Bladder postoperative medication accurate regulation and control method and system based on artificial intelligence
By integrating multimodal data through artificial intelligence technology, individualized status representations are constructed and medication plans are optimized, solving the problems of lag and inappropriateness in traditional postoperative bladder medication plans, achieving personalized precision medication, and improving safety and scientific rigor.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE FIRST MEDICAL CENT CHINESE PLA GENERAL HOSPITAL
- Filing Date
- 2026-03-11
- Publication Date
- 2026-05-05
AI Technical Summary
Traditional postoperative bladder surgery medication regimens rely on physician experience and cannot dynamically capture changes in the patient's physiological state, leading to delayed or inappropriate medication. Furthermore, the lack of systematic assessment of drug interactions makes it difficult to balance efficacy, safety, and cost-effectiveness.
An artificial intelligence-based approach is used to obtain patients' physiological indicators, medication history, and clinical symptoms through multimodal feature fusion. Feature extraction and semantic mapping are performed to construct individualized state representations. Combined with pharmacokinetic models and interaction rules, candidate drug regulation schemes are generated. The optimal scheme is then solved through multi-objective optimization, and the scheme is adjusted and iteratively updated in real time.
It enables precise medication based on the individual patient's condition, improves the safety and scientific nature of medication regimens, dynamically balances efficacy, side effects and costs, and achieves truly personalized precision medication.
Smart Images
Figure CN121983228A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to artificial intelligence technology, and more particularly to a method and system for precise regulation of postoperative bladder medication based on artificial intelligence. Background Technology
[0002] Bladder surgery is a common procedure in urology, and postoperative medication management directly affects the quality of patient recovery and the incidence of complications. Traditional postoperative bladder medication regimens rely heavily on the clinical physician's experience, determining standardized dosages and frequencies based on the patient's basic physiological indicators and the type of surgery. In clinical practice, physicians typically refer to the recommended dosage range in the drug instructions, making linear adjustments based on the patient's age, weight, liver and kidney function, and assessing the medication's effectiveness based on routine postoperative monitoring indicators such as body temperature, blood pressure, and urine output. This experience-driven medication model meets basic treatment needs in most routine cases and has formed a relatively mature clinical pathway system.
[0003] Due to significant individual differences among patients, the metabolic rate, time to peak blood concentration, and tissue distribution characteristics of the same dose of drug can vary by several times in different patients, leading to underdosing or overdosing in some patients. Traditional regimens struggle to dynamically capture continuous changes in the patient's physiological state during postoperative recovery and cannot adjust medication parameters in a timely manner based on real-time fluctuations in the patient's condition, often resulting in delayed medication or inappropriate adjustment frequencies. When multiple drugs are used in combination, existing regimens lack a systematic assessment mechanism for complex drug interactions, which may lead to unintended effects such as enhanced or weakened drug efficacy, increasing the risk of adverse reactions. In terms of efficacy assessment, relying solely on subjective symptom descriptions and limited objective indicators makes it difficult to accurately quantify the comprehensive benefits of medication regimens, failing to achieve a precise balance between efficacy, safety, and cost-effectiveness, thus affecting overall treatment efficiency. Summary of the Invention
[0004] This invention provides a method and system for precise regulation of postoperative bladder medication based on artificial intelligence, which can solve the problems in the prior art.
[0005] A first aspect of the present invention provides a method for precise regulation of postoperative bladder medication based on artificial intelligence, comprising: Obtain postoperative physiological index data, medication history data, and clinical symptom data of patients; perform feature extraction and semantic mapping on the physiological index data, medication history data, and clinical symptom data based on multimodal feature fusion to obtain individualized status representation of patients; Based on the individualized patient status characterization, pharmacokinetic parameters and efficacy response curves under different medication regimens are calculated through time-dependent modeling, and a set of candidate medication regulation regimens is generated by combining drug interaction constraint rules. Based on the set of candidate drug control schemes and postoperative recovery goals, the optimal drug control scheme is solved under the constraints of maximizing efficacy, minimizing side effects, and drug cost. The weight coefficients of each optimization goal are dynamically adjusted based on the risk characteristics in the individualized patient status representation. Medication adjustments are made based on the optimal medication control plan, and postoperative actual efficacy data and adverse reaction data are collected to form feedback information. The feedback information is then used to iteratively update the drug efficacy prediction parameters and weighting coefficients.
[0006] Based on multimodal feature fusion, feature extraction and semantic mapping are performed on the physiological indicator data, the medication history data, and the clinical symptom data to obtain a patient's individualized status representation, including: Temporal features were extracted from physiological indicator data to obtain physiological indicator temporal feature vectors; drug effect correlation analysis was performed on medication history data to obtain medication history correlation feature vectors; and symptom severity was quantified from clinical symptom data to obtain clinical symptom quantitative feature vectors. The time-series feature vectors of the physiological indicators, the medication history-related feature vectors, and the clinical symptom quantification feature vectors are cross-modal semantically aligned. The association weights between each modality feature are calculated through an attention weighting mechanism. Based on the association weights, each modality feature is adaptively fused to obtain a fused feature vector. The fused feature vector is then subjected to a nonlinear mapping transformation to obtain a patient-specific state representation.
[0007] Based on the individualized patient status characterization, pharmacokinetic parameters and efficacy response curves under different medication regimens are calculated through time-dependent modeling. A set of candidate medication regulation regimens is then generated by combining drug interaction constraint rules, including: Based on the individualized state representation of the patient, a patient-specific baseline state for predicting drug efficacy is constructed. Based on the baseline state for predicting drug efficacy and historical medication time series data, the residual influence coefficient of previous medication on the current drug efficacy is calculated by recursive time series dependency modeling, and the time series residual influence matrix is obtained. The time-series residual effect matrix is coupled with the medication regimen to be evaluated to predict the pharmacokinetic parameter change trajectory and efficacy response curve under the medication regimen to be evaluated; Obtain a drug interaction constraint rule base, and perform safety constraint screening on the efficacy response curve based on the drug interaction constraint rule base to eliminate medication regimens that violate drug interaction constraints; The medication regimens selected through the aforementioned safety constraints are combined and arranged to generate a set of candidate medication control regimens.
[0008] The time-series residual effect matrix is coupled with the medication regimen to be evaluated to predict the pharmacokinetic parameter change trajectory and efficacy response curve under the medication regimen to be evaluated, including: The drug dosage and administration time information in the medication regimen to be evaluated are converted into a time-series input vector, and the time-series input vector is coupled with the time-series residual effect matrix by a time-by-time weighted operation to obtain a modified metabolic parameter sequence containing residual effects. Based on the metabolic parameter values at each time point in the modified metabolic parameter sequence, the drug absorption rate, distribution rate, and elimination rate at the corresponding time point are calculated, and the drug absorption rate, distribution rate, and elimination rate at consecutive time points are combined to form a trajectory of changes in drug metabolism kinetic parameters. Based on the drug absorption rate and elimination rate at each moment in the trajectory of the changes in the pharmacokinetic parameters, the drug blood concentration value at each moment is calculated, and the drug blood concentration values at each moment are arranged in time sequence to form a blood concentration time series curve. The blood concentration time series curve is mapped to the therapeutic effect space to obtain the therapeutic effect response curve.
[0009] Based on the set of candidate drug control schemes and postoperative recovery goals, the optimal drug control scheme is determined under the constraints of maximizing efficacy, minimizing side effects, and controlling medication costs. Based on the risk characteristics in the individualized patient status representation, the patient risk level is identified, and adaptive weight coefficients are assigned to the efficacy maximization target, side effect minimization target, and medication cost constraint target based on the patient risk level, so as to obtain the patient-specific target weight configuration. The patient-specific target weight configuration is weighted and summed with the efficacy prediction value, side effect prediction value, and medication cost value of each candidate scheme in the candidate medication control scheme set to obtain the comprehensive evaluation score of each candidate scheme. The candidate schemes in the candidate drug control scheme set are ranked according to the comprehensive evaluation score, and the candidate scheme with the highest comprehensive evaluation score is selected as the optimal drug control scheme from the ranking results.
[0010] Medication adjustments are made based on the optimal medication control plan, and feedback information is generated by collecting postoperative efficacy data and adverse reaction data, including: Based on the drug type, dosage and administration time information in the optimal medication control plan, a medication execution instruction is generated and sent to the medication execution terminal to complete the medication adjustment. At the same time, the execution timestamp and execution status identifier of the medication execution instruction are recorded. During the preset monitoring period after the medication execution instruction is executed, physiological indicator change data of the patient is collected by physiological monitoring equipment as postoperative efficacy data, and adverse reaction symptom description and severity identification of the patient are obtained through the adverse reaction collection interface as adverse reaction data. The execution timestamp, execution status identifier, postoperative actual efficacy data, and adverse reaction data are associated and bound to generate feedback information containing medication execution information and actual response information.
[0011] A second aspect of the present invention provides an artificial intelligence-based system for precise control of postoperative bladder medication, comprising: The data acquisition unit is used to acquire the patient's postoperative physiological index data, medication history data, and clinical symptom data; based on multimodal feature fusion, the physiological index data, medication history data, and clinical symptom data are used to extract features and perform semantic mapping to obtain the patient's individualized state representation; The scheme generation unit is used to calculate the pharmacokinetic parameters and efficacy response curves of different medication regimens based on the individualized state characterization of the patient through time-series dependency modeling, and to generate a set of candidate medication regulation schemes in combination with drug interaction constraint rules. The scheme optimization unit is used to solve the optimal drug control scheme based on the set of candidate drug control schemes and postoperative recovery goals, under the constraints of maximizing efficacy, minimizing side effects, and drug cost, and dynamically adjust the weight coefficients of each optimization objective based on the risk characteristics in the individualized state representation of the patient. The feedback update unit is used to adjust the medication based on the optimal medication control plan, collect postoperative actual efficacy data and adverse reaction data to form feedback information, and use the feedback information to iteratively update the drug efficacy prediction parameters and weight coefficients.
[0012] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.
[0013] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.
[0014] The beneficial effects of this application are as follows: This invention integrates heterogeneous data such as postoperative physiological indicators, medication history, and clinical symptoms of patients through multimodal feature fusion technology, achieving a comprehensive and accurate representation of the individualized state of patients. It effectively overcomes the problem of inaccurate patient state characterization caused by the single data dimension in traditional methods, and provides a reliable data foundation for subsequent medication decisions.
[0015] By modeling drug metabolism kinetic parameters and efficacy response curves through time-dependent relationships, it is possible to accurately predict the dynamic changes of different medication regimens in the patient's body. Combined with drug interaction constraint rules, a candidate regimen set is generated, avoiding drug incompatibilities and adverse interactions, and significantly improving the safety and scientific nature of medication regimens.
[0016] The multi-objective optimization framework constructed in this invention achieves a dynamic balance between maximizing efficacy, minimizing side effects, and drug cost constraints. It adaptively adjusts the weight coefficients of each optimization objective based on the risk characteristics in the individualized state representation of the patient, so that the drug control plan can fully consider the treatment effect, while also taking into account the patient's safety and economic burden, thus realizing truly personalized and precise drug use. Attached Figure Description
[0017] Figure 1 This is a flowchart illustrating the artificial intelligence-based method for precise control of postoperative bladder medication according to an embodiment of the present invention. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.
[0020] Figure 1 This is a flowchart illustrating the artificial intelligence-based method for precise control of postoperative bladder medication according to an embodiment of the present invention. Figure 1 As shown, the method includes: Obtain postoperative physiological index data, medication history data, and clinical symptom data of patients; perform feature extraction and semantic mapping on the physiological index data, medication history data, and clinical symptom data based on multimodal feature fusion to obtain individualized status representation of patients; Based on the individualized patient status characterization, pharmacokinetic parameters and efficacy response curves under different medication regimens are calculated through time-dependent modeling, and a set of candidate medication regulation regimens is generated by combining drug interaction constraint rules. Based on the set of candidate drug control schemes and postoperative recovery goals, the optimal drug control scheme is solved under the constraints of maximizing efficacy, minimizing side effects, and drug cost. The weight coefficients of each optimization goal are dynamically adjusted based on the risk characteristics in the individualized patient status representation. Medication adjustments are made based on the optimal medication control plan, and postoperative actual efficacy data and adverse reaction data are collected to form feedback information. The feedback information is then used to iteratively update the drug efficacy prediction parameters and weighting coefficients.
[0021] In one optional implementation, feature extraction and semantic mapping are performed on the physiological indicator data, the medication history data, and the clinical symptom data based on multimodal feature fusion to obtain a patient's individualized state representation, including: Temporal features were extracted from physiological indicator data to obtain physiological indicator temporal feature vectors; drug effect correlation analysis was performed on medication history data to obtain medication history correlation feature vectors; and symptom severity was quantified from clinical symptom data to obtain clinical symptom quantitative feature vectors. The time-series feature vectors of the physiological indicators, the medication history-related feature vectors, and the clinical symptom quantification feature vectors are cross-modal semantically aligned. The association weights between each modality feature are calculated through an attention weighting mechanism. Based on the association weights, each modality feature is adaptively fused to obtain a fused feature vector. The fused feature vector is then subjected to a nonlinear mapping transformation to obtain a patient-specific state representation.
[0022] After acquiring the patient's postoperative physiological data, a sliding time window was used to segment the blood pressure, heart rate, body temperature, urine output, and blood biochemical index sequences. For the data within each time window, statistical features, including mean, variance, peak value, and rate of change, were extracted. Simultaneously, a one-dimensional convolutional neural network was used to extract local temporal pattern features, capturing the dynamic trends of physiological indicators at different postoperative time periods. The extracted statistical features and deep features were concatenated to form a 128-dimensional temporal feature vector of physiological indicators, which reflects the temporal evolution of the patient's postoperative physiological state.
[0023] For historical medication data, a drug action association graph is constructed, with previously used drugs as nodes and synergistic or antagonistic effects between drugs as edge weights. A graph neural network is used to perform message passing on this association graph, aggregating neighborhood information for each drug node to obtain implicit association patterns between drugs. Simultaneously, numerical features such as dosage, duration of use, and dosing interval are extracted and concatenated with the node embedding vector output by the graph neural network to obtain a 96-dimensional historical medication association feature vector. This vector encodes the complex relationships between a patient's past medications.
[0024] For clinical symptom data, a symptom severity scoring system was established, converting discrete symptom descriptions such as pain level, frequency of nausea and vomiting, and degree of difficulty urinating into continuous numerical scores from 0 to 10. For the textual descriptive symptoms recorded by doctors, a pre-trained language model in the medical field was used to extract semantic features, which were then mapped into fixed-dimensional vector representations through a fully connected layer. The quantized symptom scores were concatenated with the textual semantic features to form a 64-dimensional quantized feature vector of clinical symptoms.
[0025] In the cross-modal semantic alignment stage, the three types of feature vectors are mapped to a unified semantic space through independent linear projection layers, with a unified dimension of 256. A multi-head attention mechanism is used to calculate the semantic similarity between features of different modalities, where physiological indicator features are used as query vectors, and medication history features and clinical symptom features are used as key-value pairs. Attention scores are calculated by scaling the dot product. After the scores are normalized by softmax, an association weight matrix is obtained, where each element of the matrix represents the association strength between a certain physiological indicator feature and a medication history or clinical symptom feature.
[0026] Based on the calculated association weights, the three modal features are weighted and fused. Specifically, each dimension of the medication history feature vector is multiplied by its corresponding weight coefficient, and the clinical symptom feature vector is also weighted. Then, the weighted features are added element-wise to the physiological indicator features. To enhance the expressive power of the fused features, a gating mechanism is introduced to dynamically adjust the contribution of each modality. The gating value is calculated using the sigmoid function to control the passing ratio of each modality feature, resulting in a fused feature vector with a dimension of 256.
[0027] The fused feature vector is input into a nonlinear mapping network containing two fully connected layers. The first layer has 512 neurons and uses ReLU as the activation function, while the second layer has an output dimension of 128. Batch normalization layers are added during the mapping process to stabilize the training process, and dropout is used to prevent overfitting. The final result is a patient-specific state representation vector, which comprehensively encodes the deep semantic information of the patient's physiological state, medication background, and clinical symptoms, providing accurate individualized feature input for subsequent medication regimen generation.
[0028] In one optional implementation, based on the patient's individualized state characterization, pharmacokinetic parameters and efficacy response curves under different medication regimens are calculated through time-dependent modeling, and a set of candidate medication regulation regimens is generated by combining drug interaction constraint rules, including: Based on the individualized state representation of the patient, a patient-specific baseline state for predicting drug efficacy is constructed. Based on the baseline state for predicting drug efficacy and historical medication time series data, the residual influence coefficient of previous medication on the current drug efficacy is calculated by recursive time series dependency modeling, and the time series residual influence matrix is obtained. The time-series residual effect matrix is coupled with the medication regimen to be evaluated to predict the pharmacokinetic parameter change trajectory and efficacy response curve under the medication regimen to be evaluated; Obtain a drug interaction constraint rule base, and perform safety constraint screening on the efficacy response curve based on the drug interaction constraint rule base to eliminate medication regimens that violate drug interaction constraints; The medication regimens selected through the aforementioned safety constraints are combined and arranged to generate a set of candidate medication control regimens.
[0029] Postoperative physiological data, including liver and kidney function indicators, historical blood drug concentrations, and metabolic enzyme activity parameters, were acquired from patients. These data, along with clinical symptom data such as pain scores and inflammatory response levels, were normalized to form a multidimensional feature vector. An attention mechanism was used to assign dynamic weights to different features. The influence of liver and kidney function indicators on drug metabolism rate was weighted at 0.4 to 0.6, and the influence of historical blood drug concentrations on residual effects was weighted at 0.3 to 0.5. A weighted fusion calculation was then performed to obtain a baseline state vector reflecting the patient's current metabolic capacity and drug sensitivity. This vector includes clearance parameters, volume of distribution parameters, and an individualized drug sensitivity coefficient. The clearance parameter was corrected based on creatinine clearance and liver enzyme levels, while the volume of distribution parameter was calculated based on the patient's body weight and fluid distribution ratio.
[0030] Medication timestamps, drug types, dosages, and routes of administration for patients within the past 72 hours were extracted to construct a time-series medication sequence. This sequence was encoded using a Long Short-Term Memory (LSTM) network, with a time decay factor λ ranging from 0.1 to 0.3, and then processed using an exponential decay function. Calculate the residual effect strength of preceding drug administration at the current time. For analgesics with long half-lives, the residual effect coefficient remains above 0.6 within 24 hours after administration; for antibiotics, the residual effect coefficient remains above 0.4 within 12 hours after administration. Fill the matrix structure with the residual effect coefficients of different drugs at different time points to form a time-series residual effect matrix with the dimension of the number of drug types × the number of time steps.
[0031] The drug dosage, dosing interval parameters, and time-series residual effect matrix of the treatment regimen to be evaluated are multiplied element-wise, and the clearance parameter in the baseline state vector is superimposed. The dynamic changes in blood drug concentration over time are calculated using a system of differential equations from a compartment model. Specifically, a first-order kinetic equation is used to describe the drug absorption process, with the absorption rate constant adjusted according to the route of administration. The absorption constant is set to infinity for intravenous administration and 0.5 to 2.0 per hour for oral administration. The elimination process is described using the Michaelis-Menten equation, and the maximum elimination rate is proportional to the clearance parameter in the baseline state vector. Based on the blood drug concentration curve and pharmacodynamic parameters, the Sigmoid function is used to calculate the drug concentration. Calculate the intensity of the therapeutic response, where For maximum therapeutic effect, The effective concentration is 50%, and n is the Hill coefficient, which is dynamically adjusted based on the drug sensitivity coefficient in the individualized patient status characterization.
[0032] The drug interaction constraint rule base retrieves interaction rules involving all drugs in the current treatment regimen under evaluation. The rule base includes rules for competitive inhibition of drug-metabolizing enzymes, competitive renal excretion, and synergistic or antagonistic effects. For drug combinations using the CYP3A4 metabolic pathway, a safety threshold of no more than 1.5 for the coefficient of increase in blood drug concentration is set. For drug combinations with a risk of cumulative nephrotoxicity, the sum of the renal clearances of the two drugs must not reduce the glomerular filtration rate by more than 20%. The peak concentration, trough concentration, and area under the curve in the efficacy response curve are compared with the safety concentration window, and regimens with peak concentrations exceeding 1.2 times the upper limit of therapeutic concentration or trough concentrations below 0.8 times the lower limit of therapeutic concentration are eliminated.
[0033] Single-drug regimens that pass safety screening will be combined and arranged according to dosing time and dosage level. For cases requiring combination therapy, combinations will be made in the priority order of analgesics, anti-infectives, and anti-inflammatory drugs to ensure that the efficacy of the primary therapeutic target drug is not interfered with by adjuvant drugs. The generated candidate regimen set includes the dosing schedule, the types and dosages of drugs at each time point, and the predicted efficacy response curve and drug metabolism parameter change trajectory within 72 hours, with the number of regimens controlled between 50 and 200.
[0034] In one optional implementation, the time-series residual effect matrix is coupled with the medication regimen to be evaluated to predict the pharmacokinetic parameter change trajectory and efficacy response curve under the medication regimen to be evaluated, including: The drug dosage and administration time information in the medication regimen to be evaluated are converted into a time-series input vector, and the time-series input vector is coupled with the time-series residual effect matrix by a time-by-time weighted operation to obtain a modified metabolic parameter sequence containing residual effects. Based on the metabolic parameter values at each time point in the modified metabolic parameter sequence, the drug absorption rate, distribution rate, and elimination rate at the corresponding time point are calculated, and the drug absorption rate, distribution rate, and elimination rate at consecutive time points are combined to form a trajectory of changes in drug metabolism kinetic parameters. Based on the drug absorption rate and elimination rate at each moment in the trajectory of the changes in the pharmacokinetic parameters, the drug blood concentration value at each moment is calculated, and the drug blood concentration values at each moment are arranged in time sequence to form a blood concentration time series curve. The blood concentration time series curve is mapped to the therapeutic effect space to obtain the therapeutic effect response curve.
[0035] The drug dosage information, dosing time information, and dosing method information of the medication regimen to be evaluated are obtained. The dosage values of each drug are normalized, and the dose intensity of each drug at each dosing time is represented by a floating-point vector. The dosing time is represented by a time offset relative to a reference time, and the dosing time is encoded as a timestamp sequence. The drug dosage intensity vector is concatenated with the timestamp sequence to form a vector with dimension [missing information]. The temporal input vector is given by T, where T represents the number of time steps and D represents the feature dimension.
[0036] The feature vector at each time step of the time-series input vector is multiplied by the time-series residual effect matrix. The time-series residual effect matrix records the residual effect intensity of historical drug use on the current time step, and the element values in the matrix represent the influence coefficients of residual drug concentration on the current metabolic parameters at different time intervals. For the feature vector at time step $t$ of the time-series input vector, the influence coefficient vector corresponding to that time step is extracted from the time-series residual effect matrix. The feature vector and the influence coefficient vector are multiplied element-wise to obtain the corrected feature vector considering the residual effect. The corrected feature vectors of all historical time steps before this time step are accumulated to obtain the cumulative residual effect value at time step $t$. The original metabolic parameters are weighted and summed with the cumulative residual effect value, and the weight coefficients are determined according to the time decay function to obtain the corrected metabolic parameter value at that time step. The corrected metabolic parameter sequence is constructed by traversing all time steps.
[0037] The absorption constant, volume of distribution, and elimination constant at time t are extracted from the modified metabolic parameter sequence. Based on a one-compartment pharmacokinetic model, the drug absorption rate at that time is calculated by multiplying the administered dose by the absorption constant. The plasma drug concentration at that time is extracted, and the distribution rate in tissues is obtained by dividing the plasma drug concentration by the volume of distribution. The elimination rate at that time is calculated by multiplying the plasma drug concentration by the elimination constant. This calculation is repeated for consecutive time points to obtain the absorption rate, distribution rate, and elimination rate values at each time point. These three rate values are arranged in chronological order to form three independent kinetic parameter change trajectory curves.
[0038] Numerical integration was performed on the drug absorption and elimination rates in the pharmacokinetic parameter trajectory. The initial blood drug concentration was set to zero. At time t, the blood drug concentration was calculated by subtracting the cumulative elimination from the total drug absorbed up to that time. The trapezoidal integral method was used to numerically integrate the absorption and elimination rate curves to obtain the cumulative absorption and elimination at each time point. The calculated blood drug concentration values at each time point were arranged in chronological order to form a time-series blood drug concentration curve.
[0039] A mapping function between blood drug concentration and therapeutic efficacy is constructed, described by an sigmoid dose-response curve. Blood drug concentration values at each time point are extracted from the time-series curve, and these values are substituted into the mapping function to calculate the corresponding efficacy score. The efficacy score is quantified on a percentage basis, with higher scores indicating more significant treatment effects. The efficacy scores calculated at each time point are arranged chronologically to form a therapeutic response curve. This curve reflects the evolution of the patient's efficacy over time under the evaluated medication regimen and is used for subsequent construction and solution of the objective function.
[0040] In one optional implementation, the optimal medication control scheme is solved based on the set of candidate medication control schemes and the postoperative recovery target, under the constraints of maximizing efficacy, minimizing side effects, and medication cost. Based on the risk characteristics in the individualized patient status representation, the patient risk level is identified, and adaptive weight coefficients are assigned to the efficacy maximization target, side effect minimization target, and medication cost constraint target based on the patient risk level, so as to obtain the patient-specific target weight configuration. The patient-specific target weight configuration is weighted and summed with the efficacy prediction value, side effect prediction value, and medication cost value of each candidate scheme in the candidate medication control scheme set to obtain the comprehensive evaluation score of each candidate scheme. The candidate schemes in the candidate drug control scheme set are ranked according to the comprehensive evaluation score, and the candidate scheme with the highest comprehensive evaluation score is selected as the optimal drug control scheme from the ranking results.
[0041] After the candidate drug control protocol set is formed, risk feature vectors are extracted from the individualized patient status representation, including age factors, underlying disease indicators, liver and kidney function parameters, and previous drug allergy records. A fuzzy C-means clustering algorithm is used to map the patient risk feature vectors to a pre-defined risk level system, which classifies patients into low-risk, medium-risk, and high-risk levels. For low-risk patients, the weight coefficient α for maximizing efficacy is set to 0.6, the weight coefficient β for minimizing side effects is set to 0.25, and the weight coefficient γ for drug cost constraints is set to 0.15. For medium-risk patients, α is adjusted to 0.45, β is adjusted to 0.4, and γ remains at 0.15. For high-risk patients, α is set to 0.3, β is increased to 0.55, and γ is set to 0.15. This adaptive weight configuration strategy ensures that drug safety is prioritized for high-risk patients.
[0042] The efficacy prediction value E_i, side effect prediction value S_i, and medication cost value C_i of the i-th candidate drug treatment regimen are extracted from the candidate treatment regimen set. The efficacy prediction value is obtained by calculating the patient's symptom improvement rate through a time-dependent relationship model. The side effect prediction value is quantified based on the drug interaction matrix and the patient's individual metabolic capacity parameters. The medication cost value is calculated by summing the unit price of each drug and the total cost over the treatment cycle. To eliminate differences in the dimensions of different indicators, the efficacy prediction value is normalized and mapped to the interval between 0 and 1. The side effect prediction value is normalized after taking the reciprocal, so that a larger value represents a smaller side effect. The medication cost value is also normalized.
[0043] Perform a weighted summation operation, and the comprehensive evaluation score Q_i of the i-th candidate solution is obtained through the formula... The calculated values are: E_i', S_i', and C_i', where E_i', S_i', and C_i' are the normalized efficacy value, side effect value, and cost value, respectively. All candidate schemes in the candidate drug control scheme set are traversed, and the comprehensive evaluation score of each scheme is calculated. The calculation results are then stored in the scoring vector.
[0044] A quicksort algorithm is used to sort the score vectors in descending order of their comprehensive evaluation scores, resulting in a ranking of candidate solutions. The candidate solution with the highest comprehensive evaluation score is extracted from the ranking results; this solution includes elements such as the specific drug name, dosage, dosing frequency, and treatment cycle. When the score difference between the highest-scoring solution and the second-highest-scoring solution is less than a preset threshold of 0.05, a clinical physician decision support interface is introduced, simultaneously pushing both solutions to the clinical end for the physician's final decision. The optimal medication control solution is output to the execution module, while simultaneously recording the solution generation timestamp and the patient's risk level identifier, providing a traceability basis for subsequent feedback iterations.
[0045] In one optional implementation, medication adjustments are performed based on the optimal medication control plan, and feedback information is generated by collecting postoperative actual efficacy data and adverse reaction data, including: Based on the drug type, dosage and administration time information in the optimal medication control plan, a medication execution instruction is generated and sent to the medication execution terminal to complete the medication adjustment. At the same time, the execution timestamp and execution status identifier of the medication execution instruction are recorded. During the preset monitoring period after the medication execution instruction is executed, physiological indicator change data of the patient is collected by physiological monitoring equipment as postoperative efficacy data, and adverse reaction symptom description and severity identification of the patient are obtained through the adverse reaction collection interface as adverse reaction data. The execution timestamp, execution status identifier, postoperative actual efficacy data, and adverse reaction data are associated and bound to generate feedback information containing medication execution information and actual response information.
[0046] After determining the optimal medication control plan, it is necessary to translate the plan into executable medication operations and monitor patient responses in real time. Specifically, the drug type field, dosage value, and administration time parameter in the optimal medication control plan are parsed and converted into structured medication execution instructions according to a standardized medication execution protocol. This instruction includes elements such as drug identification code, dosage unit, route of administration, and execution time, forming an instruction message that conforms to medical information exchange standards. The generated medication execution instructions are sent to the medication execution terminal via an encrypted communication channel. This terminal can be an intravenous infusion control device, a smart infusion pump, or a pharmacy dispensing system. At the moment the instruction is sent, the execution timestamp generated by the system clock is automatically recorded, accurate to the second; simultaneously, an execution status identifier field is set, initially marked as "pending execution," updated to "in execution" when the medication execution terminal confirms receipt and begins execution, and marked as "completed" upon completion.
[0047] After medication administration, the postoperative monitoring process is initiated. The preset monitoring period is based on the drug's half-life and clinical routines, typically covering 2 to 72 hours after medication. During this period, physiological indicator changes are continuously acquired through a data interface established with the patient's vital signs monitoring equipment. The monitoring equipment includes an electrocardiogram monitor, blood pressure monitor, body temperature sensor, and urine output monitoring device, with the data collection frequency set every 15 minutes or adjusted as required by the physician. The collected physiological indicators include key parameters such as heart rate, blood pressure, body temperature, urine output, and blood oxygen saturation. Each data collection is accompanied by a collection time label, forming a time-series physiological indicator sequence, which serves as an objective basis for assessing the actual postoperative efficacy. Simultaneously, an adverse reaction collection interface is established, integrating a module for medical staff input and a module for patient self-reporting. Medical staff describe adverse reaction symptoms such as nausea, dizziness, and rash observed by patients during ward rounds using a standardized symptom classification coding system; patients self-report discomfort symptoms through a mobile application. For each adverse reaction record, severity is marked according to clinical grading standards, and divided into three levels: mild, moderate, and severe. Mild means that the symptoms are mild and do not affect daily activities; moderate means that medical intervention is required but there is no danger to life; and severe means that life is endangered and emergency treatment is required.
[0048] After data collection, an association and binding operation is performed. Using the execution timestamp of the medication administration instruction as the anchor point, and the timestamp and its corresponding execution status identifier as the index key, all postoperative actual efficacy data and adverse reaction data collected within the preset monitoring period are associated. The data structure consists of two parts: a medication administration information segment and an actual response information segment. The medication administration information segment stores fields such as drug type, dosage, timestamp, and execution status; the actual response information segment stores a sequence of physiological indicators sorted by time, a list of adverse reaction symptoms, and an array of severity identifiers. A timestamp matching mechanism binds the two segments of information into a unified feedback information record, ensuring that each medication adjustment forms a complete causal traceability chain with its resulting physiological response and adverse reaction. The generated feedback information is stored in a structured data format in the feedback database, providing training samples for subsequent parameter iteration updates.
[0049] A second aspect of the present invention provides an artificial intelligence-based system for precise control of postoperative bladder medication, comprising: The data acquisition unit is used to acquire the patient's postoperative physiological index data, medication history data, and clinical symptom data; based on multimodal feature fusion, the physiological index data, medication history data, and clinical symptom data are used to extract features and perform semantic mapping to obtain the patient's individualized state representation; The scheme generation unit is used to calculate the pharmacokinetic parameters and efficacy response curves of different medication regimens based on the individualized state characterization of the patient through time-series dependency modeling, and to generate a set of candidate medication regulation schemes in combination with drug interaction constraint rules. The scheme optimization unit is used to solve the optimal drug control scheme based on the set of candidate drug control schemes and postoperative recovery goals, under the constraints of maximizing efficacy, minimizing side effects, and drug cost, and dynamically adjust the weight coefficients of each optimization objective based on the risk characteristics in the individualized state representation of the patient. The feedback update unit is used to adjust the medication based on the optimal medication control plan, collect postoperative actual efficacy data and adverse reaction data to form feedback information, and use the feedback information to iteratively update the drug efficacy prediction parameters and weight coefficients.
[0050] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.
[0051] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.
[0052] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.
[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for precise regulation of postoperative bladder medication based on artificial intelligence, characterized in that, include: Obtain postoperative physiological indicators, medication history, and clinical symptom data from patients; Based on multimodal feature fusion, feature extraction and semantic mapping are performed on the physiological indicator data, the medication history data and the clinical symptom data to obtain a personalized patient status representation; Based on the individualized patient status characterization, pharmacokinetic parameters and efficacy response curves under different medication regimens are calculated through time-dependent modeling, and a set of candidate medication regulation regimens is generated by combining drug interaction constraint rules. Based on the set of candidate drug control schemes and postoperative recovery goals, the optimal drug control scheme is solved under the constraints of maximizing efficacy, minimizing side effects, and drug cost. The weight coefficients of each optimization goal are dynamically adjusted based on the risk characteristics in the individualized patient status representation. Medication adjustments are made based on the optimal medication control plan, and postoperative actual efficacy data and adverse reaction data are collected to form feedback information. The feedback information is then used to iteratively update the drug efficacy prediction parameters and weighting coefficients.
2. The method according to claim 1, characterized in that, Based on multimodal feature fusion, feature extraction and semantic mapping are performed on the physiological indicator data, the medication history data, and the clinical symptom data to obtain a patient's individualized status representation, including: Temporal features were extracted from physiological indicator data to obtain physiological indicator temporal feature vectors; drug effect correlation analysis was performed on medication history data to obtain medication history correlation feature vectors; and symptom severity was quantified from clinical symptom data to obtain clinical symptom quantitative feature vectors. The time-series feature vectors of the physiological indicators, the medication history-related feature vectors, and the clinical symptom quantification feature vectors are cross-modal semantically aligned. The association weights between each modality feature are calculated through an attention weighting mechanism. Based on the association weights, each modality feature is adaptively fused to obtain a fused feature vector. The fused feature vector is then subjected to a nonlinear mapping transformation to obtain a patient-specific state representation.
3. The method according to claim 1, characterized in that, Based on the individualized patient status characterization, pharmacokinetic parameters and efficacy response curves under different medication regimens are calculated through time-dependent modeling. A set of candidate medication regulation regimens is then generated by combining drug interaction constraint rules, including: Based on the individualized state representation of the patient, a patient-specific baseline state for predicting drug efficacy is constructed. Based on the baseline state for predicting drug efficacy and historical medication time series data, the residual influence coefficient of previous medication on the current drug efficacy is calculated by recursive time series dependency modeling, and the time series residual influence matrix is obtained. The time-series residual effect matrix is coupled with the medication regimen to be evaluated to predict the pharmacokinetic parameter change trajectory and efficacy response curve under the medication regimen to be evaluated; Obtain a drug interaction constraint rule base, and perform safety constraint screening on the efficacy response curve based on the drug interaction constraint rule base to eliminate medication regimens that violate drug interaction constraints; The medication regimens selected through the aforementioned safety constraints are combined and arranged to generate a set of candidate medication control regimens.
4. The method according to claim 3, characterized in that, The time-series residual effect matrix is coupled with the medication regimen to be evaluated to predict the pharmacokinetic parameter change trajectory and efficacy response curve under the medication regimen to be evaluated, including: The drug dosage and administration time information in the medication regimen to be evaluated are converted into a time-series input vector, and the time-series input vector is coupled with the time-series residual effect matrix by a time-by-time weighted operation to obtain a modified metabolic parameter sequence containing residual effects. Based on the metabolic parameter values at each time point in the modified metabolic parameter sequence, the drug absorption rate, distribution rate, and elimination rate at the corresponding time point are calculated, and the drug absorption rate, distribution rate, and elimination rate at consecutive time points are combined to form a trajectory of changes in drug metabolism kinetic parameters. Based on the drug absorption rate and elimination rate at each moment in the trajectory of the changes in the pharmacokinetic parameters, the drug blood concentration value at each moment is calculated, and the drug blood concentration values at each moment are arranged in time sequence to form a blood concentration time series curve. The blood concentration time series curve is mapped to the therapeutic effect space to obtain the therapeutic effect response curve.
5. The method according to claim 1, characterized in that, Based on the set of candidate drug control schemes and postoperative recovery goals, the optimal drug control scheme is determined under the constraints of maximizing efficacy, minimizing side effects, and controlling medication costs. Based on the risk characteristics in the individualized patient status representation, the patient risk level is identified, and adaptive weight coefficients are assigned to the efficacy maximization target, side effect minimization target, and medication cost constraint target based on the patient risk level, so as to obtain the patient-specific target weight configuration. The patient-specific target weight configuration is weighted and summed with the efficacy prediction value, side effect prediction value, and medication cost value of each candidate scheme in the candidate medication control scheme set to obtain the comprehensive evaluation score of each candidate scheme. The candidate schemes in the candidate drug control scheme set are ranked according to the comprehensive evaluation score, and the candidate scheme with the highest comprehensive evaluation score is selected as the optimal drug control scheme from the ranking results.
6. The method according to claim 1, characterized in that, Medication adjustments are made based on the optimal medication control plan, and feedback information is generated by collecting postoperative efficacy data and adverse reaction data, including: Based on the drug type, dosage and administration time information in the optimal medication control plan, a medication execution instruction is generated and sent to the medication execution terminal to complete the medication adjustment. At the same time, the execution timestamp and execution status identifier of the medication execution instruction are recorded. During the preset monitoring period after the medication execution instruction is executed, physiological indicator change data of the patient is collected by physiological monitoring equipment as postoperative efficacy data, and adverse reaction symptom description and severity identification of the patient are obtained through the adverse reaction collection interface as adverse reaction data. The execution timestamp, execution status identifier, postoperative actual efficacy data, and adverse reaction data are associated and bound to generate feedback information containing medication execution information and actual response information.
7. An artificial intelligence-based precision control system for post-bladder surgery medication, used to implement the method of any one of claims 1-6, characterized in that, include: The data acquisition unit is used to acquire the patient's postoperative physiological index data, medication history data, and clinical symptom data; based on multimodal feature fusion, the physiological index data, medication history data, and clinical symptom data are used to extract features and perform semantic mapping to obtain the patient's individualized state representation; The scheme generation unit is used to calculate the pharmacokinetic parameters and efficacy response curves of different medication regimens based on the individualized state characterization of the patient through time-series dependency modeling, and to generate a set of candidate medication regulation schemes in combination with drug interaction constraint rules. The scheme optimization unit is used to solve the optimal drug control scheme based on the set of candidate drug control schemes and postoperative recovery goals, under the constraints of maximizing efficacy, minimizing side effects, and drug cost, and dynamically adjust the weight coefficients of each optimization objective based on the risk characteristics in the individualized state representation of the patient. The feedback update unit is used to adjust the medication based on the optimal medication control plan, collect postoperative actual efficacy data and adverse reaction data to form feedback information, and use the feedback information to iteratively update the drug efficacy prediction parameters and weight coefficients.
8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.
9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.