A multi-element knowledge-driven pediatric perioperative respiratory system anesthesia risk assessment method, system, terminal and medium

Abstract

The application relates to the technical field of anesthesia risk assessment, and discloses a multi-element knowledge driven anesthesia risk assessment method, system, terminal and medium for a child's respiratory system during a perioperative period, which comprises the following steps: acquiring heterogeneous clinical knowledge, processing the heterogeneous clinical knowledge, obtaining risk benchmark characterization, risk probability prior distribution and medical graph embedding features; acquiring real-time multi-modal data of a target child, and obtaining a knowledge calibrated data driven risk probability distribution according to the real-time multi-modal data, the risk probability prior distribution and the medical graph embedding features; and obtaining a final risk assessment distribution and a structured assessment report according to the risk benchmark characterization and the data driven risk probability distribution. The application can improve the accuracy of anesthesia risk assessment for children.

Description

Technical Field

[0001] This application relates to the field of anesthesia risk assessment technology, and in particular to a multi-knowledge-driven method, system, terminal, and medium for assessing the risk of respiratory anesthesia in children during the perioperative period. Background Technology

[0002] Perioperative respiratory adverse events are the most common cause of serious adverse events during anesthesia and surgery in children, and represent the most frequent and dangerous safety issue among anesthesia-related complications. The risks of perioperative respiratory anesthesia in children are characterized by significant individual variability, high complexity of conditions, and rapid changes in risk. Children of different ages exhibit significant differences in airway structure, physiological function, drug response, and compensatory capacity, leading to inconsistent risk profiles for the same anesthesia regimen among different children. However, current risk assessment methods for perioperative respiratory anesthesia in children primarily rely on anesthesiologists' comprehensive judgment based on medical history, physical signs, and personal experience, lacking structured and intelligent dynamic assessment tools.

[0003] In the field of anesthesiology, artificial intelligence (AI) is gradually becoming a crucial technology for promoting precision medicine and intelligent decision-making. Machine learning, deep learning, and natural language processing methods can efficiently analyze massive amounts of complex clinical data, demonstrating stronger capabilities in pattern recognition and risk prediction compared to traditional statistical analysis and experience-based judgment. Currently, AI has shown good application potential in adult anesthesia and intensive care, but its adaptation and implementation in perioperative respiratory anesthesia risk assessment for children still have significant shortcomings.

[0004] Existing general medical prediction models tend to be data-driven in isolation, making it difficult to effectively incorporate child-specific prior clinical knowledge. On the one hand, conventional models often rely on single data sources and fail to organically integrate multi-source, heterogeneous clinical medical information, resulting in severely inaccurate assessment results when faced with highly complex individual differences among children. On the other hand, algorithms that rely solely on data training struggle to effectively incorporate child-specific prior clinical knowledge such as airway structure and pathophysiology, often neglecting valuable clinical expert experience and authoritative guideline consensus. This leads to a lack of clinical interpretability in the models, making it difficult to form a decision-making loop that aligns with medical logic.

[0005] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0006] The main purpose of this application is to provide a multi-dimensional knowledge-driven method, system, terminal, and medium for assessing the risk of respiratory anesthesia in children during the perioperative period. This aims to address the problem that existing general medical prediction models used in anesthesia risk assessment tend to be data-driven and fail to effectively incorporate children-specific clinical prior knowledge, resulting in low accuracy in pediatric anesthesia risk assessment.

[0007] The first aspect of this application provides a multi-knowledge-driven method for assessing the risk of respiratory anesthesia in children during the perioperative period. This method includes the following steps: Heterogeneous clinical knowledge is acquired and processed to obtain risk benchmark representation, risk probability prior distribution and medical graph embedding features; Acquire real-time multimodal data of the target child, and obtain a knowledge-calibrated data-driven risk probability distribution based on the real-time multimodal data, the prior risk probability distribution, and the medical graph embedding features; Based on the risk benchmark characterization and the data-driven risk probability distribution, the final risk assessment distribution and structured assessment report are obtained.

[0008] Optionally, in one embodiment of this application, the heterogeneous clinical knowledge includes clinical guidelines and expert consensus, clinical expert experience, and electronic medical record text data; The process of processing the heterogeneous clinical knowledge to obtain risk benchmark representation, risk probability prior distribution, and medical graph embedding features specifically includes: Key risk conditions were extracted from the clinical guidelines and expert consensus, and rule-driven risk benchmarks were derived based on these key risk conditions. A soft-constrained probability prior network is obtained by modeling based on the clinical expert experience. The conditional probability is then calibrated using the soft-constrained probability prior network based on historical medical record statistics to obtain the risk probability prior distribution. A knowledge graph is constructed based on the electronic medical record text data, and the knowledge graph is transformed into medical graph embedding features using graph embedding.

[0009] Optionally, in one embodiment of this application, obtaining the knowledge-calibrated data-driven risk probability distribution based on the real-time multimodal data, the prior risk probability distribution, and the medical graph embedding features specifically includes: Cross-modal attention fusion is performed based on the real-time multimodal data and the medical graph embedding features to obtain a dynamic data-driven representation; Prior constraint optimization is performed based on the dynamic data-driven representation and the prior risk probability distribution to obtain a knowledge-calibrated data-driven risk probability distribution.

[0010] Optionally, in one embodiment of this application, the real-time multimodal data includes vital sign parameters, medical imaging data, and physiological waveform signals; The step of performing cross-modal attention fusion based on the real-time multimodal data and the medical graph embedding features to obtain a dynamic data-driven representation specifically includes: The vital signs parameters, medical imaging data, and physiological waveform signals are input into a pre-trained encoder for feature extraction and unified representation to obtain a high-dimensional multimodal feature vector. Using the medical image embedding features as the query and the high-dimensional multimodal feature vector as the key and value, attention weights are calculated to obtain a dynamic data-driven representation: ; ; ; ; in, For dynamic data-driven representation, For attention mechanisms, It is a flexible maximum value function. For medical graph embedding features, It is a high-dimensional multimodal feature vector. To query the weight matrix, For query vector, The key weight matrix, For key vectors, It is the transpose symbol. For value weight matrix, For value vectors, is the dimension of the key vector.

[0011] Optionally, in one embodiment of this application, the step of performing prior constraint optimization based on the dynamic data-driven representation and the prior risk probability distribution to obtain a knowledge-calibrated data-driven risk probability distribution specifically includes: The dynamic data-driven representation is input into the Transformer encoding module for high-order semantic modeling to obtain high-order features. The high-order features are then nonlinearly mapped through a multilayer perceptron to output the initial risk probability distribution. Construct a joint loss function based on the aforementioned prior distribution of risk probabilities: ; in, For the joint loss function, Let cross-entropy be the loss function. For the initial risk probability distribution, This is a true risk label. For adaptive dynamic weight parameters, Let KL divergence be the KL divergence. This represents the prior distribution of risk probability. Based on the initial risk probability distribution and the joint loss function, a knowledge-calibrated data-driven risk probability distribution is output.

[0012] Optionally, in one embodiment of this application, obtaining the final risk assessment distribution and structured assessment report based on the risk benchmark characterization and the data-driven risk probability distribution specifically includes: The risk benchmark characterization is processed to obtain the rule-based risk probability distribution; Based on the rule-based risk probability distribution and the data-driven risk probability distribution, the final risk assessment distribution is obtained; A structured assessment report is output based on the final risk assessment distribution.

[0013] Optionally, in one embodiment of this application, obtaining the final risk assessment distribution based on the rule-based risk probability distribution and the data-driven risk probability distribution specifically involves: The rule-based risk probability distribution and the data-driven risk probability distribution are calculated using a dynamic gating mechanism formula to obtain the final risk assessment distribution: ; in, For the final risk assessment distribution, For dynamic gating coefficients, To drive risk probability distribution, This represents the probability distribution of rule-based risks.

[0014] A second aspect of this application also provides a multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment system, wherein the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment system is used to implement the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method described in any of the above solutions; the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment system includes: A multivariate clinical knowledge encoding module is used to acquire heterogeneous clinical knowledge and process the heterogeneous clinical knowledge to obtain risk benchmark representation, risk probability prior distribution and medical graph embedding features. The inference module is used to acquire real-time multimodal data of the target child and, based on the real-time multimodal data, the prior risk probability distribution, and the medical graph embedding features, obtain a knowledge-calibrated data-driven risk probability distribution. The decision fusion and expression module is used to obtain the final risk assessment distribution and structured assessment report based on the risk benchmark characterization and the data-driven risk probability distribution.

[0015] A third aspect of this application also provides a terminal, wherein the terminal includes: a memory, a processor, and a multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment program stored in the memory and executable on the processor, wherein when the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment program is executed by the processor, it implements the steps of the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method as described above.

[0016] A fourth aspect of this application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment program, and when the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment program is executed by a processor, it implements the steps of the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method as described above.

[0017] Beneficial effects: This application provides a multi-dimensional knowledge-driven method, system, terminal, and medium for assessing the risk of respiratory anesthesia in children during the perioperative period. By constructing a multi-level knowledge coding framework and optimization constraint mechanism, this application realizes the collaborative modeling of multimodal data and heterogeneous clinical knowledge. Furthermore, through an adaptive gating fusion mechanism, it achieves adaptive compensation of expert experience in scenarios with abnormal data, thereby improving the accuracy of pediatric anesthesia risk assessment. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a preferred embodiment of the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method of this application; Figure 2 This is a schematic diagram of the structured representation and unified coding of multiple clinical knowledge in a preferred embodiment of the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method of this application; Figure 3 This is a schematic diagram of a multi-modal fusion risk assessment model guided by multi-knowledge, which is a preferred embodiment of the multi-knowledge-driven risk assessment method for perioperative respiratory anesthesia in children according to this application. Figure 4 This is a schematic diagram of an interpretable hybrid decision-making mechanism in a preferred embodiment of the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method of this application; Figure 5 This is a structural diagram of a preferred embodiment of the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment system of this application; Figure 6 This is a structural diagram of a preferred embodiment of the terminal of this application.

[0020] Explanation of reference numerals in the attached figures: 100. Multivariate clinical knowledge coding module; 200. Reasoning module; 300. Decision fusion and expression module. Detailed Implementation

[0021] To make the objectives, technical solutions, and effects of this application clearer and more explicit, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only possible technical implementations of this application and not all possible implementations. Based on the embodiments in this application, those skilled in the art can obtain other embodiments without creative effort, and these embodiments are also within the protection scope of this application.

[0022] First, let's introduce the terms used in the embodiments of this application: AI, Artificial Intelligence; DAG, Directed Acyclic Graph; URI, Upper Respiratory Infection; OSAS, Obstructive Sleep Apnea Syndrome; KL, Kullback-Leibler divergence, KL divergence (relative entropy, used to measure the difference between two probability distributions); CE, Cross-Entropy (commonly used in loss functions); MLP, Multilayer Perceptron (a type of feedforward neural network). Softmax, the softmax function, is a function that maps a vector to a probability distribution. Attention, Attention mechanism (focusing the model on important features). Transformer, Transformer model, Transformer model (a deep learning architecture based on self-attention). GAT stands for Graph Attention Network.

[0023] To address the shortcomings of existing technologies in pediatric anesthesia risk assessment, such as subjectivity and lack of quantitative tools, and the limitations of data-driven methods in terms of model reliability, stability, and interpretability due to sample scarcity and individual differences, this application proposes a multi-dimensional knowledge-driven intelligent assessment method for pediatric perioperative respiratory anesthesia risk. This method aims to solve the problem of difficulty in unified modeling of explicit rules, implicit experience, and medical knowledge graphs due to heterogeneous expression forms by constructing a multi-level knowledge coding framework. Furthermore, it addresses the disconnect between traditional data-driven methods and clinical prior knowledge, resulting in low generalization performance and poor prediction robustness in small sample scenarios, through cross-modal attention mechanisms and optimization mechanisms incorporating prior constraints. Finally, it solves the challenge of coordinating symbolic reasoning and numerical prediction across different data qualities through an adaptive gating fusion mechanism, thereby improving the robustness and clinical interpretability of anesthesia risk assessment.

[0024] This application combines the advantages of mobility, multi-view visual monitoring, and localized intelligent analysis. The system is carried by a mobile all-in-one hardware device equipped with dual cameras to collect real-time facial and body images of the children being tested. The system also has the function of inputting and comprehensively processing data such as vital signs, physiological signals, and clinical text.

[0025] Specifically, this application achieves assessment through three stages: knowledge encoding, risk modeling, and adaptive decision-making. First, the system organizes clinical guidelines to construct a directed acyclic graph decision system for symbolic reasoning. Simultaneously, expert experience and medical atlases are modeled as probabilistic prior networks and low-dimensional graph embeddings, respectively, achieving a unified representation of heterogeneous knowledge. Then, using knowledge graph embedding as the query and multimodal features such as facial features, body features, and vital signs as keys and values, a cross-modal attention mechanism guides the fusion process. A prediction model is trained using a knowledge constraint mechanism incorporating cross-entropy and KL divergence, outputting data-driven risk distribution results. Furthermore, the system utilizes an adaptive gating mechanism to dynamically adjust the gating coefficients based on the integrity of the input data, adaptively fusing symbolic reasoning and data prediction results, and outputting a structured assessment report from three levels: risk factors, triggering rules, and similar cases.

[0026] Compared with existing technologies, this application achieves collaborative modeling of multimodal data and heterogeneous clinical knowledge by constructing a multi-level knowledge coding framework and optimizing constraint mechanisms. This solves the problems of data-driven methods being disconnected from prior knowledge and having poor generalization and robustness in small sample scenarios. Through an adaptive gating fusion mechanism, it achieves adaptive compensation of expert experience in data anomaly scenarios, improving the quantitative accuracy, environmental robustness, and multidimensional clinical interpretability of the model in anesthesia risk assessment in complex clinical environments.

[0027] The multi-knowledge-driven intelligent assessment system for perioperative respiratory anesthesia risk in children proposed in this application is carried out by a mobile all-in-one device equipped with dual cameras to collect real-time images of children's faces and bodies, and integrates vital signs, physiological signals and clinical text to form a multimodal input.

[0028] Specifically, this application achieves localized assessment through three stages: First, in the knowledge encoding stage, clinical guidelines are compiled to construct a directed acyclic graph decision system for symbolic reasoning. Simultaneously, expert experience and medical atlases are modeled as probabilistic prior networks and low-dimensional graph embeddings, respectively, to achieve a unified representation of heterogeneous knowledge. Second, in the risk modeling stage, knowledge graph embedding is used as the query, and a cross-modal attention mechanism guides the fusion of multimodal features. A knowledge constraint mechanism including cross-entropy and KL divergence is used to train the model, outputting risk distribution results. Further, in the adaptive decision-making stage, a gating mechanism is used to dynamically adjust coefficients based on data integrity, adaptively fusing symbolic reasoning and data prediction results to achieve adaptive compensation of expert experience. A structured assessment report is output from three levels: risk factors, triggering rules, and similar cases.

[0029] The technical solutions of this application 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.

[0030] The preferred embodiment of this application describes a multi-knowledge-driven method for assessing the risk of respiratory anesthesia in children during the perioperative period. Figure 1 As shown, the multi-knowledge-driven method for assessing the risk of perioperative respiratory anesthesia in children includes the following steps: In step S10, heterogeneous clinical knowledge is acquired and processed to obtain risk benchmark representation, risk probability prior distribution and medical graph embedding features.

[0031] In one possible implementation, the heterogeneous clinical knowledge includes clinical guidelines and expert consensus, clinical expert experience, and electronic medical record text data; step S10 specifically includes: extracting key risk conditions from the clinical guidelines and expert consensus, and obtaining a rule-driven risk benchmark representation based on the key risk conditions; modeling based on the clinical expert experience to obtain a soft-constrained probability prior network, and calibrating the conditional probabilities based on historical medical record statistics through the soft-constrained probability prior network to obtain a risk probability prior distribution; constructing a knowledge graph based on the electronic medical record text data, and converting the knowledge graph into medical graph embedding features using graph embedding.

[0032] like Figure 2 As shown, to address the challenges of complex clinical knowledge sources and heterogeneous expression in perioperative respiratory anesthesia risk assessment for children, a multivariate clinical knowledge coding framework needs to be constructed to uniformly model and computably represent knowledge from clinical guidelines, expert experience, and electronic medical records. First, based on guidelines and expert consensus on difficult airway management in children, perioperative respiratory management, assessment of surgical timing for URI patients, and anesthesia management of OSAS, key risk conditions such as age <1 year, recent history of respiratory infection, airway grade ≥III, and sleep apnea symptoms are extracted and organized into a rule graph structure. The system uses a directed acyclic graph to represent the logical dependencies between different rules. Through symbolic reasoning mechanisms, rule matching and path triggering are performed on input cases to achieve interpretable coarse-grained risk reasoning and output an independent rule-driven risk benchmark representation. This representation, as the explicit logical backbone, provides deterministic clinical logical constraints and basic calibration information for subsequent risk modeling.

[0033] Meanwhile, the system performs independent parallel coding for tacit knowledge in clinical experience and textual medical records: on the one hand, it models expert experience that is difficult to explicitly regularize as a soft-constrained probabilistic prior network, describes the dependencies of different risk factors through a probabilistic graphical model, and calibrates the conditional probabilities by combining historical case statistics to obtain the prior distribution of risk probabilities. On the other hand, based on the text graph construction method, the system automatically mines the relationships between diseases, symptoms, examination indicators, and complications from electronic medical record text using graph attention networks, constructs a medical knowledge graph of perioperative anesthesia risks, and obtains medical graph embedding features using graph embedding methods. This is used to characterize typical risk combination patterns and individualized risk triggers. Thus, step S10 achieves the independent symbolization and vectorization encoding of the three knowledge sources.

[0034] In step S20, real-time multimodal data of the target child is acquired, and a knowledge-calibrated data-driven risk probability distribution is obtained based on the real-time multimodal data, the prior risk probability distribution, and the medical graph embedding features.

[0035] In one possible implementation, step S20 specifically includes: performing cross-modal attention fusion based on the real-time multimodal data and the medical graph embedding features to obtain a dynamic data-driven representation; and performing prior constraint optimization based on the dynamic data-driven representation and the prior risk probability distribution to obtain a knowledge-calibrated data-driven risk probability distribution.

[0036] In one possible implementation, the real-time multimodal data includes vital sign parameters, medical imaging data, and physiological waveform signals. The specific implementation of obtaining the dynamic data-driven representation involves: inputting the vital sign parameters, medical imaging data, and physiological waveform signals into a pre-trained encoder for feature extraction and unified representation to obtain a high-dimensional multimodal feature vector; using the medical image embedding features as a query, and the high-dimensional multimodal feature vector as a key and value, calculating attention weights to obtain the dynamic data-driven representation. ; ; ; ; in, For dynamic data-driven representation, For attention mechanisms, It is a flexible maximum value function. For medical graph embedding features, It is a high-dimensional multimodal feature vector. To query the weight matrix, For query vector, The key weight matrix, For key vectors, It is the transpose symbol. For value weight matrix, For value vectors, is the dimension of the key vector.

[0037] In one possible implementation, obtaining the knowledge-calibrated data-driven risk probability distribution involves: inputting the dynamic data-driven representation into a Transformer encoding module for high-order semantic modeling to obtain high-order features; performing nonlinear mapping on the high-order features using a multilayer perceptron to output an initial risk probability distribution; and constructing a joint loss function based on the prior risk probability distribution. ; in, For the joint loss function, Let cross-entropy be the loss function. For the initial risk probability distribution, This is a true risk label. For adaptive dynamic weight parameters, Let KL divergence be the KL divergence. This represents the prior distribution of risk probability. Based on the initial risk probability distribution and the joint loss function, a knowledge-calibrated data-driven risk probability distribution is output.

[0038] like Figure 3 As shown, in the risk modeling stage, the knowledge codes at each level independently exported in step S10 are used for deep collaborative modeling with the multimodal real-time data of the current case. The multimodal real-time data includes vital sign parameters, medical imaging data, and physiological waveform signals, etc., and features are extracted and uniformly represented using pre-trained encoders for the corresponding modalities to obtain high-dimensional multimodal feature vectors. Building upon this foundation, the system employs a hierarchical interaction and fusion mechanism to jointly model knowledge features and multimodal data features, thereby achieving dynamic characterization and accurate prediction of perioperative respiratory anesthesia risks in children. First, embedding features using electronic medical record images. As a knowledge-oriented query, it utilizes real-time multimodal features. As keys and values, a cross-modal attention mechanism guides the model to focus on key clinical features associated with known risk patterns, such as abnormal respiratory rate, blood oxygen fluctuations, and airway structural features, ultimately outputting a highly personalized, dynamic, data-driven representation. In the above formula, , , It is a learnable linear projection matrix used to map heterogeneous features to a unified representation space; This is a scaling factor used to prevent gradient vanishing in the Softmax function. This mechanism uses knowledge-guided filtering to remove noise from multimodal data, achieving semantic enrichment of key risk features.

[0039] Then, the dynamically fused features output by the cross-modal attention mechanism are... The input is processed by a Transformer encoding module for high-order semantic modeling, and combined with a multilayer perceptron architecture to achieve the final nonlinear mapping, outputting a data-driven risk probability distribution. To incorporate expert experience and knowledge into the risk modeling process, this implementation plan introduces medical probability prior constraints, using the risk probability prior distribution output in step S10. As knowledge supervision information during the training phase, the following knowledge-constrained joint loss function for optimization learning is explicitly constructed. In the above formula, This is a true label of the risks that actually occur in clinical practice. This is the cross-entropy loss function, used to ensure the accuracy of the model's risk prediction. The KL divergence term is used to constrain the data-driven risk distribution output by the deep model to maintain spatial consistency with the prior probability distribution of expert experience. An adaptive dynamic weighting parameter is used to balance prediction accuracy with prior knowledge constraints.

[0040] This step guides multimodal data alignment at the feature layer through graph embedding features and implements regularization constraints at the loss function layer using probabilistic priors. While ensuring that the prediction results conform to clinical medical logic, it improves the stability and generalization ability of the model in extreme scenarios such as small samples, low probability and rare perioperative complications, and finally outputs risk assessment results that have been calibrated by knowledge.

[0041] In step S30, based on the risk benchmark characterization and the data-driven risk probability distribution, the final risk assessment distribution and structured assessment report are obtained.

[0042] In one possible implementation, step S30 specifically includes: processing the risk benchmark characterization to obtain a rule-based risk probability distribution; obtaining a final risk assessment distribution based on the rule-based risk probability distribution and the data-driven risk probability distribution; and outputting a structured assessment report based on the final risk assessment distribution.

[0043] In one possible implementation, the final risk assessment distribution is specifically achieved as follows: The rule-based risk probability distribution and the data-driven risk probability distribution are calculated using a dynamic gating mechanism formula to obtain the final risk assessment distribution: ; in, For the final risk assessment distribution, For dynamic gating coefficients, To drive risk probability distribution, This represents the probability distribution of rule-based risks.

[0044] like Figure 4 As shown, in the final decision-making stage, a hybrid decision-making mechanism integrating symbolic reasoning and neural network reasoning is constructed to achieve stable and interpretable perioperative risk assessment. Specifically, this hybrid decision-making mechanism employs two decision-making approaches in parallel: on the one hand, the system performs symbolic reasoning based on the clinical decision rule system, representing the risk benchmark driven by the rules output in step one. The linear transformation matrix is ​​used to map the rule risk probability distribution of the decision layer. This is used to provide a benchmark judgment with medical consensus; simultaneously, the predicted distribution is driven by the multimodal fusion feature output data enriched in step two. This is used to capture the individualized dynamic risk characteristics of patients. To achieve adaptive fusion of the two types of decision outcomes, this implementation scheme designs a learnable dynamic gating mechanism to obtain the final risk assessment distribution. In the above formula, These are learnable dynamic gating coefficients, with values ​​ranging from 0 to 1. This gating mechanism allows the model to increase the weighting coefficients when the real-time multimodal data is of reliable quality and rich in features. This allows the system to fully leverage the statistical learning capabilities and generalization advantages of neural networks; however, in situations involving missing data, abnormal noise, or rare extreme conditions, the system will automatically reduce... Enhance the understanding of explicit guidelines and rules By adjusting the dependence ratio, the predictive bias of the deep model can be corrected through deterministic clinical logic, thereby significantly improving the robustness of the decision-making system.

[0045] This application determines risk levels based on the adaptively fused risk assessment distribution and collaboratively outputs a structured assessment report from three levels: risk factors, triggering rules, and similar cases. At the risk factor level, the system analyzes cross-modal attention weights, quantitatively extracts and ranks the dynamic individualized triggers that contribute most to the current case. At the triggering rule level, the system backtracks the symbolic reasoning path of the directed acyclic graph, explicitly outputting the clinical guideline entries activated by the current case, thus providing interpretable evidence support for the explicit guideline rule dimension. At the similar case level, the system utilizes the low-dimensional embedding features of the knowledge graph to retrieve and present typical cases with highly similar clinical manifestations and corresponding treatment plans from the historical database.

[0046] Next, referring to the accompanying drawings, a multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment system according to an embodiment of this application is described, which is used to implement the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method described in any of the above schemes.

[0047] Figure 5 This is a structural diagram of the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment system according to an embodiment of this application.

[0048] like Figure 5 As shown, the multi-dimensional knowledge-driven pediatric perioperative respiratory anesthesia risk assessment system includes: a multi-dimensional clinical knowledge coding module 100, a reasoning module 200, and a decision fusion and expression module 300.

[0049] Specifically, the multivariate clinical knowledge encoding module 100 is used to acquire heterogeneous clinical knowledge and process the heterogeneous clinical knowledge to obtain risk benchmark representation, risk probability prior distribution and medical graph embedding features. The inference module 200 is used to acquire real-time multimodal data of the target child, and obtain a knowledge-calibrated data-driven risk probability distribution based on the real-time multimodal data, the prior risk probability distribution, and the medical graph embedding features. The decision fusion and expression module 300 is used to obtain the final risk assessment distribution and structured assessment report based on the risk benchmark characterization and the data-driven risk probability distribution.

[0050] Figure 6 A structural diagram of a terminal provided in an embodiment of this application. The terminal may include: The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0051] When the processor 502 executes the program, it implements the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method provided in the above embodiments.

[0052] Furthermore, the terminal also includes: Communication interface 503 is used for communication between memory 501 and processor 502.

[0053] The memory 501 is used to store computer programs that can run on the processor 502.

[0054] Memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0055] If the memory 501, processor 502, and communication interface 503 are implemented independently, then the communication interface 503, memory 501, and processor 502 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EIS) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0056] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0057] Processor 502 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0058] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described multi-knowledge-driven method for assessing the risk of respiratory anesthesia in children during the perioperative period.

[0059] One embodiment of this application provides a computer program product, including a computer program that, when executed by a processor, implements the features described in this application. Figure 1 The corresponding embodiments provide a multivariate knowledge-driven method for assessing the risk of respiratory anesthesia in children during the perioperative period.

[0060] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to user-analyzed data, user-stored data, user-displayed data, etc.) and signals involved in this invention are all information, data and signals authorized by the user or fully authorized by all parties; and the collection, use and processing of relevant information, data and signals comply with the laws, regulations and standards of relevant countries and regions.

[0061] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0062] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0063] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0064] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable storage medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable storage medium could be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0065] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0066] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0067] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0068] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

[0069] It should be understood that the application of this application is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 therein. Such 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 this application.

Claims

1. A multi-knowledge-driven method for assessing the risk of respiratory anesthesia in children during the perioperative period, characterized in that, The multi-knowledge-driven approach to assessing pediatric perioperative respiratory anesthesia risk includes: Heterogeneous clinical knowledge is acquired and processed to obtain risk benchmark representation, risk probability prior distribution and medical graph embedding features; Acquire real-time multimodal data of the target child, and obtain a knowledge-calibrated data-driven risk probability distribution based on the real-time multimodal data, the prior risk probability distribution, and the medical graph embedding features; Based on the risk benchmark characterization and the data-driven risk probability distribution, the final risk assessment distribution and structured assessment report are obtained.

2. The multi-knowledge-driven method for assessing the risk of pediatric perioperative respiratory anesthesia according to claim 1, characterized in that, The heterogeneous clinical knowledge includes clinical guidelines and expert consensus, clinical expert experience, and electronic medical record text data; The process of processing the heterogeneous clinical knowledge to obtain risk benchmark representation, risk probability prior distribution, and medical graph embedding features specifically includes: Key risk conditions were extracted from the clinical guidelines and expert consensus, and rule-driven risk benchmarks were derived based on these key risk conditions. A soft-constrained probability prior network is obtained by modeling based on the clinical expert experience. The conditional probability is then calibrated using the soft-constrained probability prior network based on historical medical record statistics to obtain the risk probability prior distribution. A knowledge graph is constructed based on the electronic medical record text data, and the knowledge graph is transformed into medical graph embedding features using graph embedding.

3. The multi-knowledge-driven method for assessing the risk of pediatric perioperative respiratory anesthesia according to claim 1, characterized in that, The step of obtaining a knowledge-calibrated data-driven risk probability distribution based on the real-time multimodal data, the prior risk probability distribution, and the medical graph embedding features specifically includes: Cross-modal attention fusion is performed based on the real-time multimodal data and the medical graph embedding features to obtain a dynamic data-driven representation; Prior constraint optimization is performed based on the dynamic data-driven representation and the prior risk probability distribution to obtain a knowledge-calibrated data-driven risk probability distribution.

4. The multi-knowledge-driven method for assessing the risk of pediatric perioperative respiratory anesthesia according to claim 3, characterized in that, The real-time multimodal data includes vital signs parameters, medical imaging data, and physiological waveform signals; The step of performing cross-modal attention fusion based on the real-time multimodal data and the medical graph embedding features to obtain a dynamic data-driven representation specifically includes: The vital signs parameters, medical imaging data, and physiological waveform signals are input into a pre-trained encoder for feature extraction and unified representation to obtain a high-dimensional multimodal feature vector. Using the medical image embedding features as the query and the high-dimensional multimodal feature vector as the key and value, attention weights are calculated to obtain a dynamic data-driven representation: ； ； ； ； in, For dynamic data-driven representation, For attention mechanisms, It is a flexible maximum value function. For medical graph embedding features, It is a high-dimensional multimodal feature vector. To query the weight matrix, For query vector, The key weight matrix is... For key vectors, It is the transpose symbol. For value weight matrix, For value vectors, is the dimension of the key vector.

5. The multi-knowledge-driven method for assessing the risk of pediatric perioperative respiratory anesthesia according to claim 4, characterized in that, The step of optimizing prior constraints based on the dynamic data-driven representation and the prior risk probability distribution to obtain a knowledge-calibrated data-driven risk probability distribution specifically includes: The dynamic data-driven representation is input into the Transformer encoding module for high-order semantic modeling to obtain high-order features. The high-order features are then nonlinearly mapped through a multilayer perceptron to output the initial risk probability distribution. Construct a joint loss function based on the aforementioned prior distribution of risk probabilities: ； in, For the joint loss function, Let cross-entropy be the loss function. For the initial risk probability distribution, This is a true risk label. For adaptive dynamic weight parameters, Let KL divergence be the KL divergence. This represents the prior distribution of risk probability. Based on the initial risk probability distribution and the joint loss function, a knowledge-calibrated data-driven risk probability distribution is output.

6. The multi-knowledge-driven method for assessing the risk of pediatric perioperative respiratory anesthesia according to claim 3, characterized in that, The process of obtaining the final risk assessment distribution and structured assessment report based on the risk benchmark characterization and the data-driven risk probability distribution specifically includes: The risk benchmark characterization is processed to obtain the rule-based risk probability distribution; Based on the rule-based risk probability distribution and the data-driven risk probability distribution, the final risk assessment distribution is obtained; A structured assessment report is output based on the final risk assessment distribution.

7. The multi-knowledge-driven method for assessing the risk of pediatric perioperative respiratory anesthesia according to claim 6, characterized in that, The final risk assessment distribution is obtained based on the rule-based risk probability distribution and the data-driven risk probability distribution, specifically as follows: The rule-based risk probability distribution and the data-driven risk probability distribution are calculated using a dynamic gating mechanism formula to obtain the final risk assessment distribution: ； in, For the final risk assessment distribution, For dynamic gating coefficients, To drive risk probability distribution, This represents the probability distribution of rule-based risks.

8. A multi-dimensional knowledge-driven risk assessment system for pediatric perioperative respiratory anesthesia, characterized in that, The multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment system is used to implement the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method according to any one of claims 1-7, wherein the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment system includes: A multivariate clinical knowledge encoding module is used to acquire heterogeneous clinical knowledge and process the heterogeneous clinical knowledge to obtain risk benchmark representation, risk probability prior distribution and medical graph embedding features. The inference module is used to acquire real-time multimodal data of the target child and, based on the real-time multimodal data, the prior risk probability distribution, and the medical graph embedding features, obtain a knowledge-calibrated data-driven risk probability distribution. The decision fusion and expression module is used to obtain the final risk assessment distribution and structured assessment report based on the risk benchmark characterization and the data-driven risk probability distribution.

9. A terminal, characterized in that, The terminal includes: a memory, a processor, and a multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment program stored in the memory and executable on the processor. When the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment program is executed by the processor, it implements the steps of the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment program, which, when executed by a processor, implements the steps of the multi-knowledge-driven pediatric perioperative respiratory anesthesia risk assessment method as described in any one of claims 1-7.

Images