optimizing the dosage of paliperidone. By obtaining the status data of a patient, preprocessing the status data, and extracting features from the preprocessed status data, feature vectors of blood drug concentration, HRV time-domain indexes, and smoking amount are obtained; the feature vectors of blood drug concentration, HRV time-domain indexes, and smoking amount are input into a pre-constructed dosage optimization model to obtain the optimal dosage of paliperidone for the patient output by the dosage optimization model. The present invention can improve the accuracy of drug dosage prediction, facilitate the formulation of personalized treatment plans, and enhance drug efficacy.

[0026] Figure 1 It is a schematic flowchart of the method for optimizing the dosage of paliperidone provided by the embodiment of the present invention. As Figure 1 shown, a method for optimizing the dosage of paliperidone is provided, which is applied to a client and includes the following steps: Step 110, Step 120, and Step 130. The process steps of this method are only a possible implementation manner of the present invention.

[0027] Step 110: Obtain the status data of the patient, where the status data includes the blood drug concentration of paliperidone, the HRV time-domain index of heart rate variability, and the smoking amount.

[0028] Among them, the HRV time-domain index is a mathematical parameter that quantifies heart rate variability by statistically analyzing the time series fluctuations of consecutive R-R intervals (R-R intervals), reflecting the dynamic balance ability of the autonomic nervous system (sympathetic / parasympathetic nervous system).

[0029] Optionally, detection techniques such as high-performance liquid chromatography-tandem mass spectrometry and immunoassay are used to detect the blood drug concentration of paliperidone in the patient; the Therapeutic Drug Monitoring (TDM) technique is used to obtain the blood drug concentration of paliperidone in the patient.

[0030] Optionally, the HRV time-domain indexes of the patient are collected through wearable devices, such as SDNN, RMSSD, etc.; SDNN represents the standard deviation of all normal R-R intervals (NN intervals) and is used to evaluate the overall autonomic nerve tension (combined activity of sympathetic + parasympathetic nerves). The higher the value (>50 ms), the stronger the physiological adaptation ability; RMSSD represents the root mean square of the differences between adjacent R-R intervals and is used to evaluate the parasympathetic nerve activity (vagal tone). It is sensitive to deep breathing and stress changes. The higher the value (>30 ms), the stronger the resilience.

[0031] Optionally, record the patient's daily smoking amount (such as 1 cigarette / day) and smoking time.

[0032] It should be noted that the blood drug concentration, HRV time-domain indexes, and smoking amount are time-series data within the current time period; the HRV time-domain indexes are early biomarkers of paliperidone cardiac toxicity, and the HRV time-domain indexes are the gold standard for quantifying the impact of paliperidone on the autonomic nervous system; smoking significantly reduces paliperidone exposure through CYP1A2 enzyme induction, and there is a negative correlation between the smoking amount and the blood drug concentration of paliperidone.

[0033] Step 120: Preprocess the status data to obtain the preprocessed status data, and extract features from the preprocessed status data to obtain the feature vectors of the blood drug concentration, the feature vectors of the HRV time-domain indexes, and the feature vectors of the smoking amount.

[0034] Optionally, align the blood drug concentration, HRV time-domain indexes, and smoking amount, and unify the asynchronous data streams to millisecond-level timestamps.

[0035] Optionally, interpolate the blood drug concentration, HRV time-domain indexes, and smoking amount, and use cubic spline interpolation to fill in the missing points.

[0036] Optionally, extract features from the preprocessed blood drug concentration to obtain the feature vectors of the blood drug concentration; extract features from the preprocessed HRV time-domain indexes to obtain the feature vectors of the HRV time-domain indexes; extract features from the preprocessed smoking amount to obtain the feature vectors of the smoking amount.

[0037] Optionally, perform normalization processing on the feature vectors of the blood drug concentration, the feature vectors of the HRV time-domain indexes, and the feature vectors of the smoking amount.

[0038] Step 130: Input the feature vectors of the blood drug concentration, the feature vectors of the HRV time-domain indexes, and the feature vectors of the smoking amount into the pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model.

[0039] Among them, the dose optimization model is trained based on the historical status data of historical patients and the historical doses of paliperidone for historical patients; the historical status data includes historical blood drug concentration, historical HRV time-domain indexes, and historical smoking amount.

[0040] Optionally, the dose optimization model is constructed based on a reinforcement learning model.

[0041] Optionally, the dose optimization model is trained based on the historical status data, historical action data (i.e., the historical doses of paliperidone), and historical reward data of historical patients.

[0042] In some embodiments, the dose optimization model is a reinforcement learning model based on a long short-term memory network (LSTM). The dose optimization model includes an LSTM layer, and the LSTM layer includes a metabolic state gating layer, which is used to dynamically correct the metabolic rate parameter of paliperidone for the patient.

[0043] Among them, the LSTM layer includes an input layer, a metabolic state gating layer, and an output layer; the metabolic state gating layer introduces a dynamic correction term for the HRV frequency domain index (low-frequency power LF / high-frequency power HF ratio); the LF / HF ratio is used to quantify the sympathetic-vagal balance state of the autonomic nervous system and has a key warning value in the cardiac safety monitoring of paliperidone.

[0044] It should be noted that the HRV frequency domain index can capture cardiac risks earlier than the time domain index; the HRV frequency domain index has a natural normalization characteristic and can be seamlessly integrated into the gating mechanism; the HRV frequency domain index can directly quantify the sympathetic intensity, accurately predict the fluctuations of the metabolic rate parameter, and has strong anti-interference ability.

[0045] In the embodiments of the present invention, by obtaining the state data of the patient, preprocessing the state data, and extracting features from the preprocessed state data, feature vectors of blood drug concentration, HRV time domain index, and smoking amount are obtained; the feature vectors of blood drug concentration, HRV time domain index, and smoking amount are input into a pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model, improving the accuracy of drug dose prediction, facilitating the formulation of personalized treatment plans, and enhancing the drug efficacy.

[0046] In some embodiments, step 130 inputs the feature vectors of blood drug concentration, HRV time domain index, and smoking amount into a pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model, including: Step 131: Input the feature vectors of blood drug concentration, HRV time domain index, and smoking amount into a pre-constructed dose optimization model; Step 132: Based on the dose optimization model, fuse the feature vectors of blood drug concentration, HRV time domain index, and smoking amount to obtain a fused feature vector, and calculate the reward values of multiple doses of paliperidone according to the fused feature vector, and determine the optimal dose from multiple doses; Among them, the reward value is calculated according to the efficacy reward data and the side effect penalty data.

[0047] Optionally, the weights of the eigenvectors of the blood drug concentration, the eigenvectors of the HRV time-domain indices, and the eigenvectors of the smoking amount are adaptively adjusted, and the eigenvectors of the blood drug concentration, the eigenvectors of the HRV time-domain indices, and the eigenvectors of the smoking amount are weighted and fused.

[0048] In some embodiments, according to the fused eigenvector, the reward values of multiple doses of paliperidone are calculated, and the optimal dose is determined from the multiple doses, including: Based on the fused eigenvector, the metabolic rate parameter of the patient's paliperidone is dynamically corrected, and the blood drug concentration after the patient takes multiple doses of paliperidone is predicted to obtain the predicted blood drug concentration values corresponding to the multiple doses of paliperidone; According to the predicted blood drug concentration values corresponding to the multiple doses of paliperidone, the reward values of the multiple doses of paliperidone are calculated, and the paliperidone dose with the highest reward value is determined as the optimal dose.

[0049] Optionally, the adverse reactions after the patient takes multiple doses of paliperidone are predicted to obtain the predicted adverse reaction results corresponding to the multiple doses of paliperidone.

[0050] Optionally, according to the predicted blood drug concentration values and the predicted adverse reaction results corresponding to the multiple doses of paliperidone, the reward values of the multiple doses of paliperidone are calculated.

[0051] Optionally, when the blood drug concentration is within the target range (20 - 60 ng / mL), the reward is +10; if an adverse reaction occurs, such as the confirmation of extrapyramidal symptoms (EPS), the penalty is -5. EPS is the most common adverse reaction of paliperidone, which is caused by the excessive blockade of dopamine D2 receptors in the central nervous system by the drug.

[0052] In some embodiments, the eigenvectors of the blood drug concentration, the eigenvectors of the HRV time-domain indices, and the eigenvectors of the smoking amount are fused to obtain a fused eigenvector, including: Calculate the attention weights of the eigenvectors of the blood drug concentration, the eigenvectors of the HRV time-domain indices, and the eigenvectors of the smoking amount respectively; Based on the attention weights, the eigenvectors of the blood drug concentration, the eigenvectors of the HRV time-domain indices, and the eigenvectors of the smoking amount are weighted and fused to obtain a fused eigenvector.

[0053] Optionally, the eigenvectors of the blood drug concentration, the eigenvectors of the HRV time-domain indices, and the eigenvectors of the smoking amount are mapped to a unified dimension.

[0054] Optionally, according to different scenarios, the weights of the blood drug concentration, the HRV time-domain indices, and the smoking amount are dynamically adjusted.

[0055] For example, during the renal function mutation period, increase the weight of the blood drug concentration; during the early stage of EPS, increase the weight of the HRV time domain index; during the smoking cessation transition period, increase the weight of the smoking amount.

[0056] It can be understood that through the attention mechanism, the contribution weights of the blood drug concentration characteristics, HRV time domain index characteristics, and smoking amount characteristics to the dose decision are dynamically quantified. Furthermore, the adaptive fusion of multi-modal characteristics can be achieved, thereby accurately predicting the optimal drug use strategy and improving the interpretability of clinical drug use decisions.

[0057] Figure 2 This is a schematic flowchart of the training process of the dose optimization model provided by the embodiments of the present invention. As Figure 2 shown, in some embodiments, the training process of the dose optimization model includes: Step 210: Obtain local clinical data, and construct training data according to the local clinical data. The training data includes historical status data, historical action data, and historical reward data of historical patients; Among them, the historical status data includes: the historical blood drug concentration of paliperidone, the historical HRV time domain index, and the historical smoking amount; the historical action data is the historical dose of paliperidone; the historical reward data is calculated based on historical efficacy data and historical adverse reaction penalty data.

[0058] Optionally, perform preprocessing on the local clinical data, such as data cleaning, alignment, interpolation, normalization, etc.

[0059] Optionally, dynamically update the identifiers of historical patients.

[0060] Step 220: Based on the training data, train the local initial dose optimization model. After training, obtain the local model parameters.

[0061] Optionally, sort the gradients of the local model parameters and retain the local model parameters with the top 10% gradients.

[0062] Step 230: Perform differential privacy encryption on the local model parameters to obtain the ciphertext of the local model parameters.

[0063] Step 240: Send the ciphertext of the local model parameters to the server side; Among them, the server side is used to perform weighted aggregation on the ciphertexts of the local model parameters of different clients to obtain aggregated parameters, and iteratively update the initial dose optimization model based on the aggregated parameters to obtain the dose optimization model, and send the dose optimization model to the client.

[0064] Step 250: Receive the dose optimization model sent by the server side.

[0065] It can be understood that by using the method of federated learning to train the initial dose optimization model to obtain the dose optimization model, and transmitting the local model parameters in an encrypted manner, the security of data transmission is improved, the communication overhead is reduced, and the efficiency of model training and the performance of the model are enhanced.

[0066] In some embodiments, the initial dose optimization model includes an initial LSTM layer, and the initial LSTM layer includes a metabolic state gating layer. Training the local initial dose optimization model includes: Performing feature extraction on historical blood drug concentrations to obtain a feature vector of historical blood drug concentrations; Performing feature extraction on historical HRV time domain indicators to obtain a feature vector of historical HRV time domain indicators; Performing feature extraction on historical smoking amounts to obtain a feature vector of historical smoking amounts; Fusing the feature vector of historical blood drug concentrations, the feature vector of historical HRV time domain indicators, and the feature vector of historical smoking amounts to obtain a historical fusion feature vector; Inputting the historical fusion feature vector and the historical dose into the initial LSTM layer to obtain a blood drug concentration prediction sequence corresponding to the historical dose of paliperidone output by the initial LSTM layer; Based on the historical blood drug concentration sequence and the blood drug concentration prediction sequence, calculating the weighted mean square error, and iteratively optimizing the parameters of the initial LSTM layer according to the weighted mean square error.

[0067] Optionally, calculating the respective historical attention weights of the feature vector of historical blood drug concentrations, the feature vector of historical HRV time domain indicators, and the feature vector of historical smoking amounts.

[0068] Optionally, based on the historical attention weights, performing weighted fusion on the feature vector of historical blood drug concentrations, the feature vector of historical HRV time domain indicators, and the feature vector of historical smoking amounts to obtain a historical fusion feature vector.

[0069] The following describes the dose optimization device for paliperidone provided by the embodiments of the present invention. The dose optimization device for paliperidone described below can be correspondingly referred to the paliperidone dose optimization method described above.

[0070] Figure 3 is a schematic structural diagram of the dose optimization device for paliperidone provided by the embodiments of the present invention. As Figure 3 shown, the dose optimization device 300 for paliperidone is applied to the client and includes: An acquisition unit 310, configured to acquire the status data of the patient, where the status data includes the blood drug concentration of paliperidone, the heart rate variability HRV time domain indicator, and the smoking amount; A data processing unit 320, configured to preprocess the status data to obtain preprocessed status data, and perform feature extraction on the preprocessed status data to obtain feature vectors of blood drug concentration, feature vectors of HRV time domain indexes, and feature vectors of smoking amount; A prediction unit 330, configured to input the feature vectors of blood drug concentration, feature vectors of HRV time domain indexes, and feature vectors of smoking amount into a pre-constructed dose optimization model, and obtain the optimal dose of paliperidone for the patient output by the dose optimization model; Wherein, the dose optimization model is trained based on the historical status data of historical patients and the historical doses of paliperidone of historical patients; the historical status data includes historical blood drug concentration, historical HRV time domain indexes, and historical smoking amount.

[0071] Optionally, the dose optimization model is a reinforcement learning model based on a long short-term memory network (LSTM). The dose optimization model includes an LSTM layer, and the LSTM layer includes a metabolic status gating layer, which is configured to dynamically correct the metabolic rate parameters of the patient's paliperidone.

[0072] Optionally, inputting the feature vectors of blood drug concentration, feature vectors of HRV time domain indexes, and feature vectors of smoking amount into a pre-constructed dose optimization model, and obtaining the optimal dose of paliperidone for the patient output by the dose optimization model includes: Inputting the feature vectors of blood drug concentration, feature vectors of HRV time domain indexes, and feature vectors of smoking amount into a pre-constructed dose optimization model; Based on the dose optimization model, fusing the feature vectors of blood drug concentration, feature vectors of HRV time domain indexes, and feature vectors of smoking amount to obtain a fused feature vector, and calculating the reward values of multiple doses of paliperidone according to the fused feature vector, and determining the optimal dose from multiple doses; Wherein, the reward value is calculated according to the efficacy reward data and the side effect penalty data.

[0073] Optionally, calculating the reward values of multiple doses of paliperidone according to the fused feature vector, and determining the optimal dose from multiple doses includes: Based on the fused feature vector, dynamically correcting the metabolic rate parameters of the patient's paliperidone, predicting the blood drug concentration after the patient takes multiple doses of paliperidone, and obtaining the blood drug concentration prediction values corresponding to multiple doses of paliperidone; Calculating the reward values of multiple doses of paliperidone according to the blood drug concentration prediction values corresponding to multiple doses of paliperidone, and determining the paliperidone dose with the highest reward value as the optimal dose.

[0074] Optionally, the feature vectors of blood drug concentration, the feature vectors of HRV time domain indexes, and the feature vectors of smoking amount are fused to obtain a fused feature vector, including: Calculate the attention weights of the feature vectors of blood drug concentration, the feature vectors of HRV time domain indexes, and the feature vectors of smoking amount respectively; Based on the attention weights, perform weighted fusion on the feature vectors of blood drug concentration, the feature vectors of HRV time domain indexes, and the feature vectors of smoking amount to obtain a fused feature vector.

[0075] Optionally, the training process of the dose optimization model includes: Obtain local clinical data, and construct training data according to the local clinical data. The training data includes historical state data, historical action data, and historical reward data of historical patients; Among them, the historical state data includes: historical blood drug concentration of paliperidone, historical HRV time domain indexes, and historical smoking amount; the historical action data is the historical dose of paliperidone; the historical reward data is calculated based on historical efficacy data and historical adverse reaction penalty data; Based on the training data, train the local initial dose optimization model. After the training is completed, obtain local model parameters; Perform differential privacy encryption on the local model parameters to obtain the ciphertext of the local model parameters; Send the ciphertext of the local model parameters to the server side; the server side is used to perform weighted aggregation on the ciphertexts of the local model parameters of different clients to obtain aggregated parameters, and perform iterative update on the initial dose optimization model based on the aggregated parameters to obtain the dose optimization model, and send the dose optimization model to the client; Receive the dose optimization model sent by the server side.

[0076] Optionally, the initial dose optimization model includes an initial LSTM layer, and the initial LSTM layer includes a metabolic state gating layer. Training the local initial dose optimization model includes: Extract features from the historical blood drug concentration to obtain the feature vector of the historical blood drug concentration; Extract features from the historical HRV time domain indexes to obtain the feature vector of the historical HRV time domain indexes; Extract features from the historical smoking amount to obtain the feature vector of the historical smoking amount; Fuse the feature vector of the historical blood drug concentration, the feature vector of the historical HRV time domain indexes, and the feature vector of the historical smoking amount to obtain a historical fused feature vector; Input the historical fused feature vector and the historical dose into the initial LSTM layer to obtain the predicted blood drug concentration sequence corresponding to the historical dose of paliperidone output by the initial LSTM layer; Based on the historical blood drug concentration sequence and the predicted blood drug concentration sequence, calculate the weighted mean square error, and iteratively optimize the parameters of the initial LSTM layer according to the weighted mean square error.

[0077] It should be noted here that the dosage optimization device for paliperidone provided in the embodiments of the present invention can implement all the method steps implemented in the above-mentioned embodiments of the paliperidone dosage optimization method, and can achieve the same technical effects. Therefore, the same parts and beneficial effects as those in the method embodiments will not be specifically described in this embodiment.

[0078] Figure 4 It is a schematic structural diagram of an electronic device provided in an embodiment of the present invention. As Figure 4 shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440. Among them, the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call the logical instructions in the memory 430 to execute the paliperidone dosage optimization method, which includes: obtaining the status data of the patient, where the status data includes the blood drug concentration of paliperidone, the HRV time-domain index of heart rate variability, and the smoking amount; preprocessing the status data to obtain the preprocessed status data, and performing feature extraction on the preprocessed status data to obtain a feature vector of the blood drug concentration, a feature vector of the HRV time-domain index, and a feature vector of the smoking amount; inputting the feature vector of the blood drug concentration, the feature vector of the HRV time-domain index, and the feature vector of the smoking amount into a pre-constructed dosage optimization model to obtain the optimal dosage of paliperidone for the patient output by the dosage optimization model; where the dosage optimization model is trained based on the historical status data of historical patients and the historical dosage of paliperidone of historical patients; the historical status data includes historical blood drug concentration, historical HRV time-domain index, and historical smoking amount.

[0079] In addition, when the logical instructions in the above-mentioned memory 430 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0080] On the other hand, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is used to implement the method for optimizing the dose of paliperidone provided by the above-mentioned various methods. The method includes: obtaining the status data of a patient, where the status data includes the blood drug concentration of paliperidone, the HRV time-domain index of heart rate variability, and the smoking amount; preprocessing the status data to obtain the preprocessed status data, and extracting features from the preprocessed status data to obtain the feature vectors of the blood drug concentration, the feature vectors of the HRV time-domain index, and the feature vectors of the smoking amount; inputting the feature vectors of the blood drug concentration, the feature vectors of the HRV time-domain index, and the feature vectors of the smoking amount into a pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model; where the dose optimization model is trained based on the historical status data of historical patients and the historical doses of paliperidone of historical patients; the historical status data includes historical blood drug concentration, historical HRV time-domain index, and historical smoking amount.

[0081] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0082] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or equivalently replace some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for optimizing the dosage of paliperidone, characterized in that, Applied to the client, including: Obtain the status data of the patient, where the status data includes the blood drug concentration of paliperidone, the HRV time domain index of heart rate variability, and the smoking amount; Preprocess the status data to obtain preprocessed status data, and extract features from the preprocessed status data to obtain a feature vector of blood drug concentration, a feature vector of HRV time domain index, and a feature vector of smoking amount; Input the feature vector of blood drug concentration, the feature vector of HRV time domain index, and the feature vector of smoking amount into a pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model; Among them, the dose optimization model is trained based on the historical status data of historical patients and the historical dose of paliperidone of historical patients; the historical status data includes historical blood drug concentration, historical HRV time domain index, and historical smoking amount.

2. The dosage optimization method of paliperidone according to claim 1, characterized in that, The dose optimization model is a reinforcement learning model based on the long short-term memory network LSTM. The dose optimization model includes an LSTM layer, and the LSTM layer includes a metabolic state gating layer, which is used to dynamically correct the metabolic rate parameter of paliperidone for the patient.

3. The dosage optimization method of paliperidone according to claim 1, wherein The step of inputting the feature vector of blood drug concentration, the feature vector of HRV time domain index, and the feature vector of smoking amount into a pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model includes: Input the feature vector of blood drug concentration, the feature vector of HRV time domain index, and the feature vector of smoking amount into a pre-constructed dose optimization model; Based on the dose optimization model, fuse the feature vector of blood drug concentration, the feature vector of HRV time domain index, and the feature vector of smoking amount to obtain a fused feature vector. According to the fused feature vector, calculate the reward values of multiple doses of paliperidone, and determine the optimal dose from the multiple doses; Among them, the reward value is calculated based on efficacy reward data and side effect penalty data.

4. The dosage optimization method of paliperidone according to claim 3, characterized in that, The step of calculating the reward values of multiple doses of paliperidone according to the fused feature vector and determining the optimal dose from the multiple doses includes: Based on the fused feature vector, dynamically correct the metabolic rate parameter of paliperidone for the patient, and predict the blood drug concentration after the patient takes multiple doses of paliperidone to obtain the blood drug concentration prediction values corresponding to multiple doses of paliperidone; According to the blood drug concentration prediction values corresponding to multiple doses of paliperidone, calculate the reward values of multiple doses of paliperidone, and determine the paliperidone dose with the highest reward value as the optimal dose.

5. The dosage optimization method of paliperidone according to claim 3, characterized in that, The step of fusing the feature vector of blood drug concentration, the feature vector of HRV time domain index, and the feature vector of smoking amount to obtain a fused feature vector includes: Calculate the respective attention weights of the feature vector of blood drug concentration, the feature vector of HRV time domain index, and the feature vector of smoking amount; Based on the attention weights, perform weighted fusion on the feature vector of blood drug concentration, the feature vector of HRV time domain index, and the feature vector of smoking amount to obtain the fused feature vector.

6. The method for optimizing the dosage of paliperidone according to claim 1, wherein, The training process of the dose optimization model includes: Obtain local clinical data, and construct training data according to the local clinical data. The training data includes historical status data, historical action data, and historical reward data of historical patients; Among them, the historical status data includes: historical blood drug concentration of paliperidone, historical HRV time domain index, and historical smoking amount; the historical action data is the historical dose of paliperidone; the historical reward data is calculated based on historical efficacy data and historical adverse reaction penalty data; Based on the training data, train the initial dose optimization model locally. After the training is completed, obtain the local model parameters; Perform differential privacy encryption on the local model parameters to obtain the ciphertext of the local model parameters; Send the ciphertext of the local model parameters to the server side; the server side is used to perform weighted aggregation on the ciphertexts of the local model parameters of different clients to obtain aggregation parameters, and iteratively update the initial dose optimization model based on the aggregation parameters to obtain the dose optimization model, and send the dose optimization model to the client; Receive the dose optimization model sent by the server side.

7. The dosage optimization method of paliperidone according to claim 6, characterized in that, The initial dose optimization model includes an initial LSTM layer, and the initial LSTM layer includes a metabolic state gating layer. Training the initial dose optimization model locally includes: Extract features from the historical blood drug concentration to obtain the feature vector of the historical blood drug concentration; Extract features from the historical HRV time domain index to obtain the feature vector of the historical HRV time domain index; Extract features from the historical smoking amount to obtain the feature vector of the historical smoking amount; Fuse the feature vector of the historical blood drug concentration, the feature vector of the historical HRV time domain index, and the feature vector of the historical smoking amount to obtain a historical fusion feature vector; Input the historical fusion feature vector and the historical dose into the initial LSTM layer to obtain the blood drug concentration prediction sequence corresponding to the historical dose of paliperidone output by the initial LSTM layer; Based on the historical blood drug concentration sequence and the blood drug concentration prediction sequence, calculate the weighted mean square error, and iteratively optimize the parameters of the initial LSTM layer according to the weighted mean square error.

8. A dosage optimization device for paliperidone, characterized in that, When applied to the client, it includes: An acquisition unit for acquiring the status data of the patient. The status data includes the blood drug concentration of paliperidone, the heart rate variability HRV time domain index, and the smoking amount; A data processing unit for preprocessing the status data to obtain the preprocessed status data, and extracting features from the preprocessed status data to obtain the feature vector of the blood drug concentration, the feature vector of the HRV time domain index, and the feature vector of the smoking amount; A prediction unit for inputting the feature vector of the blood drug concentration, the feature vector of the HRV time domain index, and the feature vector of the smoking amount into the pre-constructed dose optimization model to obtain the optimal dose of paliperidone for the patient output by the dose optimization model; Among them, the dose optimization model is trained based on the historical state data of historical patients and the historical doses of paliperidone of historical patients; the historical state data includes historical blood drug concentration, historical HRV time domain index, and historical smoking amount.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the dose optimization method of paliperidone according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the dose optimization method of paliperidone according to any one of claims 1 to 7.

Images