A medication recommendation system for procedural sedation in children based on multi-objective optimization

Abstract

The present invention relates to the field of medical artificial intelligence technology, and discloses a pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization, comprising: a data acquisition and preprocessing module, which is used to acquire multi-source heterogeneous clinical data and perform data cleaning and feature extraction to generate a multi-source heterogeneous clinical feature set; a multi-objective optimization decision module, which is used to establish a number of objective functions, and adopt an improved multi-objective optimization method to optimize and solve each objective function to obtain an optimal medication regimen; a medication regimen module, which is used to automatically generate a specific medication regimen based on a scheme generation method with time constraints according to the optimal medication regimen; a safety inspection module, which is used to perform a multi-level safety review on a specific medication regimen, and obtain a multi-level safety review result by calculating a feasibility judgment function through risk assessment; the present invention improves the scientificity and reliability of the medication regimen, so that each patient child can obtain a sedation regimen that best suits his or her individual characteristics.

Description

Technical Field

[0001] The present invention relates to the field of medical artificial intelligence technology, and in particular to a pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization. Background Art

[0002] Procedural sedation in children is an important part of clinical diagnosis and treatment, and is of great significance for improving the success rate of examinations and operations and alleviating the pain of children. However, procedural sedation in children still faces many challenges in clinical practice.

[0003] In the existing technology, the formulation of procedural sedation medication regimens for children mainly relies on the personal experience of doctors and the general recommendations of clinical guidelines. This method has the following shortcomings: first, due to the large age span and significant individual differences of children, standardized medication regimens are difficult to meet personalized needs; second, the combined use of multiple sedatives involves complex drug interactions, and traditional experience is difficult to accurately evaluate their comprehensive effects; third, the existing regimens are difficult to take into account multiple goals such as sedation effect, safety and economy at the same time. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention provides a pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization. By comprehensively considering individual differences, multi-objective balance, safety assurance and other factors, an intelligent medication regimen is formulated to solve the problems of lack of clinical data, single objective and insufficient consideration of safety assurance in the pediatric procedural sedation medication regimen in the prior art, thereby improving the standardization and personalization of pediatric procedural sedation medication.

[0005] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0006] A pediatric procedural sedation medication recommendation system based on multi-objective optimization, including:

[0007] The data collection and preprocessing module is used to collect multi-source heterogeneous clinical data, clean the multi-source heterogeneous clinical data, extract and fuse the cleaned multi-source heterogeneous clinical data, and generate a multi-source heterogeneous clinical feature set;

[0008] The multi-objective optimization decision-making module is used to define decision variables and objective constraints based on multi-source heterogeneous clinical feature sets, establish sedation effect objective function, safety objective function, and economic objective function, and use the improved multi-objective optimization method to optimize and solve each objective function to obtain the optimal medication plan;

[0009] Among them, the best medication regimen is the one that can achieve success in one medication and does not excessively increase the cost for sedative medication for children under the condition of safety.

[0010] The medication scheme module is used to automatically generate a specific medication scheme based on the optimal medication scheme and the scheme generation method based on the timing constraints;

[0011] Among them, the specific medication regimen includes the type of drug and the dosage of the drug;

[0012] The safety inspection module is used to conduct multi-level safety audits on specific medication plans. Through risk assessment, medication dosage constraints, medication time interval constraints, and combined medication constraints, it calculates the feasibility judgment function and obtains multi-level safety audit results.

[0013] Furthermore, the data collection and preprocessing module includes a raw data collection module, a data cleaning module, and a feature engineering module;

[0014] The raw data acquisition module is used to collect multi-source heterogeneous clinical data;

[0015] The multi-source heterogeneous clinical data include basic information of the patient, clinical test data, electronic medical record text, previous medication records, and surgical operation related information; the basic information of the patient includes the patient's age, gender, weight, fasting status, and body temperature;

[0016] The data cleaning module is used to detect outliers, interpolate missing values, structure unstructured text, and convert data formats uniformly for multi-source heterogeneous clinical data, thus obtaining cleaned multi-source heterogeneous clinical data.

[0017] The feature engineering module is used to extract key features, select and reduce dimensions, and combine and interact features of multi-source heterogeneous clinical data after data cleaning to generate a feature set of multi-source heterogeneous clinical data.

[0018] Furthermore, the structured processing of unstructured text adopts bidirectional long short-term memory network and conditional random field model processing to unify the structure of unstructured text in multi-source heterogeneous clinical data and generate structured text of multi-source heterogeneous clinical data. The specific process is as follows:

[0019] The unstructured text in the multi-source heterogeneous clinical data is input into the input layer and word embedding layer of the bidirectional long short-term memory network and the conditional random field model in turn to generate a word vector matrix, namely:

[0020]

[0021] Where n represents the length of the unstructured text input sequence in the multi-source heterogeneous clinical data, X represents the unstructured text input sequence in the multi-source heterogeneous clinical data, X1 represents the unstructured text input at the first moment, X2 represents the unstructured text input at the second moment, and X t represents the unstructured text input at time t, X1 Indicates the initial status information of the consultation, X 2 Indicates medication preparation information, X 3 Indicates medication regimen information, X 4 Indicates the effective date of medication, X 5 Indicates the end of medication operation, X 6 represents the medication recovery period information, E represents the word vector matrix, e1 represents the d-dimensional vector of the first word, and e2 represents the d-dimensional vector of the second word;

[0022] Among them, the initial status information of the consultation includes the basic information of the child, previous medication records, and allergy history records; the medication preparation information includes the pre-sedation assessment results, examination and operation types, and the planned sedation depth; the medication regimen information includes the first medication dose, medication method, and baseline values ​​of monitoring indicators, and the monitoring indicators are blood oxygen saturation, pulse rate, and respiratory rate; the medication effective period information includes changes in vital signs, sedation depth scores, and adverse reaction records; the medication operation end information includes the completion of the operation, the cumulative medication dose, and the sedation effect evaluation status; the medication recovery period information includes the recovery time, recovery status, and follow-up records;

[0023] The word vector matrix is ​​input into the bidirectional long short-term memory network layer, and the context information of the word vector matrix is ​​obtained and input into the hidden layer to obtain the hidden state of the word vector matrix, that is:

[0024]

[0025] Among them, LSTM represents long short-term memory network, h t represents the hidden state at time t, represents the hidden state of the forward LSTM network at time t, that is, the state evolution information from the admission of the patient to the current moment, represents the hidden state of the forward long short-term memory network at time t-1, represents the hidden state of the LSTM network after the tth moment, that is, the state information from the expected awakening to the current moment, represents the hidden state of the backward long short-term memory network at time t-1, e t Represents the word vector at time t;

[0026] The hidden state of the word vector matrix is ​​input into the conditional random field layer to generate a structured text sequence of multi-source heterogeneous clinical data, namely:

[0027]

[0028] Among them, Y represents the annotation sequence, that is, the optimized structured text sequence, Y1 represents the annotation label of the first word, Y2 represents the annotation label of the second word, and Y τrepresents the annotation label of the τth word, P(Y|X) represents the conditional probability distribution of the structured text sequence, exp represents the exponential function, score(X,Y) represents the clinical rationality of the structured text sequence, including the appropriateness of the medication regimen, the correctness of the temporal logic, and the degree of compliance with medical regulations, and Z(X) represents the normalization factor, that is, the overall space of all possible medication regimens under the current clinical condition.

[0029] Furthermore, feature selection uses the least absolute shrinkage and selection operator regression method based on L1 regularization to screen key features and generate the screened key features, namely:

[0030]

[0031] Among them, min γ represents the minimum value, n′ represents the number of samples of multi-source heterogeneous clinical data, z represents the target variable of the key feature, F represents the key feature matrix, γ represents the key feature weight vector, λ represents the regularization parameter, ‖·‖1 and ‖·‖2 represent the L1 norm and L2 norm respectively;

[0032] Among them, key characteristics include age, weight, ASA rating, past medical history, and examination type.

[0033] Furthermore, a deep cross network is introduced for feature combination and interaction to generate a multi-source heterogeneous clinical data feature set. The specific process is as follows:

[0034] The reduced-dimensional feature vectors generated after feature selection and dimensionality reduction are sequentially input into several cross layers of the cross network for feature interaction, and the cross features are output, namely:

[0035]

[0036] Among them, f l+1 represents the cross feature of the l+1th layer, f0 represents the dimension reduction feature vector, T represents the transpose, w1 represents the weight vector of the first cross layer, o1 represents the bias vector of the first cross layer, and f1 represents the cross feature of the first layer;

[0037] The cross features are input into the fully connected network of the deep network for deep feature extraction and fusion to generate a multi-source heterogeneous clinical data feature set.

[0038] Furthermore, the decision variables and target constraints are as follows:

[0039] Decision variables include drug type, dosage, and administration time; target constraints include dosage limits and contraindications.

[0040] Furthermore, the sedation effect objective function, safety objective function, and economic objective function are:

[0041]

[0042] Among them, q1 represents the sedation effect index, f1(q1) represents the sedation effect objective function, and eff i (q1) represents the score of the i-th sedation effect index q1, ω i represents the weight of the ith sedation effect index q1, q2 represents the safety index, f2(q2) represents the safety objective function, and risk j (q2) represents the risk probability of the jth security indicator q2, serverity j represents the risk severity of the j-th safety indicator q2, q3 represents the economic indicator, f3(q3) represents the economic objective function, c k represents the unit cost of the k-th drug, dose k (q3) represents the consumption of the kth drug.

[0043] Furthermore, when the improved multi-objective optimization method is used to optimize and solve each objective function, the particle swarm optimization method is introduced to adaptively and dynamically adjust the weights of the sedation effect objective function, the safety objective function, and the economic objective function, and by updating the position of each particle, the optimal weights of the sedation effect objective function, the safety objective function, and the economic objective function are obtained to obtain the optimal medication plan;

[0044] Among them, the position of each particle is updated as follows:

[0045]

[0046] in, represents the velocity of particle s at time t+1, represents the velocity of particle s at time t, represents the inertia weight, π1 and π2 represent learning factors, r1 and r2 represent random numbers, and pbest s represents the historical optimal position of particle s, gbest represents the global optimal position of the particle swarm, represents the position of particle s at time t, represents the position of particle s at time t+1.

[0047] Furthermore, according to the optimal medication regimen, the specific process of automatically generating a specific medication regimen based on the scheme generation method of timing constraints is as follows:

[0048] Based on the optimal medication regimen, obtain the drug combination set, namely:

[0049]

[0050] Among them, M represents the drug combination set, m1 represents the first drug, m2 represents the second drug, and m a represents the ath drug, τ a represents the onset time of drug a, δ a Indicates the duration of action of drug a, α a represents the peak action time of drug a, β a represents the elimination half-life of drug a;

[0051] The drug combination is centrally arranged in a time sequence, namely:

[0052] T1={(m a ,t a )|m a ∈M,t a ∈R +}

[0053] Among them, T1 represents the medication sequence of the drugs in the drug combination, t a Indicates the time of taking drug a, R + represents a positive real number;

[0054] According to the medication sequence of the drugs in the drug combination set, the drug sequence minimization objective function is established, that is:

[0055]

[0056] Among them, minJ(T1) represents the drug time sequence minimization objective function, θ a represents the weight coefficient of the objective function of minimizing the time series of drug a, E a represents the optimization target of medication timing of drug a, η represents the regularization coefficient, R ′ represents the timing regularization term;

[0057] Establish the constraints of the drug time sequence minimization objective function, namely:

[0058] The drug interval constraints are:

[0059]

[0060] The action superposition constraint is:

[0061] C(t)=∑ a c ′ a (tt a )≤C max ;

[0062] The operation timing constraints are:

[0063]

[0064] Among them, t b represents the duration of drug b administration, Δτ ab Indicates the drug interval threshold, m b represents the bth drug, C(t) represents the overall drug intensity at the tth moment, c ′ a represents the time effect function of the ath drug, C max Indicates the maximum allowable superposition strength, t start Indicates the start time of the medication regimen, T op Indicates the operation start time, d op Indicates the duration of the operation;

[0065] Based on the constraints of the objective function of minimizing the drug sequence, the drug sequence of the drug combination set is solved to generate a specific medication plan, that is, the drug types and drug dosages under the medication sequence.

[0066] Furthermore, a multi-level safety review is conducted on the specific medication regimen. Through risk assessment, medication dosage constraints, medication time interval constraints, and combined medication constraints, the feasibility judgment function is calculated. The specific process of obtaining the multi-level safety review results is as follows:

[0067] For each specific medication regimen, a risk assessment based on fuzzy rules is used to calculate the risk of the specific medication regimen, namely:

[0068]

[0069] Among them, R1(x) represents the risk of specific medication regimen x, μ z′ (x) represents the membership function of the z′th rule of the specific medication regimen x, w1 z′ represents the weight of the z′th rule of the specific medication regimen X, μ(x) represents the membership function of the specific medication regimen X, C1 represents the optimal value defined by the expert, i.e., the central value, and σ represents the acceptable deviation range, i.e., the standard deviation;

[0070] Establish dosing constraints, namely:

[0071] G1(x)=d(X)-D max ≤0

[0072] Among them, g1(x) represents the dosage constraint of the specific medication regimen x, D(x) represents the drug dosage vector of the specific medication regimen x, and D max The vector representing the maximum allowable dose of the drug for a specific medication regimen x;

[0073] Establish medication interval constraints, namely:

[0074] g2(x)=|T1a1 -T1 b1 |-I min ≥0

[0075] Among them, g2(x) represents the medication time interval constraint of the specific medication regimen x, T1 a1 Indicates the administration time point of drug a1 in the specific medication regimen x, T1 b1 Indicates the administration time point of drug b1 in the specific medication regimen x, I min Indicates the minimum safe dosing interval;

[0076] Establish constraints on combination medication, namely:

[0077] g3(x)=S int (x)-S th ≤0;

[0078] Among them, g3(x) represents the combined medication constraints of the specific medication regimen x, S int (x) represents the drug interaction risk score function for a specific medication regimen x, S th represents the drug interaction risk threshold;

[0079] Calculate the feasibility judgment function and obtain the multi-level security audit results, namely:

[0080] F′(x)=∧(g d′ (x)≤0),d′=1,2,3

[0081] Among them, F ′ (x) represents the feasibility result of the specific medication regimen x, that is, the multi-level safety review result, ∧ represents the judgment function, d ′ represents the specific constraint quantity, where when F ′ When (x) is 1, the current specific medication regimen x is feasible; when F ′ When (x) is 0, the current specific medication regimen x is not feasible.

[0082] The present invention has the following beneficial effects:

[0083] 1. The proposed system for recommending medication regimens for procedural sedation in children based on multi-objective optimization realizes accurate medication recommendations based on individual characteristics of children. It not only considers basic characteristics of children such as age, weight, and medical history, but also dynamically adjusts the regimen according to specific types of surgery or operations, and supports customized personalized regimens under special circumstances. This data-driven personalized recommendation method significantly reduces the subjectivity of traditional empirical medication and improves the scientificity and reliability of medication regimens. At the same time, by realizing refined management of treatment regimens, the system enables each child to obtain the sedation regimen that best suits their individual characteristics.

[0084] 2. Effectively reduce the risk of adverse reactions, that is, by establishing a comprehensive safety early warning mechanism, ensuring medication safety from multiple levels, and being able to monitor drug interaction risks in real time, automatically identify potential medication contraindications, and provide multi-level safety measures; this intelligent risk management method significantly improves medication safety, not only reducing the probability of medication errors, but also reducing the incidence of adverse reactions; especially in emergency situations, the system's rapid response and plan recommendation functions greatly improve the response capabilities of medical staff and provide strong protection for the safety of children;

[0085] 3. Significantly improve the efficiency of medical decision-making, that is, through intelligent decision support, optimize the clinical decision-making process, generate personalized medication plans in a short time, significantly shorten the plan formulation time, and reduce the workload of medical staff; at the same time, the standardized decision-making process provided by the system improves the standardization and consistency of clinical decision-making; and in terms of medical resource utilization, by optimizing the drug use strategy, improving the consistency of the diagnosis and treatment process, reducing unnecessary medical waste, and achieving the unity of economy and effectiveness;

[0086] 4. It has unique advantages in data management and knowledge accumulation, namely, it accumulates a large amount of valuable clinical data through standardized data collection and processing processes; these data not only support the continuous optimization of the system, but also provide valuable data resources for clinical research; at the same time, the system's learning ability enables it to continuously draw experience from new clinical cases, optimize recommendation strategies, and achieve continuous performance improvement. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 This is a flow chart of a pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization proposed by the present invention;

[0088] Figure 2 It is a simplified structural diagram of the bidirectional long short-term memory network and conditional random field model;

[0089] Figure 3 A simplified structural diagram of a deep cross network. DETAILED DESCRIPTION

[0090] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0091] like Figure 1As shown in the figure, a pediatric procedural sedation medication recommendation system based on multi-objective optimization includes:

[0092] The data collection and preprocessing module is used to collect multi-source heterogeneous clinical data, clean the multi-source heterogeneous clinical data, extract and fuse the features of the cleaned multi-source heterogeneous clinical data, and generate a multi-source heterogeneous clinical feature set.

[0093] In this embodiment, since pediatric clinical data has the characteristics of diverse sources and heterogeneous formats, including structured test data and unstructured clinical text records; these data are often scattered in different information systems, and there are problems of uneven quality and inconsistent standards; especially when processing clinical records written by doctors, how to accurately extract key information related to sedation is an important technical challenge. In addition, the common problems of missing values ​​and outliers in clinical data also bring difficulties to the effective use of data; therefore, in this embodiment, by performing data cleaning and feature engineering operations on the collected multi-source heterogeneous clinical data, these complex data are effectively integrated and processed to achieve the recommendation of procedural sedation medication regimens for children, that is, to achieve precision medication. In addition, in the data collection and preprocessing module, the system adopts a dual-track data collection mechanism: on the one hand, clinical specialty information is collected through a registered sedation data collection system; on the other hand, basic medical record information is automatically obtained from the hospital information system (HIS) through a standardized data interface; clinical specialty information includes fasting status, ASA rating, vital signs, previous sedation history, history of anesthesia-related adverse reactions, and special conditions (such as airway abnormalities, cardiopulmonary function status, etc.); the system adopts a structured form design to standardize the data entry format, and is equipped with an intelligent reminder function to promptly prompt the lack or abnormality of key information; in the data preprocessing stage, the system performs data standardization, including unit unification and outlier processing, and performs medical entity recognition and relationship extraction on unstructured text; in the feature engineering stage, the system screens key features through algorithms such as minimum absolute shrinkage and selection operator regression (Lasso regression) and random forest, and uses principal component analysis (PCA) to discover potential correlations between features, while focusing on maintaining the interpretability of features to ensure that doctors can understand and verify the results.

[0094] Specifically, the data collection and preprocessing module includes a raw data collection module, a data cleaning module, and a feature engineering module;

[0095] The raw data acquisition module is used to collect multi-source heterogeneous clinical data; the multi-source heterogeneous clinical data includes the basic information of the patient child, clinical test data, electronic medical record text, previous medication records and surgical operation related information; the basic information of the patient child includes the patient child's age, gender, weight, fasting status and body temperature.

[0096] The data cleaning module is used to perform outlier detection, missing value interpolation, structured processing of unstructured text, and unified conversion of data formats on multi-source heterogeneous clinical data to obtain multi-source heterogeneous clinical data after data cleaning.

[0097] In this embodiment, the extraction of clinical text information is the key basis for the system to implement personalized medication plans; in actual clinical work, a large amount of valuable patient information exists in the form of unstructured text in medical documents such as electronic medical records, surgical records, and nursing records; at the same time, these text data contain important information such as patients' previous medication reactions, special situation records, and treatment effect evaluations, which are of great guiding significance for formulating accurate sedation plans. Therefore, in view of the unique professionalism and complexity of medical texts, the present invention designs named entity recognition (NER) and relationship extraction technology based on deep learning, and selects a bidirectional long short-term memory network and a conditional random field model (BiLSTM-CRF) as the core algorithm, mainly based on the following considerations: First, entities in medical texts (such as disease names, drug names, symptom descriptions, etc.) are usually composed of multiple phrases, and there are a large number of abbreviations and synonymous expressions. BiLSTM can effectively capture these long-distance semantic dependencies; secondly, the recognition of medical entities needs to consider contextual information, and the bidirectional structure can make full use of the contextual information for judgment; finally, the introduction of the CRF layer can learn the transfer rules between labels to ensure that the output annotation sequence meets medical standards; among them, the simplified structure of the bidirectional long short-term memory network and the conditional random field model is as follows: Figure 2 As shown in the figure, it includes input layer, word embedding layer, bidirectional long short-term memory network layer (BiLSTM), hidden state layer and conditional random field layer (CRF layer). The specific operation process is as follows:

[0098] The unstructured text in the multi-source heterogeneous clinical data is input into the input layer and word embedding layer of the bidirectional long short-term memory network and the conditional random field model in turn to generate a word vector matrix, namely:

[0099] Where n represents the length of the unstructured text input sequence in the multi-source heterogeneous clinical data, X represents the unstructured text input sequence in the multi-source heterogeneous clinical data, X1 represents the unstructured text input at the first moment, X2 represents the unstructured text input at the second moment, and X t represents the unstructured text input at time t, X 1 Indicates the initial status information of the consultation, X 2 Indicates medication preparation information, X 3 Indicates medication regimen information, X 4 Indicates the effective date of medication, X 5 Indicates the end of medication operation, X 6represents the medication recovery period information, E represents the word vector matrix, e1 represents the d-dimensional vector of the first word, and e2 represents the d-dimensional vector of the second word.

[0100] Among them, the initial status information of the consultation includes the basic information of the patient child, previous medication records, and allergy history records; the medication preparation information includes the pre-sedation assessment results, examination and operation types, and the planned sedation depth; the medication plan information includes the first medication dose, medication method, and baseline values ​​of monitoring indicators. The monitoring indicators are blood oxygen saturation, pulse rate, and respiratory rate; the medication effective period information includes changes in vital signs, sedation depth scores, and adverse reaction records; the medication operation end information includes the completion of the operation, the cumulative medication dose, and the sedation effect evaluation status; the medication awakening period information includes the awakening time, recovery status, and follow-up records.

[0101] The word vector matrix is ​​input into the bidirectional long short-term memory network layer, and the context information of the word vector matrix is ​​obtained and input into the hidden layer to obtain the hidden state of the word vector matrix, that is:

[0102]

[0103] Among them, LSTM represents long short-term memory network, h t represents the hidden state at time t, represents the hidden state of the forward LSTM network at time t, that is, the state evolution information from the admission of the patient to the current moment, represents the hidden state of the forward long short-term memory network at time t-1, represents the hidden state of the LSTM network after the tth moment, that is, the state information from the expected awakening to the current moment, represents the hidden state of the backward long short-term memory network at time t-1, e t Represents the word vector at the tth moment.

[0104] The hidden state of the word vector matrix is ​​input into the conditional random field layer to generate a structured text sequence of multi-source heterogeneous clinical data, namely:

[0105]

[0106] Among them, Y represents the annotation sequence, that is, the optimized structured text sequence, Y1 represents the annotation label of the first word, Y2 represents the annotation label of the second word, and Y τ represents the annotation label of the τth word, P(Y|X) represents the conditional probability distribution of the structured text sequence, exp represents the exponential function, score(X,Y) represents the clinical rationality of the structured text sequence, including the appropriateness of the medication regimen, the correctness of the temporal logic, and the degree of compliance with medical regulations, and Z(X) represents the normalization factor, that is, the overall space of all possible medication regimens under the current clinical condition.

[0107] In addition, the output structured text includes information such as drug name, symptom name and dosage; at the same time, the normalization factor Z(X) also helps to balance the risks and benefits of different schemes based on the constraints of clinical practice. In summary, the structured processing of unstructured text adopts bidirectional long short-term memory network and conditional random field model. Compared with the traditional rule-based method, this method has a higher recognition accuracy for medical text, especially in the processing of non-standard expressions. It shows strong robustness; secondly, it can identify the implicit medical relationship in the text, such as the association between symptoms and medication reactions, the interaction between different drugs, etc. This information is of great value for the subsequent optimization of medication regimens; finally, the processing speed of this method can meet the clinical real-time requirements. At the same time, it can also extract key information related to sedation from complex clinical texts, providing reliable data support for the subsequent personalized program formulation. For example, the system can automatically identify the adverse reactions that occurred in the patient's previous sedation process, and convert this information into quantitative risk assessment indicators to assist doctors in formulating safer sedation programs.

[0108] The feature engineering module is used to extract key features, select and reduce dimensions, and combine and interact features of multi-source heterogeneous clinical data after data cleaning to generate a feature set of multi-source heterogeneous clinical data.

[0109] In this embodiment, feature engineering can play the role of a bridge between raw data and model algorithms in the system. In clinical practice, the factors that affect the formulation of sedation plans are numerous and complex, including the patient's basic information (age, weight, etc.), clinical indicators (vital signs, laboratory tests, etc.), past medical history, etc. How to effectively integrate and utilize these heterogeneous data directly affects the accuracy and reliability of the final plan. Among them, this embodiment adopts a two-stage feature engineering strategy to solve the problem of feature selection and feature combination respectively.

[0110] In the feature selection stage (feature selection and dimensionality reduction), the Lasso regression algorithm based on L1 regularization is used. The main considerations for choosing this algorithm are: first, medical data are often highly correlated. Lasso regression can automatically achieve sparse selection of features and screen out truly important features by introducing the L1 norm penalty term; second, the algorithm has good interpretability and can quantitatively evaluate the importance of each feature, which is crucial to the credibility of the medical decision-making system. Specifically:

[0111] Feature selection uses the least absolute shrinkage and selection operator regression method based on L1 regularization to screen key features and generate the selected key features, namely:

[0112]

[0113] Among them, min γ represents the minimum value, n′ represents the number of samples of multi-source heterogeneous clinical data, z is the target variable of the key feature, F represents the key feature matrix, γ represents the key feature weight vector, λ represents the regularization parameter, ‖·‖1 and ‖·‖2 represent the L1 norm and L2 norm, respectively; the key features include age, weight, ASA rating, past medical history, and examination type.

[0114] In this embodiment, F∈R n′×k′ ,γ∈R k′ , k ′ Represents the key feature dimension, λ>0, and is used to control feature sparsity.

[0115] Specifically, a deep cross network is introduced for feature combination and interaction to generate a multi-source heterogeneous clinical data feature set. The specific process is as follows:

[0116] The reduced-dimensional feature vectors generated after feature selection and dimensionality reduction are sequentially input into several cross layers of the cross network for feature interaction, and the cross features are output, namely:

[0117]

[0118] Among them, f l+1 represents the cross features of the l+1th layer, f0 represents the reduced dimension feature vector, T represents the transpose, w1 represents the weight vector of the first cross layer, o1 represents the bias vector of the first cross layer, and f1 represents the cross features of the first layer.

[0119] The cross features are input into the fully connected network of the deep network for deep feature extraction and fusion to generate a multi-source heterogeneous clinical data feature set.

[0120] In this embodiment, since traditional feature combination methods often rely on manual design and are difficult to discover high-order feature interaction patterns, the present invention introduces a deep cross network for feature combination. The deep cross network can automatically learn the combination relationship between features through a special cross-layer structure, such as: the interaction relationship between age-weight-dose, the association between previous medication history-current symptoms-drug selection, and the combined impact of examination type-sedation depth-drug time; through the design of the cross-layer, the features are bitwise multiplied and weighted, making the feature interaction explainable, and can also handle high-dimensional sparse features well. Among them, the simplified structure of the deep cross network is as follows: Figure 3As shown in the figure, it includes a cross network and a deep network. The cross network includes three cross layers, and the deep network is a fully connected layer. Each cross layer realizes feature interaction in the following ways: (1) Explicit feature cross: The nonlinear relationship between features, such as the interaction between age and weight, is captured by performing the outer product operation between the original feature vector and the current layer feature. (2) Feature weight learning: The importance of different feature combinations is dynamically adjusted using learnable weight parameters. (3) Residual connection: The cross feature is added to the original feature to enhance the feature expression while retaining the original information. (4) Deep feature combination: A high-order feature interaction is constructed through a multi-layer cross structure to realize the combination of multi-dimensional features such as age-weight-dose. The fully connected network performs feature processing in the following ways: (1) Nonlinear transformation: The feature is nonlinearly transformed through an activation function (such as ReLU) to enhance the expression ability of the model. (2) Hierarchical feature learning: A more abstract feature representation is extracted layer by layer through a multi-layer structure. (3) Feature fusion: Feature information from different sources is deeply fused to generate a unified feature representation. (4) Dimensional mapping: The high-dimensional feature space is mapped to a feature space suitable for subsequent tasks.

[0121] The multi-objective optimization decision-making module is used to define decision variables and objective constraints based on a multi-source heterogeneous clinical feature set, establish a sedation effect objective function, a safety objective function, and an economic objective function, and use an improved multi-objective optimization method to optimize and solve each objective function to obtain the optimal medication plan; among which, the decision variables include drug type, dosage, and administration time; the objective constraints include dosage limits and contraindications; among which, the optimal medication plan is a sedation medication plan for children with patients that is successful in one medication under safety conditions and does not excessively increase costs.

[0122] In this embodiment, since the optimization of the sedation scheme is essentially a multi-objective decision-making problem, it is necessary to balance multiple objectives such as effect, safety and economy at the same time, and the traditional single-objective optimization method is difficult to deal with the trade-off relationship between objectives, and the simple weighted sum is easy to fall into the local optimum. Therefore, the present invention designs an improved multi-objective optimization method to optimize and solve each objective function, that is, introduces a particle swarm optimization algorithm to optimize the weights of the established multi-objective function, but since the traditional multi-objective particle swarm optimization method uses fixed weights, it is difficult to adapt to the needs of different clinical scenarios. Therefore, the present invention analyzes historical data and establishes a dynamic weight adjustment model based on patient characteristics in the particle swarm optimization method, so that the algorithm can automatically adjust the importance of each objective according to the specific situation.

[0123] Specifically, the sedation effect objective function, safety objective function, and economic objective function are:

[0124]

[0125] Among them, q1 represents the sedation effect index, f1(q1) represents the sedation effect objective function, and eff i (q1) represents the score of the i-th sedation effect index q1, ω i represents the weight of the ith sedation effect index q1, q2 represents the safety index, f2(q2) represents the safety objective function, and risk j (q2) represents the risk probability of the jth security indicator q2, serverity j represents the risk severity of the j-th safety indicator q2, q3 represents the economic indicator, f3(q3) represents the economic objective function, c k represents the unit cost of the k-th drug, dose k (q3) represents the consumption of the kth drug.

[0126] In this embodiment, three core dimensions are considered in the design of the objective function: first, sedation effect, through comprehensive evaluation of indicators such as sedation depth and onset time; second, safety, including sedation failure, risk of adverse reactions, drug interactions, etc.; and finally, economy, taking into account drug costs and resource consumption.

[0127] Specifically, when the improved multi-objective optimization method is used to optimize and solve each objective function, the particle swarm optimization method is introduced to adaptively and dynamically adjust the weights of the sedation effect objective function, the safety objective function, and the economic objective function, and the optimal weights of the sedation effect objective function, the safety objective function, and the economic objective function are obtained by updating the position of each particle to obtain the optimal medication plan; wherein, the position of each particle is updated as:

[0128]

[0129] in, represents the velocity of particle s at time t+1, represents the velocity of particle s at time t, represents the inertia weight, π1 and π2 represent learning factors, r1 and r2 represent random numbers, and pbest s represents the historical optimal position of particle s, gbest represents the global optimal position of the particle swarm, represents the position of particle s at time t, represents the position of particle s at time t+1.

[0130] The medication scheme module is used to automatically generate a specific medication scheme based on the optimal medication scheme and the scheme generation method based on the timing constraints; wherein the specific medication scheme includes the type of medicine and the dosage of the medicine.

[0131] In this embodiment, the medication scheme module adopts a timing constraint-based scheme generation method (TCPG), which solves the multi-drug timing scheduling problem through a dynamic programming method. Specifically, first, the clinical medication timing constraints are formalized into a mathematical model; second, the time dynamic characteristics of drug action are considered; finally, the adjustability of the scheme is achieved; the solution process adopts a hierarchical dynamic programming strategy: the first-level planning determines the order of administration of each drug; the second-level planning optimizes the specific administration time point; the third-level planning handles the real-time adjustment requirements; the details are as follows:

[0132] Based on the optimal medication regimen, obtain the drug combination set, namely:

[0133]

[0134] Among them, M represents the drug combination set, m1 represents the first drug, m2 represents the second drug, and m a represents the ath drug, τ a represents the onset time of drug a, δ a Indicates the duration of action of drug a, α a represents the peak action time of drug a, β a Represents the elimination half-life of drug a.

[0135] The drug combination is centrally arranged in a time sequence, namely:

[0136] T1={(m a ,t a )|m a ∈M,t a ∈R +}

[0137] Among them, T1 represents the medication sequence of the drugs in the drug combination, t a Indicates the time of taking drug a, R + Represents a positive real number.

[0138] According to the medication sequence of the drugs in the drug combination set, the drug sequence minimization objective function is established, that is:

[0139]

[0140] Among them, minJ(T1) represents the drug time sequence minimization objective function, θ a represents the weight coefficient of the objective function of minimizing the time series of drug a, E a represents the optimization target of medication timing of drug a, η represents the regularization coefficient, R ′ represents the timing regularization term.

[0141] Establish the constraints of the drug time sequence minimization objective function, namely:

[0142] The drug interval constraints are:

[0143]

[0144] The action superposition constraint is:

[0145] C(t)=∑ a c ′ a (tt a )≤C max .

[0146] The operation timing constraints are:

[0147]

[0148] Among them, t a Indicates the time of taking drug a, t b represents the duration of drug b administration, Δτ ab Indicates the drug interval threshold, m b represents the bth drug, C(t) represents the overall drug intensity at the tth moment, c ′ a represents the time effect function of the ath drug, C max Indicates the maximum allowable superposition strength, t start Indicates the start time of the medication regimen, t op Indicates the operation start time, d op Indicates the duration of the operation.

[0149] Based on the constraints of the objective function of minimizing the drug sequence, the drug sequence of the drug combination set is solved to generate a specific medication plan, that is, the drug types and drug dosages under the medication sequence.

[0150] The safety inspection module is used to conduct multi-level safety audits on specific medication plans. Through risk assessment, medication dosage constraints, medication time interval constraints, and combined medication constraints, it calculates the feasibility judgment function and obtains multi-level safety audit results.

[0151] In this embodiment, since the safety constraint processing is the last line of defense to ensure the reliability of the system recommendation plan, considering the particularity of medical decision-making, the present invention designs a multi-level safety constraint system based on fuzzy rules; the reason for choosing fuzzy rules is that the safety constraints in clinical practice are often not strict judgments, but have a certain degree of fuzziness and tolerance, and fuzzy rules can better simulate the decision-making process of doctors. The specific operation process is as follows:

[0152] For each specific medication regimen, a risk assessment based on fuzzy rules is used to calculate the risk of the specific medication regimen, namely:

[0153]

[0154] Among them, R1(x) represents the risk of specific medication regimen x, μ z′ (x) represents the membership function of the z′th rule of the specific medication regimen x, w1 z′ represents the zth order of a specific medication regimen x ′ The weight of the rule, μ(x) represents the membership function of the specific medication regimen x, c1 represents the optimal value defined by the expert, that is, the central value, and σ represents the acceptable deviation range, that is, the standard deviation.

[0155] In this embodiment, the membership function of the specific medication regimen x is in the form of a Gaussian function.

[0156] Establish dosing constraints, namely:

[0157] g1(x)=D(x)-D max ≤0

[0158] Among them, g1(x) represents the dosage constraint of the specific medication regimen x, D(x) represents the drug dosage vector of the specific medication regimen x, and D max A vector representing the maximum allowable doses of the drug for a specific dosing regimen x.

[0159] Establish medication interval constraints, namely:

[0160] g2(x)=|T1 a1 -T1 b1 |-I min ≥0

[0161] Among them, g2(x) represents the medication time interval constraint of the specific medication regimen x, T1 a1 Indicates the administration time point of drug a1 in the specific medication regimen x, T1 b1 Indicates the administration time point of drug b1 in the specific medication regimen x, I min Indicates the minimum safe dosing interval.

[0162] Establish constraints on combination medication, namely:

[0163] g3(x)=S int (x)-S th ≤0;

[0164] Among them, g3(x) represents the combined medication constraints of the specific medication regimen x, S int (x) represents the drug interaction risk score function for a specific medication regimen x, S th Represents the drug interaction risk threshold.

[0165] In this embodiment, the purpose of designing the above three types of constraints is: using dosage constraints to ensure that the single dosage and cumulative dosage are within a safe range; using medication time interval constraints to manage the administration sequence and intervals of different drugs; and using combination medication constraints to evaluate the risks of combined use of multiple drugs.

[0166] Calculate the feasibility judgment function and obtain the multi-level security audit results, namely:

[0167] F′(x)=∧(g d′ (x)≤0),d′=1,2,3

[0168] Among them, F′(x) represents the feasibility result of the specific medication plan x, that is, the multi-level safety review result, ∧ represents the judgment function, and d′ represents the specific constraint. When F′(x) is 1, the current specific medication plan x is feasible; when F ′ When (x) is 0, the current specific medication regimen x is not feasible.

[0169] Therefore, through the combination of the above modules, the system can generate personalized medication plans that meet both clinical needs and safety requirements; each algorithm module is designed with a corresponding parameter tuning mechanism, which can be optimized and adjusted according to the actual application effect; at the same time, the system can also realize real-time monitoring of algorithm performance to ensure that the calculation is completed within a clinically acceptable time range; this constraint processing mechanism has shown significant advantages in practice: first, the system can evaluate the safety of the plan in real time and warn of potential risks; second, when the constraints are violated, the system will automatically provide adjustment suggestions to guide doctors to revise the plan; finally, through continuous learning and updating, the system's constraint rule base is continuously optimized, and the accuracy of safety assessment is gradually improved. Finally, the above four system modules are used to realize the full process intelligence from data processing to plan generation, providing reliable technical support for clinical decision-making.

[0170] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0171] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A pediatric procedural sedation medication recommendation system based on multi-objective optimization, characterized in that: include: The data collection and preprocessing module is used to collect multi-source heterogeneous clinical data, clean the multi-source heterogeneous clinical data, extract and fuse the cleaned multi-source heterogeneous clinical data, and generate a multi-source heterogeneous clinical feature set; The multi-objective optimization decision-making module is used to define decision variables and objective constraints based on multi-source heterogeneous clinical feature sets, establish sedation effect objective function, safety objective function, and economic objective function, and use the improved multi-objective optimization method to optimize and solve each objective function to obtain the optimal medication plan; Among them, the best medication regimen is the one that can achieve success in one medication and does not excessively increase the cost for sedative medication for children under the condition of safety. The medication scheme module is used to automatically generate a specific medication scheme based on the optimal medication scheme and the scheme generation method based on the timing constraints; Among them, the specific medication regimen includes the type of drug and the dosage of the drug; The safety inspection module is used to conduct multi-level safety audits on specific medication plans. Through risk assessment, medication dosage constraints, medication time interval constraints, and combined medication constraints, it calculates the feasibility judgment function and obtains multi-level safety audit results.

2. The pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization according to claim 1, characterized in that: The data collection and preprocessing module includes raw data collection module, data cleaning module, and feature engineering module; The raw data acquisition module is used to collect multi-source heterogeneous clinical data; The multi-source heterogeneous clinical data include basic information of the patient, clinical test data, electronic medical record text, previous medication records, and surgical operation related information; the basic information of the patient includes the patient's age, gender, weight, fasting status, and body temperature; The data cleaning module is used to detect outliers, interpolate missing values, structure unstructured text, and convert data formats uniformly for multi-source heterogeneous clinical data, thus obtaining cleaned multi-source heterogeneous clinical data. The feature engineering module is used to extract key features, select and reduce dimensions, and combine and interact features of multi-source heterogeneous clinical data after data cleaning to generate a feature set of multi-source heterogeneous clinical data.

3. The pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization according to claim 2 is characterized in that: The structured processing of unstructured text adopts bidirectional long short-term memory network and conditional random field model to unify the structure of unstructured text in multi-source heterogeneous clinical data and generate structured text of multi-source heterogeneous clinical data. The specific process is as follows: The unstructured text in the multi-source heterogeneous clinical data is input into the input layer and word embedding layer of the bidirectional long short-term memory network and the conditional random field model in turn to generate a word vector matrix, namely: Where n represents the length of the unstructured text input sequence in the multi-source heterogeneous clinical data, X represents the unstructured text input sequence in the multi-source heterogeneous clinical data, X1 represents the unstructured text input at the first moment, X2 represents the unstructured text input at the second moment, and X t represents the unstructured text input at time t, X 1 Indicates the initial status information of the consultation, X 2 Indicates medication preparation information, X 3 Indicates medication regimen information, X 4 Indicates the effective date of medication, X 5 Indicates the end of medication operation, X 6 represents the medication recovery period information, E represents the word vector matrix, e1 represents the d-dimensional vector of the first word, and e2 represents the d-dimensional vector of the second word; Among them, the initial status information of the consultation includes the basic information of the child, previous medication records, and allergy history records; the medication preparation information includes the pre-sedation assessment results, examination and operation types, and the planned sedation depth; the medication regimen information includes the first medication dose, medication method, and baseline values ​​of monitoring indicators, and the monitoring indicators are blood oxygen saturation, pulse rate, and respiratory rate; the medication effective period information includes changes in vital signs, sedation depth scores, and adverse reaction records; the medication operation end information includes the completion of the operation, the cumulative medication dose, and the sedation effect evaluation status; the medication recovery period information includes the recovery time, recovery status, and follow-up records; The word vector matrix is ​​input into the bidirectional long short-term memory network layer, and the context information of the word vector matrix is ​​obtained and input into the hidden layer to obtain the hidden state of the word vector matrix, that is: Among them, LSTM represents long short-term memory network, h t represents the hidden state at time t, represents the hidden state of the forward LSTM network at time t, that is, the state evolution information from the admission of the patient to the current moment, represents the hidden state of the forward long short-term memory network at time t-1, represents the hidden state of the LSTM network after the tth moment, that is, the state information from the expected awakening to the current moment, represents the hidden state of the backward long short-term memory network at time t-1, e t Represents the word vector at time t; The hidden state of the word vector matrix is ​​input into the conditional random field layer to generate a structured text sequence of multi-source heterogeneous clinical data, namely: Among them, Y represents the annotation sequence, that is, the optimized structured text sequence, Y1 represents the annotation label of the first word, Y2 represents the annotation label of the second word, and Y τ represents the annotation label of the τth word, P(Y|X) represents the conditional probability distribution of the structured text sequence, exp represents the exponential function, score(X,Y) represents the clinical rationality of the structured text sequence, including the appropriateness of the medication regimen, the correctness of the temporal logic, and the degree of compliance with medical regulations, and Z(X) represents the normalization factor, that is, the overall space of all possible medication regimens under the current clinical condition.

4. The pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization according to claim 3 is characterized in that: Feature selection uses the least absolute shrinkage and selection operator regression method based on L1 regularization to screen key features and generate the selected key features, namely: Among them, min γ represents the minimum value, n′ represents the number of samples of multi-source heterogeneous clinical data, z represents the target variable of the key feature, F represents the key feature matrix, γ represents the key feature weight vector, λ represents the regularization parameter, ‖·‖1 and ‖·‖2 represent the L1 norm and L2 norm respectively; Among them, key characteristics include age, weight, ASA rating, past medical history, and examination type.

5. The pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization according to claim 4, characterized in that: When combining and interacting features, a deep cross network is introduced to perform combined interaction to generate a multi-source heterogeneous clinical data feature set. The specific process is as follows: The reduced-dimensional feature vectors generated after feature selection and dimensionality reduction are sequentially input into several cross layers of the cross network for feature interaction, and the cross features are output, namely: Among them, f l+1 represents the cross feature of the l+1th layer, f0 represents the dimension reduction feature vector, T represents the transpose, w1 represents the weight vector of the first cross layer, o1 represents the bias vector of the first cross layer, and f1 represents the cross feature of the first layer; The cross features are input into the fully connected network of the deep network for deep feature extraction and fusion to generate a multi-source heterogeneous clinical data feature set.

6. The pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization according to claim 5, characterized in that: Decision variables and target constraints, specifically: Decision variables include drug type, dosage, and administration time; target constraints include dosage limits and contraindications.

7. The pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization according to claim 6, characterized in that: The sedation effect objective function, safety objective function, and economic objective function are: Among them, q1 represents the sedation effect index, f1(q1) represents the sedation effect objective function, and eff i (q1) represents the score of the i-th sedation effect index q1, ω i represents the weight of the ith sedation effect index q1, q2 represents the safety index, f2(q2) represents the safety objective function, and risk j (q2) represents the risk probability of the jth security indicator q2, serverity j represents the risk severity of the j-th safety indicator q2, q3 represents the economic indicator, f3(q3) represents the economic objective function, c k represents the unit cost of the k-th drug, dose k (q3) represents the consumption of the kth drug.

8. The pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization according to claim 7, characterized in that: When the improved multi-objective optimization method is used to optimize and solve each objective function, the particle swarm optimization method is introduced to adaptively and dynamically adjust the weights of the sedation effect objective function, the safety objective function, and the economic objective function, and by updating the position of each particle, the optimal weights of the sedation effect objective function, the safety objective function, and the economic objective function are obtained to obtain the optimal medication plan; Among them, the position of each particle is updated as follows: in, represents the velocity of particle s at time t+1, represents the velocity of particle s at time t, represents the inertia weight, π1 and π2 represent learning factors, r1 and r2 represent random numbers, and pbest s represents the historical optimal position of particle s, gbest represents the global optimal position of the particle swarm, represents the position of particle s at time t, represents the position of particle s at time t+1.

9. The pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization according to claim 8, characterized in that: According to the optimal medication plan, the specific process of automatically generating a specific medication plan based on the scheme generation method of time constraints is as follows: Based on the optimal medication regimen, obtain the drug combination set, namely: Among them, M represents the drug combination set, m1 represents the first drug, m2 represents the second drug, and m a represents the ath drug, τ a represents the onset time of drug a, δ a Indicates the duration of action of drug a, α a represents the peak action time of drug a, β a represents the elimination half-life of drug a; The drug combination is centrally arranged in a time sequence, namely: T1={(m a ,t a )|m a ∈M,t a ∈R + } Among them, T1 represents the medication sequence of the drugs in the drug combination, t a Indicates the time of taking drug a, R + represents a positive real number; According to the medication sequence of the drugs in the drug combination set, the drug sequence minimization objective function is established, that is: Among them, minJ(T1) represents the objective function of minimizing the drug time sequence, represents the weight coefficient of the objective function of minimizing the time series of drug a, E a represents the optimization target of medication timing of drug a, η represents the regularization coefficient, and R′ represents the timing regularization term; Establish the constraints of the drug time sequence minimization objective function, namely: The drug interval constraints are: The action superposition constraint is: C(t)=∑ a c′ a (t-t a )≤C max ; The operation timing constraints are: Among them, t b represents the duration of drug b administration, Δτ ab Indicates the drug interval threshold, m b represents the bth drug, C(t) represents the overall drug intensity at the tth moment, c′ a represents the time effect function of the ath drug, C max Indicates the maximum allowable superposition strength, t start Indicates the start time of the medication regimen, t op Indicates the operation start time, d op Indicates the duration of the operation; Based on the constraints of the objective function of minimizing the drug sequence, the drug sequence of the drug combination set is solved to generate a specific medication plan, that is, the drug types and drug dosages under the medication sequence.

10. The pediatric procedural sedation medication regimen recommendation system based on multi-objective optimization according to claim 9, characterized in that: Conduct multi-level safety audits on specific medication regimens, calculate feasibility judgment functions through risk assessment, medication dosage constraints, medication time interval constraints, and combined medication constraints, and obtain the specific process of multi-level safety audit results as follows: For each specific medication regimen, a risk assessment based on fuzzy rules is used to calculate the risk of the specific medication regimen, namely: Among them, R1(x) represents the risk of specific medication regimen x, μ z′ (x) represents the membership function of the z′th rule of the specific medication regimen x, w1 z′ represents the weight of the z′th rule of the specific medication regimen x, μ(x) represents the membership function of the specific medication regimen x, c1 represents the optimal value defined by the expert, i.e., the central value, and σ represents the acceptable deviation range, i.e., the standard deviation; Establish dosing constraints, namely: g1(x)=D(x)-D max ≤0 Among them, g1(x) represents the dosage constraint of the specific medication regimen x, D(x) represents the drug dosage vector of the specific medication regimen x, and D max The vector representing the maximum allowable dose of the drug for a specific medication regimen x; Establish medication interval constraints, namely: g2(x)=|T1 a1 -T1 b1 |-I min ≥0 Among them, g2(x) represents the medication time interval constraint of the specific medication regimen x, T1 a1 Indicates the administration time point of drug a1 in the specific medication regimen x, T1 b1 Indicates the administration time point of drug b1 in the specific medication regimen x, I min Indicates the minimum safe dosing interval; Establish constraints on combination medication, namely: g3(x)=S int (x)-S th ≤0; Among them, g3(x) represents the combined medication constraints of the specific medication regimen x, S int (x) represents the drug interaction risk score function for a specific medication regimen x, S th represents the drug interaction risk threshold; Calculate the feasibility judgment function and obtain the multi-level security audit results, namely: F′(x)=∧(g d′ (x)≤0),d′=1,2,3 Among them, F′(x) represents the feasibility result of the specific medication plan x, that is, the multi-level safety review result, ∧ represents the judgment function, and d′ represents the specific constraint. When F′(x) is 1, the current specific medication plan x is feasible; when F′(x) is 0, the current specific medication plan x is not feasible.

Images