Preoperative assessment method, device and equipment for cardiovascular and cerebrovascular risks of non-cardiac surgery and storage medium

Abstract

The invention relates to a cardiovascular and cerebrovascular risk preoperative assessment method, device and equipment for a non-cardiac surgery and a storage medium. The method comprises the steps of performing feature extraction on multi-modal preoperative data of a target patient to obtain multi-modal data features; performing cross-modal interaction processing and global modeling and fusion on the multi-modal data features by a pre-operative feature fusion module based on cross-modal global cross attention to obtain cardiovascular and cerebrovascular risk features; and inputting the cardiovascular and cerebrovascular risk characteristics into a cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss, and outputting a cardiovascular and cerebrovascular risk preoperative assessment result based on the multi-dimensional reconstruction loss by the cardiovascular and cerebrovascular risk assessment module under the guidance of the comprehensive quantitative assessment of the to-be-predicted risk. According to the method, the efficiency and the accuracy of the pre-operative assessment of the cardiovascular and cerebrovascular risks of the non-cardiac surgery are improved, the extraction and fusion process of the multi-modal data features is optimized, and the pre-operative assessment capability and the generalization capability of the model on the cardiovascular and cerebrovascular risks are improved.

Description

Technical Field

[0001] This application belongs to the fields of medical artificial intelligence models and healthcare technologies, and particularly relates to a method, device, equipment, and storage medium for pre-operative assessment of cardiovascular and cerebrovascular risks in non-cardiac surgeries. Background Art

[0002] Traditional pre-operative assessment methods for cardiovascular and cerebrovascular risks in non-cardiac surgeries mainly collect patient information through means such as doctors' inquiries about medical history, physical examinations, and laboratory tests, and then conduct risk assessments in combination with clinical experience. However, this pre-operative risk assessment method often only focuses on aspects such as patients' physiological indicators or medical history, ignoring other important factors such as psychological indicators and social factors. At the same time, traditional pre-operative diagnosis of cardiovascular and cerebrovascular risks highly relies on doctors' clinical experience and subjective judgment when facing pre-operative modal features with complex structures and rich information. These pre-operative data not only have a wide variety but also have a high degree of heterogeneity and isomerism, making it difficult to comprehensively and accurately analyze them through a single processing method. Therefore, doctors need to spend a lot of time and energy on manual analysis and judgment, which not only greatly reduces the efficiency of diagnosis and assessment but also may lead to misjudgment or missed judgment due to human factors, affecting the formulation of surgical plans and the treatment effects of patients. Secondly, since pre-operative data usually includes various types such as high-definition medical images, gene sequencing information, and electronic medical record records, with the continuous growth of data volume and the increasing complexity of data, traditional feature modeling and analysis methods gradually show limitations in processing global feature extraction and correlation analysis between features.

[0003] With the continuous progress of medical technology and the growing clinical needs, medical artificial intelligence models play an increasingly important role in medical research and clinical practice. Multimodal medical models can efficiently integrate and fuse various types of medical data resources such as high-definition medical images, in-depth gene sequencing information, and detailed electronic medical record records, providing solid and comprehensive data support for the accurate diagnosis of diseases, the scientific prediction of risks, and the design of personalized treatment plans, and also providing strong support for the precision medicine and personalized intervention of cardiovascular and cerebrovascular diseases. However, due to technical limitations, disease progression, and other reasons, some pre-operative modal data have not been fully incorporated into the consideration scope of pre-operative risk assessment, which will have an adverse impact on the accuracy and reliability of multimodal medical models, making the model may have biases or deficiencies in predicting cerebrovascular risks. At the same time, due to the high hetero-isomerism of the collected multimodal data, it is difficult to directly use it to clearly characterize the pathological state. Therefore, how to optimize the use of pre-operative features and indicators without sacrificing model performance while making full use of existing pre-operative features and indicators has become an important direction in the current research on pre-operative assessment of cerebrovascular risks. Summary of the Invention

[0004] The present application provides a preoperative evaluation method, device, equipment, and storage medium for cardiovascular and cerebrovascular risks in non-cardiac surgery, aiming to solve at least one of the above technical problems in the prior art to a certain extent.

[0005] To solve the above problems, the present application provides the following technical solutions: A preoperative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery, including: Utilize a preoperative knowledge graph enhancement module based on a pre-trained large model to link the electronic medical record document of the target patient to the knowledge graph, construct the preoperative knowledge graph of the target patient, and use the pre-trained large model to perform risk prediction on the preoperative knowledge graph to obtain a comprehensive quantitative evaluation of the risk to be predicted; Respectively utilize a preoperative feature global interaction module based on feature aggregation graph convolution, a preoperative image feature extraction module based on multi-level low-rank matrix decomposition, and a preoperative time-series data feature extraction module based on parallel time-frequency coupling to extract features from the multimodal preoperative data of the target patient to obtain multimodal data features; wherein, the multimodal preoperative data includes electronic health record data, preoperative medical image data, and preoperative time-series data, and the multimodal data features include numerical and categorical preoperative electronic health data features, preoperative image data features, and preoperative time-series data features; Utilize a preoperative feature fusion module based on cross-modal global cross-attention to perform cross-modal interaction processing, global modeling, and fusion on the multimodal data features to obtain complete cardiovascular and cerebrovascular risk features; Input the cardiovascular and cerebrovascular risk features into a cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss. The cardiovascular and cerebrovascular risk assessment module is guided by the comprehensive quantitative evaluation of the risk to be predicted and outputs a preoperative evaluation result of the cardiovascular and cerebrovascular risks in non-cardiac surgery for the target patient based on the multi-dimensional reconstruction loss.

[0006] The technical solution adopted in the embodiment of the present application further includes: The process of utilizing a preoperative knowledge graph enhancement module based on a pre-trained large model to link the electronic medical record document of the target patient to the knowledge graph, construct the preoperative knowledge graph of the target patient, and use the pre-trained large model to perform risk prediction on the preoperative knowledge graph to obtain a comprehensive quantitative evaluation of the risk to be predicted is specifically as follows: Use a question-guided large language model to extract key information from the electronic medical record document, perform entity recognition on the key information, link each recognized entity to the corresponding node in the knowledge graph, and adjust the correlation weights of the corresponding nodes according to the entity type and contribution to obtain the preoperative knowledge graph of the target patient; The collaborative risk prediction module based on the pre-trained large model is conducive to transforming the preoperative knowledge graph into a semi-structured knowledge graph, inputting the semi-structured knowledge graph into the pre-trained large model, and using the pre-trained large model to quantitatively evaluate the correlation between the risk to be predicted and each relevant dimension, so as to obtain a comprehensive quantitative evaluation of the risk to be predicted.

[0007] The technical solutions adopted in the embodiments of this application further include: respectively using the preoperative feature global interaction module based on feature aggregation graph convolution, the preoperative image feature extraction module based on multi-level low-rank matrix decomposition, and the preoperative time-series data feature extraction module based on parallel time-frequency coupling to extract features from the multi-modal preoperative data of the target patient to obtain multi-modal data features, including: The preoperative feature global interaction module based on feature aggregation graph convolution is conducive to performing standardized preprocessing on the electronic health record data, and performing feature aggregation graph representation on the preprocessed electronic health record data to obtain numerical and categorical preoperative electronic health data features; among them, the feature extraction process of the preoperative feature global interaction module based on feature aggregation graph convolution includes: performing normal distribution standardization processing on the numerical data in the electronic health record data, performing one-hot encoding on the categorical data, representing both the numerical data and the categorical data in numerical form, and combining the numerical forms to generate a data vector x; dimensionalizing the data vector x into a square feature matrix H 0, and constructing an initial adjacency matrix in a graph convolution manner A 0 and a degree matrix D 0, so that the numerical data and the categorical data form a mutually related graph representation, and the final numerical and categorical preoperative electronic health data features are obtained through multiple layers of graph convolution.

[0008] The technical solutions adopted in the embodiments of this application further include: respectively using the preoperative feature global interaction module based on feature aggregation graph convolution, the preoperative image feature extraction module based on multi-level low-rank matrix decomposition, and the preoperative time-series data feature extraction module based on parallel time-frequency coupling to extract features from the multi-modal preoperative data of the target patient to obtain multi-modal data features, and further including: It is conducive to the pre-operative image feature extraction module based on multi-level low-rank matrix decomposition to extract features from the pre-operative medical image data, obtaining pre-operative image data features; wherein, the feature extraction process of the pre-operative image feature extraction module based on multi-level low-rank matrix decomposition includes: performing preliminary convolutional feature extraction on the pre-operative medical image data, and then using the multi-level matrix decomposition technology with low-rank constraint to gradually decompose the pre-operative medical image data into the matrix product of multiple low-rank representations, forming a new feature space, projecting the pre-operative medical image data into the new feature space, and reconstructing the pre-operative medical image data using the projection result and spatial information to obtain the final pre-operative image data features.

[0009] The technical solution adopted in the embodiment of the present application further includes: respectively using the pre-operative feature global interaction module based on feature aggregation graph convolution, the pre-operative image feature extraction module based on multi-level low-rank matrix decomposition, and the pre-operative time series data feature extraction module based on parallel time-frequency coupling to extract features from the multi-modal pre-operative data of the target patient, obtaining multi-modal data features, and further including: Using the pre-operative time series data feature extraction module based on parallel time-frequency coupling to perform multi-dimensional representation of the pre-operative time series data in the time domain and frequency domain, and extracting pre-operative time series data features with time-frequency consistency through periodic analysis and dominant frequency analysis; wherein, the pre-operative time series data feature extraction module based on parallel time-frequency coupling includes a time series analysis sub-module and a frequency domain analysis sub-module. The time series analysis sub-module captures the long-term trend and change law of the pre-operative time series data through a polynomial matrix to obtain time domain features; the frequency domain analysis sub-module uses frequency domain decomposition to transfer the pre-operative time series data to the frequency domain, samples to obtain the sequence data with the largest spectral amplitude, and transfers it back to the time domain to obtain frequency domain features; finally, the time domain features and frequency domain features are respectively passed through a linear layer and then concatenated to obtain the final pre-operative time series data features.

[0010] The technical solution adopted in the embodiment of the present application further includes: using the pre-operative feature fusion module based on cross-modal global cross-attention to perform cross-modal interaction processing, global modeling and fusion on the multi-modal data features, obtaining complete cardiovascular and cerebrovascular risk features, specifically: Taking the electronic health data features, pre-operative image data features or pre-operative time series data features as the baseline, performing cross-attention calculations with other modal data features respectively, enabling the multi-modal data features to obtain global interaction and similarity modeling, and then performing an independent self-attention calculation on the electronic health data features, pre-operative image data features and pre-operative time series data features respectively, and obtaining complete cardiovascular and cerebrovascular risk features through feature alignment.

[0011] The technical solution adopted in the embodiments of the present application further includes: inputting the cardiovascular and cerebrovascular risk characteristics into a cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss. The cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss is guided by the comprehensive quantitative assessment of the risk to be predicted, and outputs the preoperative assessment result of the cardiovascular and cerebrovascular risk of the target patient's non-cardiac surgery. Specifically: The cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss aggregates cardiovascular and cerebrovascular risk characteristics through a multi-layer perceptron and connects with an activation function. Under the guidance of the comprehensive quantitative assessment of the risk to be predicted, it outputs the cardiovascular and cerebrovascular risk prediction probability. Among them, the multi-dimensional reconstruction loss of the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss includes categorical cross-entropy loss, low-rank matrix factorization loss, and time-frequency domain coupling similarity loss. The categorical cross-entropy loss is used to measure the difference between the cardiovascular and cerebrovascular risk prediction probability output by the model and the true expert prior knowledge. The low-rank matrix factorization loss is used to ensure that the model can extract the supplementary information hidden in the preoperative image data features. The time-frequency domain coupling similarity loss is used to constrain and optimize the model during the backpropagation process.

[0012] Another technical solution adopted in the embodiments of the present application is: a preoperative assessment device for the cardiovascular and cerebrovascular risk of non-cardiac surgery, including: Preoperative knowledge graph enhancement module: used to link the electronic medical record document of the target patient to the knowledge graph, construct the preoperative knowledge graph of the target patient, and use a pre-trained large model to perform risk prediction on the preoperative knowledge graph to obtain a comprehensive quantitative assessment of the risk to be predicted; Multi-modal feature extraction module: used to respectively extract features from the multi-modal preoperative data of the target patient by using a preoperative feature global interaction module based on feature aggregation graph convolution, a preoperative image feature extraction module based on multi-level low-rank matrix factorization, and a preoperative time series data feature extraction module based on parallel time-frequency coupling to obtain multi-modal data features. Among them, the multi-modal preoperative data includes electronic health record data, preoperative medical image data, and preoperative time series data. The multi-modal data features include numerical and categorical preoperative electronic health data features, preoperative image data features, and preoperative time series data features; Preoperative feature fusion module: used to perform cross-modal interaction processing, global modeling, and fusion on the multi-modal data features by using cross-modal global cross-attention to obtain complete cardiovascular and cerebrovascular risk characteristics; Cardiovascular and cerebrovascular risk assessment module: used to receive the input cardiovascular and cerebrovascular risk characteristics, and under the guidance of the comprehensive quantitative assessment of the risk to be predicted, output the preoperative assessment result of the cardiovascular and cerebrovascular risk of the target patient's non-cardiac surgery by using multi-dimensional reconstruction loss.

[0013] Another technical solution adopted in the embodiments of the present application is: a device, the device includes a processor and a memory coupled to the processor, wherein, the memory stores program instructions for implementing the preoperative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery; the processor is configured to execute the program instructions stored in the memory to control the preoperative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery.

[0014] Another technical solution adopted in the embodiments of the present application is: a storage medium storing program instructions executable by a processor, the program instructions being used to execute the preoperative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery.

[0015] Compared with the prior art, the beneficial effects produced by the embodiments of the present application are as follows: The preoperative evaluation method, device, device, and storage medium for cardiovascular and cerebrovascular risks in non-cardiac surgery in the embodiments of the present application use a parallel multi-modal feature extraction framework as the network architecture, and introduce methods such as pre-trained large models, knowledge graphs, graph convolutional neural networks, and low-rank matrix decomposition to perform multi-dimensional processing on multi-modal preoperative data in non-cardiac surgery to obtain multi-modal data features, which is conducive to cross-modal global cross-attention to perform cross-modal interaction processing, global modeling, and fusion on multi-modal data features to obtain complete cardiovascular and cerebrovascular risk features, and the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss uses prior knowledge as a guide to perform preoperative evaluation of cardiovascular and cerebrovascular risks in non-cardiac surgery, which not only improves the efficiency and accuracy of preoperative evaluation of cardiovascular and cerebrovascular risks in non-cardiac surgery, but also optimizes the extraction and fusion process of multi-modal data features, enhances the model's preoperative evaluation ability and generalization ability for cardiovascular and cerebrovascular risks, and has good interpretability. In addition, the embodiments of the present application also help to optimize the allocation of medical resources, improve medical efficiency, reduce medical costs, and provide important references for clinical research such as exploring the causes and mechanisms of cardiovascular and cerebrovascular events. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flowchart of the preoperative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery according to the first embodiment of the present application; Figure 2 is a flowchart of the preoperative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery according to the second embodiment of the present application; Figure 3 is a schematic diagram of the knowledge graph structure of the electronic medical record document constructed by the preoperative knowledge graph enhancement module based on the pre-trained large model in the embodiment of the present application; Figure 4 is a general schematic diagram of the cross-modal global cross-attention architecture of the preoperative feature fusion module based on cross-modal global cross-attention in the embodiment of the present application; Figure 5Schematic structural diagram of a pre-operative evaluation device for cardio-cerebrovascular risks in non-cardiac surgery according to an embodiment of the present application; Figure 6 Schematic structural diagram of the device according to an embodiment of the present application; Figure 7 Schematic structural diagram of the storage medium according to an embodiment of the present application. Detailed implementation manners

[0017] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0018] The terms "first", "second", and "third" in the present application are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first", "second", and "third" may explicitly or implicitly include at least one of such features. In the description of the present application, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically and clearly defined. All directional indications (such as up, down, left, right, front, back...) in the embodiments of the present application are only used to explain the relative positional relationship and movement conditions between components in a specific posture (as shown in the drawings). If the specific posture changes, the directional indications will also change accordingly. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products, or devices.

[0019] Referring to "embodiment" herein means that a specific feature, structure, or characteristic described in conjunction with the embodiment may be included in at least one embodiment of the present application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein can be combined with other embodiments.

[0020] Specifically, please refer to Figure 1 , which is a flowchart of a pre-operative evaluation method for cardio-cerebrovascular risks in non-cardiac surgery according to the first embodiment of the present application. The pre-operative evaluation method for cardio-cerebrovascular risks in non-cardiac surgery according to the first embodiment of the present application includes the following steps: S100: The preoperative knowledge graph enhancement module based on the pre-trained large model is used to link the electronic medical record document of the target patient to the knowledge graph, construct the preoperative knowledge graph of the target patient, and use the pre-trained large model to perform risk prediction on the preoperative knowledge graph to obtain a comprehensive quantitative assessment of the risk to be predicted; S110: The preoperative feature global interaction module based on feature aggregation graph convolution, the preoperative image feature extraction module based on multi-level low-rank matrix decomposition, and the preoperative time series data feature extraction module based on parallel time-frequency coupling are respectively used to extract features from the multi-modal preoperative data of the target patient to obtain multi-modal data features; among them, the multi-modal preoperative data includes electronic health record data, preoperative medical image data, and preoperative time series data, and the extracted multi-modal data features include numerical and categorical preoperative electronic health data features, preoperative image data features, and preoperative time series data features; S120: The preoperative feature fusion module based on cross-modal global cross-attention is used to perform cross-modal interaction processing, global modeling, and fusion on the multi-modal data features to obtain complete cardiovascular and cerebrovascular risk features; S130: Input the cardiovascular and cerebrovascular risk features into the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss. The cardiovascular and cerebrovascular risk assessment module is guided by the comprehensive quantitative assessment of the risk to be predicted and outputs the preoperative assessment result of the cardiovascular and cerebrovascular risk of the non-cardiac surgery of the target patient based on the multi-dimensional reconstruction loss.

[0021] Please refer to Figure 2 , which is the flowchart of the preoperative assessment method for cardiovascular and cerebrovascular risks of non-cardiac surgery in the second embodiment of the present application. The preoperative assessment method for cardiovascular and cerebrovascular risks of non-cardiac surgery in the second embodiment of the present application includes the following steps: S200: Obtain multi-modal preoperative data such as the electronic medical record document, electronic health record data, preoperative medical image data, and preoperative time series data of the target patient; In this step, the electronic medical record document includes, but is not limited to, data such as the symptom manifestations, past medical history, medication details, and previous medical visit records of the target patient. The electronic health record data includes, but is not limited to, data such as the age, BMI (Body Mass Index), gender, blood type, and medical history of the target patient. The preoperative medical image data includes, but is not limited to, data such as the X-ray films of the target patient. The preoperative time series data includes, but is not limited to, data such as the electrocardiogram of the target patient.

[0022] S210: The preoperative knowledge graph enhancement module based on the pre-trained large model is used to extract key information from the electronic medical record document and link the key information to the knowledge graph to construct the preoperative knowledge graph of the target patient; In this step, the construction process of the preoperative knowledge graph is specifically as follows: First, a series of questions are used to guide large language models (LLMs) to extract key information from electronic medical record documents. Then, entity recognition is performed on the key information, and each recognized entity is linked to the corresponding node in the knowledge graph (KGs). The relevance weights of the corresponding nodes are adjusted according to the entity type and contribution to obtain the global relevance representation of the electronic medical record document, thereby constructing the preoperative knowledge graph of the target patient, providing strong support for subsequent cardiovascular and cerebrovascular risk assessment. Specifically, as Figure 3 shown, it is a schematic diagram of the knowledge graph structure of the electronic medical record document constructed by the preoperative knowledge graph enhancement module based on the pre-trained large model in the embodiment of the present application.

[0023] It can be understood that before using the pre-trained large model, the embodiment of the present application uses the preoperative knowledge graph enhancement module based on the pre-trained large model to construct a preoperative knowledge graph suitable for the input of the pre-trained large model, which can better adapt to the input requirements of the pre-trained large model and improve the efficiency and accuracy of model evaluation.

[0024] S220: Use the collaborative risk prediction module based on the pre-trained large model to deeply reconstruct the preoperative knowledge graph, transform it into a semi-structured knowledge graph, and input the semi-structured knowledge graph into the pre-trained large model. Use the pre-trained large model to quantitatively evaluate the correlation between the risk to be predicted and each relevant dimension to obtain a comprehensive quantitative evaluation of the risk to be predicted; In this step, through the collaborative risk prediction module based on the pre-trained large model, the constructed preoperative knowledge graph is deeply reconstructed. First, it is transformed into a more flexible and easy-to-process semi-structured knowledge graph to better adapt to the input requirements of the pre-trained large model. Then, a prompt template is used to guide the pre-trained large model to perform refined quantitative evaluation on the correlation between the risk to be predicted and each relevant dimension, obtain a comprehensive quantitative evaluation of the risk to be predicted, lay a solid foundation for subsequent cardiovascular and cerebrovascular risk assessment work, and improve the accuracy of model evaluation.

[0025] S230: Use the preoperative feature global interaction module based on feature aggregation graph convolution to perform standardized preprocessing on the electronic health record data, and perform feature aggregation graph representation on the preprocessed electronic health record data to obtain numerical and categorical preoperative electronic health data features; In this step, the feature extraction process of the preoperative feature global interaction module based on feature aggregation graph convolution specifically includes: First, numerical data such as age and BMI in the electronic health record data are subjected to normal distribution standardization processing. Then, one-hot encoding is performed on categorical data such as gender and blood type, and both numerical data and categorical data are represented in numerical form and combined to generate a data vector x. The data vector x is dimensionally expanded into a square feature matrixH 0, and construct an initial adjacency matrix in the form of graph convolution A 0 and a degree matrix D 0, so that numerical data and various types of data form an interconnected graph representation, and the final numerical and various types of preoperative electronic health data features are obtained through multi-layer graph convolution. The specific processing process can be expressed by the following formula: (1) where H i is the feature matrix of the i th layer, A i is the adjacency matrix of the i th layer, D i is the degree matrix of the i th layer, U i , V i , W i is the three weight matrices of the i th layer, is the sigmoid activation function, H n is the feature matrix of the last layer, is the feature vector finally output by the module.

[0026] It can be understood that in the embodiment of the present application, an initial adjacency matrix and a degree matrix are constructed with the idea of graph convolution, so that different data types form an interconnected graph representation, and the final numerical and various types of preoperative electronic health data features are obtained through multi-layer graph convolution, which can fully explore the hidden feature correlations in the electronic health record data, realize the global correlation modeling of the electronic health record data, provide richer and more reliable feature representations for subsequent feature fusion, and improve the accuracy of preoperative risk assessment.

[0027] S240: Use the preoperative image feature extraction module based on multi-level low-rank matrix decomposition to extract features from the preoperative medical image data to obtain preoperative image data features; In this step, the extraction process of the preoperative image data features specifically includes: first, perform preliminary convolutional feature extraction on the preoperative medical image data, and then use the multi-level matrix decomposition technology with low-rank constraints to gradually decompose the preoperative medical image data into the product of multiple matrices with low-rank representations to form a new feature space, project the preoperative medical image data into the new feature space, and reconstruct the preoperative medical image data using the projection result and spatial information to obtain the final preoperative image data features. Among them, the low-rank constraint formula is expressed as follows: (2) Among them, X is the module input image feature matrix, M i is the feature matrix after decomposition of the i th layer, Z i is the weight matrix of the matrix decomposition of the i th layer, X l is the feature matrix decomposed by the l th layer (i.e., the last layer), represents the matrix infinity norm, represents the matrix F norm.

[0028] It can be understood that in the embodiment of the present application, a low-rank constraint is added to the preoperative image feature extraction module based on multi-level low-rank matrix decomposition, which can efficiently extract the spatial semantic features of preoperative medical image data, effectively improve the feature extraction efficiency of preoperative medical image data, and ensure the stability and convergence of the model during the training process, providing support for subsequent multi-modal feature fusion and preoperative risk assessment.

[0029] S250: Use the preoperative time series data feature extraction module based on parallel time-frequency coupling to perform multi-dimensional representation of the preoperative time series data in the time domain and frequency domain, and extract the preoperative time series data features with time-frequency consistency through periodic analysis and dominant frequency analysis; In this step, when the preoperative time series data feature extraction module based on parallel time-frequency coupling performs feature extraction, it not only fully considers the periodic features contained in the preoperative time series data, but also deeply explores and reveals the potential frequency domain correlation features behind the preoperative time series data. The periodic features reflect the change rules and trends of medical data, and the frequency domain correlation features reflect the dynamic characteristics and internal mechanisms of the preoperative time series data, which can enhance the feature extraction efficiency of the preoperative time series data and provide a more solid and reliable data basis for preoperative risk prediction.

[0030] Furthermore, the preoperative time-series data feature extraction module based on parallel time-frequency coupling adopts a dual-domain parallel structure to process time-domain information and frequency-domain information in parallel, achieving a comprehensive and in-depth analysis of preoperative time-series data and greatly improving the accuracy and efficiency of feature extraction. Specifically, the preoperative time-series data feature extraction module based on parallel time-frequency coupling includes a time-series analysis sub-module and a frequency-domain analysis sub-module. Among them, the time-series analysis sub-module models the inherent overall trend of preoperative time-series data through a polynomial matrix, captures the long-term trend and variation law of preoperative time-series data, and obtains time-domain features; the frequency-domain analysis sub-module uses frequency-domain decomposition to transfer preoperative time-series data to the frequency domain, samples the sequence data with the largest spectral amplitude, and transfers it back to the time domain to obtain frequency-domain features; finally, the obtained time-domain features and frequency-domain features are respectively passed through a linear layer and then concatenated to obtain the final preoperative time-series data features. Specifically, the feature extraction process of the preoperative time-series data feature extraction module based on parallel time-frequency coupling can be expressed by the following formula:

[0031] (3) where i represents the i th parallel encoder, Q denotes the polynomial matrix, argTopK represents taking the top K maximum values K i (1) to K i (K) , Linear represents the linear layer, concat represents concatenating and aligning in dimensions, FFT represents the Fast Fourier Transform, IFFT represents the Inverse Fast Fourier Transform, A i (k) represents the kth frequency component obtained after FFT of the input of the i th encoder, M is the aggregated time-domain feature vector, and S is the aggregated frequency-domain feature vector.

[0032] S260: Facilitate the preoperative feature fusion module based on cross-modal global cross-attention to perform cross-modal interaction processing, global modeling, and fusion on numerical and categorical preoperative electronic health data features, preoperative image data features, and preoperative time-series data features to obtain complete cardiovascular and cerebrovascular risk features; In this step, in order to ensure that multimodal data features such as numerical and categorical preoperative electronic health data features, preoperative image data features, and preoperative time-series data features extracted in parallel can maintain a high degree of semantic consistency and can be fused into independent and complete cardiovascular and cerebrovascular risk features, this application introduces a preoperative feature fusion module based on cross-modal global cross-attention for cross-modal interaction processing, global modeling, and fusion. The multidimensional complementary strategy is used to model the similarity and consistency between different modal features, so that each modal data feature can fully consider the information of other modal data features, so as to strengthen the representational commonality between different modal data features, and thus exchange and integrate information globally to obtain more representative and discriminative cardiovascular and cerebrovascular risk features, enabling the model to strengthen the semantic consistency of multimodal data features, helping to overcome the fusion difficulties brought by the high heterogeneity and heterogeneity of preoperative data, not only deepening the model's understanding ability of complex multimodal preoperative data, but also significantly improving the interpretability of the model, enhancing the generalization ability of the model, further improving the accuracy and reliability of the preoperative assessment of cardiovascular and cerebrovascular risks in non-cardiac surgery, and providing more scientific and powerful decision-making support for clinicians. Specifically, as Figure 4 shown, it is the overall schematic diagram of the cross-modal global cross-attention architecture of the preoperative feature fusion module based on cross-modal global cross-attention in the embodiment of this application. Specifically, taking the electronic health data feature X, the preoperative image data feature Z, or the preoperative time-series data feature Y as the baseline, cross-attention calculations are respectively performed with other modal data features, so that the data features of the three modalities can obtain sufficient global interaction and similarity modeling. Then, independent self-attention calculations are respectively performed on the electronic health data feature, the preoperative image data feature, and the preoperative time-series data feature, and complete cardiovascular and cerebrovascular risk features are obtained through feature alignment. The cross-attention and self-attention calculation formulas are as follows: (4) where q represents the query vector, k represents the key vector, d k represents k the dimension of, v represents the value vector, and softmax represents the activation function.

[0033] S270: Input the cardiovascular and cerebrovascular risk features into the cardiovascular and cerebrovascular risk assessment module based on multidimensional reconstruction loss. Guided by the comprehensive quantitative assessment of the risk to be predicted, the preoperative assessment result of the cardiovascular and cerebrovascular risk of the target patient's non-cardiac surgery is output; In this step, the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss adopts a perceptron structure with multiple layers of neurons. It receives the cardiovascular and cerebrovascular risk features output by the preoperative feature fusion module based on cross-modal global cross-attention, and uses the comprehensive quantitative assessment generated by the collaborative risk prediction module based on the pre-trained large model to conduct cardiovascular and cerebrovascular risk assessment. Specifically, the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss aggregates the cardiovascular and cerebrovascular risk features through a multi-layer perceptron and connects them with an activation function. Under the guidance of the comprehensive quantitative assessment of the risk to be predicted, it outputs the cardiovascular and cerebrovascular risk prediction probability and outputs a probability value between 0 and 1, indicating the risk of cardiovascular and cerebrovascular events occurring. Among them, the multi-dimensional reconstruction loss of the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss consists of three parts: categorical cross-entropy loss, low-rank matrix decomposition loss, and time-frequency domain coupling similarity loss. The categorical cross-entropy loss is used to measure the difference between the cardiovascular and cerebrovascular risk prediction probability output by the model and the true expert prior knowledge. The low-rank matrix decomposition loss is used to ensure that the model can efficiently extract implicit supplementary information when processing preoperative image data features, improve the model stability and accelerate the model convergence. The time-frequency domain coupling similarity loss is used to further constrain and optimize the model during the backpropagation process to ensure the stability and consistency of the model during the reverse derivation process. Through the organic combination of these three, not only the guiding role of the prior knowledge of the pre-trained large model in model training is fully utilized, but also the model is guided to converge continuously during the training process through multi-dimensional loss optimization, and finally the accurate assessment of cardiovascular and cerebrovascular risks is achieved, further improving the preoperative assessment ability and generalization ability of the model for cardiovascular and cerebrovascular risks. Specifically, the calculation formula of the multi-dimensional reconstruction loss is as follows: (5) where is the time-frequency domain coupling similarity loss.

[0034] Based on the above, the pre-operative assessment method for cardiovascular and cerebrovascular risks in non-cardiac surgery according to the embodiments of the present application uses a parallel multi-modal feature extraction framework as the network architecture, and introduces methods such as pre-trained large models, knowledge graphs, graph convolutional neural networks, and low-rank matrix factorization to perform multi-dimensional processing on the multi-modal pre-operative data of non-cardiac surgery, obtaining multi-modal data features, and using cross-modal global cross-attention to perform cross-modal interaction processing, global modeling, and fusion on the multi-modal data features, obtaining complete cardiovascular and cerebrovascular risk features, and performing pre-operative assessment of cardiovascular and cerebrovascular risks in non-cardiac surgery based on the cardiovascular and cerebrovascular risk assessment module with multi-dimensional reconstruction loss under the guidance of prior knowledge. The embodiments of the present application integrate deep learning and prior knowledge of cardiovascular and cerebrovascular risks in the medical field, constructing stable and reliable cardiovascular and cerebrovascular risk features through multiple carefully designed network modules, not only improving the efficiency and accuracy of pre-operative assessment of cardiovascular and cerebrovascular risks in non-cardiac surgery, but also optimizing the extraction and fusion process of multi-modal data features, enhancing the model's pre-operative assessment ability and generalization ability for cardiovascular and cerebrovascular risks, and having good interpretability. In addition, the embodiments of the present application are helpful for optimizing the allocation of medical resources, improving medical efficiency, reducing medical costs, and providing important references for clinical research such as exploring the causes and mechanisms of cardiovascular and cerebrovascular events.

[0035] Please refer to Figure 5 , which is a schematic structural diagram of the device for pre-operative assessment of cardiovascular and cerebrovascular risks in non-cardiac surgery according to the embodiments of the present application. The device 40 for pre-operative assessment of cardiovascular and cerebrovascular risks in non-cardiac surgery according to the embodiments of the present application includes: Pre-operative knowledge graph enhancement module 41: used to link the electronic medical record document of the target patient to the knowledge graph, construct the pre-operative knowledge graph of the target patient, and use the pre-trained large model to perform risk prediction on the pre-operative knowledge graph, obtaining a comprehensive quantitative assessment of the risks to be predicted; Multi-modal feature extraction module 42: used to respectively use the pre-operative feature global interaction module based on feature aggregation graph convolution, the pre-operative image feature extraction module based on multi-level low-rank matrix factorization, and the pre-operative time series data feature extraction module based on parallel time-frequency coupling to extract features from the multi-modal pre-operative data of the target patient, obtaining multi-modal data features; wherein, the multi-modal pre-operative data includes electronic health record data, pre-operative medical image data, and pre-operative time series data, and the multi-modal data features include numerical and categorical pre-operative electronic health data features, pre-operative image data features, and pre-operative time series data features; Pre-operative feature fusion module 43: used to perform cross-modal interaction processing, global modeling, and fusion on the multi-modal data features using cross-modal global cross-attention, obtaining complete cardiovascular and cerebrovascular risk features; Cardiovascular risk assessment module 44: configured to receive input cardiovascular risk characteristics, and guided by the comprehensive quantitative assessment of the risk to be predicted, output a preoperative assessment result of the cardiovascular risk of the non-cardiac surgery for the target patient by using a multi-dimensional reconstruction loss.

[0036] It should be noted that for the information interaction, execution process, etc. between the above-mentioned devices / units, since they are based on the same concept as the method embodiments of the present application, their specific functions and the technical effects brought thereby can be specifically referred to in the method embodiment part, and will not be elaborated here.

[0037] The device provided by the embodiments of the present application can be applied in the foregoing method embodiments. For details, please refer to the description of the foregoing method embodiments and will not be elaborated here.

[0038] Please refer to Figure 6 , which is a schematic structural diagram of the device according to the embodiments of the present application. The device 50 includes: A memory 51 storing executable program instructions; A processor 52 connected to the memory 51; The processor 52 is configured to call the executable program instructions stored in the memory 51 and execute the following steps: facilitating the preoperative knowledge graph enhancement module based on the pre-trained large model to link the electronic medical record document of the target patient to the knowledge graph, constructing the preoperative knowledge graph of the target patient, and using the pre-trained large model to perform risk prediction on the preoperative knowledge graph to obtain a comprehensive quantitative assessment of the risk to be predicted; respectively facilitating the preoperative feature global interaction module based on feature aggregation graph convolution, the preoperative image feature extraction module based on multi-level low-rank matrix decomposition, and the preoperative time series data feature extraction module based on parallel time-frequency coupling to extract features from the multi-modal preoperative data of the target patient to obtain multi-modal data features; wherein, the multi-modal preoperative data includes electronic health record data, preoperative medical image data, and preoperative time series data, and the multi-modal data features include numerical and categorical preoperative electronic health data features, preoperative image data features, and preoperative time series data features; facilitating the preoperative feature fusion module based on cross-modal global cross-attention to perform cross-modal interaction processing, global modeling, and fusion on the multi-modal data features to obtain complete cardiovascular risk characteristics; inputting the cardiovascular risk characteristics into the cardiovascular risk assessment module based on multi-dimensional reconstruction loss, and the cardiovascular risk assessment module is guided by the comprehensive quantitative assessment of the risk to be predicted and outputs a preoperative assessment result of the cardiovascular risk of the non-cardiac surgery for the target patient based on the multi-dimensional reconstruction loss.

[0039] Among them, the processor 52 can also be referred to as a CPU (Central Processing Unit). The processor 52 may be an integrated circuit chip with the ability to process signals. The processor 52 can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0040] Please refer to Figure 7, which is a schematic structural diagram of the storage medium according to an embodiment of the present application. The storage medium according to the embodiment of the present application stores program instructions 61 that can implement the following steps: facilitating the preoperative knowledge graph enhancement module based on the pre-trained large model to link the electronic medical record document of the target patient to the knowledge graph, constructing the preoperative knowledge graph of the target patient, and using the pre-trained large model to perform risk prediction on the preoperative knowledge graph to obtain a comprehensive quantitative assessment of the risk to be predicted; respectively facilitating the preoperative feature global interaction module based on feature aggregation graph convolution, the preoperative image feature extraction module based on multi-level low-rank matrix decomposition, and the preoperative time series data feature extraction module based on parallel time-frequency coupling to extract features from the multi-modal preoperative data of the target patient to obtain multi-modal data features; wherein, the multi-modal preoperative data includes electronic health record data, preoperative medical image data, and preoperative time series data, and the multi-modal data features include numerical and categorical preoperative electronic health data features, preoperative image data features, and preoperative time series data features; facilitating the preoperative feature fusion module based on cross-modal global cross-attention to perform cross-modal interaction processing, global modeling, and fusion on the multi-modal data features to obtain complete cardiovascular and cerebrovascular risk features; inputting the cardiovascular and cerebrovascular risk features into the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss, and the cardiovascular and cerebrovascular risk assessment module is guided by the comprehensive quantitative assessment of the risk to be predicted and outputs the preoperative assessment result of the cardiovascular and cerebrovascular risk of the non-cardiac surgery of the target patient based on multi-dimensional reconstruction loss. Among them, the program instructions 61 can be stored in the above storage medium in the form of a software product, including several instructions for causing a device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods in various embodiments of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program instructions, or terminal devices such as computers, servers, mobile phones, and tablets. Among them, the server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0041] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be indirect couplings or communication connections through some interfaces, devices, or units, and can be in electrical, mechanical, or other forms.

[0042] In addition, each functional unit in various embodiments of the present application can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units. The above is only the implementation mode of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, is equally included in the patent protection scope of the present application.

Claims

1. A preoperative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery, characterized in that, Including: The preoperative knowledge graph enhancement module based on a pre-trained large model facilitates linking the electronic medical record document of the target patient to the knowledge graph, constructing the preoperative knowledge graph of the target patient, and using the pre-trained large model to perform risk prediction on the preoperative knowledge graph to obtain a comprehensive quantitative assessment of the risk to be predicted. The preoperative feature global interaction module based on feature aggregation graph convolution, the preoperative image feature extraction module based on multi-level low-rank matrix factorization, and the preoperative time series data feature extraction module based on parallel time-frequency coupling respectively facilitate feature extraction from the multi-modal preoperative data of the target patient to obtain multi-modal data features. Among them, the multi-modal preoperative data includes electronic health record data, preoperative medical image data, and preoperative time series data, and the multi-modal data features include numerical and categorical preoperative electronic health data features, preoperative image data features, and preoperative time series data features. The preoperative feature fusion module based on cross-modal global cross-attention facilitates cross-modal interaction processing, global modeling, and fusion of the multi-modal data features to obtain complete cardiovascular and cerebrovascular risk features. Input the cardiovascular and cerebrovascular risk features into the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss. The cardiovascular and cerebrovascular risk assessment module is guided by the comprehensive quantitative assessment of the risk to be predicted and outputs the preoperative cardiovascular and cerebrovascular risk assessment results for the non-cardiac surgery of the target patient based on multi-dimensional reconstruction loss.

2. The preoperative cardiovascular risk assessment method for non-cardiac surgery according to claim 1, wherein The preoperative knowledge graph enhancement module based on a pre-trained large model facilitates linking the electronic medical record document of the target patient to the knowledge graph, constructing the preoperative knowledge graph of the target patient, and using the pre-trained large model to perform risk prediction on the preoperative knowledge graph to obtain a comprehensive quantitative assessment of the risk to be predicted. Specifically: Use a question-guided large language model to extract key information from the electronic medical record document, perform entity recognition on the key information, link each recognized entity to the corresponding node in the knowledge graph, and adjust the relevance weights of the corresponding nodes according to the entity type and contribution to obtain the preoperative knowledge graph of the target patient. The collaborative risk prediction module based on a pre-trained large model facilitates converting the preoperative knowledge graph into a semi-structured knowledge graph, inputting the semi-structured knowledge graph into the pre-trained large model, and using the pre-trained large model to quantitatively evaluate the relevance between the risk to be predicted and each relevant dimension to obtain a comprehensive quantitative assessment of the risk to be predicted.

3. The preoperative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery according to claim 2, wherein The preoperative feature global interaction module based on feature aggregation graph convolution, the preoperative image feature extraction module based on multi-level low-rank matrix factorization, and the preoperative time series data feature extraction module based on parallel time-frequency coupling respectively facilitate feature extraction from the multi-modal preoperative data of the target patient to obtain multi-modal data features, including: It is beneficial for the preoperative feature global interaction module based on feature aggregation graph convolution to perform standardized preprocessing on the electronic health record data, and perform feature aggregation graph representation on the preprocessed electronic health record data to obtain numerical and categorical preoperative electronic health data features; among them, the feature extraction process of the preoperative feature global interaction module based on feature aggregation graph convolution includes: performing normal distribution standardization processing on the numerical data in the electronic health record data, and performing one-hot encoding on the categorical data, representing both the numerical data and the categorical data in numerical form, and combining the numerical forms to generate a data vector x; dimensionality expanding the data vector x into a square feature matrix H 0, and constructing an initial adjacency matrix in the form of graph convolution A 0 and a degree matrix D 0, enabling the numerical data and the categorical data to form an interrelated graph representation, and obtaining the final numerical and categorical preoperative electronic health data features through multiple layers of graph convolution.

4. The pre-operative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery according to claim 3, wherein The preoperative feature global interaction module based on feature aggregation graph convolution, the preoperative image feature extraction module based on multi-level low-rank matrix factorization, and the preoperative time series data feature extraction module based on parallel time-frequency coupling respectively facilitate feature extraction from the multi-modal preoperative data of the target patient to obtain multi-modal data features, further including: Facilitate the pre-operative image feature extraction module based on multi-level low-rank matrix decomposition to extract features from the pre-operative medical image data, obtaining pre-operative image data features; wherein, the feature extraction process of the pre-operative image feature extraction module based on multi-level low-rank matrix decomposition includes: performing preliminary convolutional feature extraction on the pre-operative medical image data, and then using the multi-level matrix decomposition technology with low-rank constraint to gradually decompose the pre-operative medical image data into the matrix product of multiple low-rank representations, forming a new feature space, projecting the pre-operative medical image data into the new feature space, and reconstructing the pre-operative medical image data using the projection result and spatial information to obtain the final pre-operative image data features.

5. The preoperative cardiovascular risk assessment method for non-cardiac surgery according to claim 4, wherein The above-mentioned steps of respectively facilitating the pre-operative feature global interaction module based on feature aggregation graph convolution, the pre-operative image feature extraction module based on multi-level low-rank matrix decomposition, and the pre-operative time series data feature extraction module based on parallel time-frequency coupling to extract features from the multi-modal pre-operative data of the target patient, obtaining multi-modal data features, further include: Using the pre-operative time series data feature extraction module based on parallel time-frequency coupling to perform multi-dimensional representations in the time domain and frequency domain on the pre-operative time series data, and extracting pre-operative time series data features with time-frequency consistency through periodic analysis and dominant frequency analysis; wherein, the pre-operative time series data feature extraction module based on parallel time-frequency coupling includes a time series analysis sub-module and a frequency domain analysis sub-module. The time series analysis sub-module captures the long-term trend and change law of the pre-operative time series data through a polynomial matrix to obtain time domain features; the frequency domain analysis sub-module uses frequency domain decomposition to transfer the pre-operative time series data to the frequency domain, samples to obtain the sequence data with the largest spectral amplitude, and transfers it back to the time domain to obtain frequency domain features; finally, the time domain features and frequency domain features are respectively passed through a linear layer and then concatenated to obtain the final pre-operative time series data features.

6. The pre-operative cardiovascular risk assessment method for non-cardiac surgery according to any one of claims 1 to 5, characterized in that, The above-mentioned steps of facilitating the pre-operative feature fusion module based on cross-modal global cross-attention to perform cross-modal interaction processing, global modeling, and fusion on the multi-modal data features, obtaining complete cardiovascular and cerebrovascular risk features, specifically: Taking the electronic health data features, pre-operative image data features, or pre-operative time series data features as the baseline, performing cross-attention calculations with other modal data features respectively, enabling the multi-modal data features to obtain global interaction and similarity modeling, and then performing an independent self-attention calculation on the electronic health data features, pre-operative image data features, and pre-operative time series data features respectively, and obtaining complete cardiovascular and cerebrovascular risk features through feature alignment.

7. The preoperative evaluation method for cardiovascular and cerebrovascular risks in non-cardiac surgery according to claim 6, characterized in that, The above-mentioned steps of inputting the cardiovascular and cerebrovascular risk features into the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss, and the cardiovascular and cerebrovascular risk assessment module based on multi-dimensional reconstruction loss is guided by the comprehensive quantitative assessment of the risk to be predicted, and outputting the pre-operative assessment result of the cardiovascular and cerebrovascular risk of the non-cardiac surgery of the target patient, specifically: The cardiovascular risk assessment module based on multi-dimensional reconstruction loss aggregates cardiovascular risk features through a multi-layer perceptron and connects with an activation function, and outputs the cardiovascular risk prediction probability under the guidance of the comprehensive quantitative assessment of the risk to be predicted; wherein, the multi-dimensional reconstruction loss of the cardiovascular risk assessment module based on multi-dimensional reconstruction loss includes categorical cross-entropy loss, low-rank matrix factorization loss, and time-frequency domain coupling similarity loss. The categorical cross-entropy loss is used to measure the difference between the cardiovascular risk prediction probability output by the model and the true expert prior knowledge. The low-rank matrix factorization loss is used to ensure that the model can extract the supplementary information hidden in the preoperative image data features. The time-frequency domain coupling similarity loss is used to constrain and optimize the model during the backpropagation process.

8. A preoperative evaluation device for cardiovascular and cerebrovascular risks in non-cardiac surgery, characterized in that, It includes: Preoperative knowledge graph enhancement module: used to link the electronic medical record document of the target patient to the knowledge graph, construct the preoperative knowledge graph of the target patient, and use a pre-trained large model to perform risk prediction on the preoperative knowledge graph to obtain a comprehensive quantitative assessment of the risk to be predicted; Multi-modal feature extraction module: used to respectively extract features from the multi-modal preoperative data of the target patient by using a preoperative feature global interaction module based on feature aggregation graph convolution, a preoperative image feature extraction module based on multi-level low-rank matrix factorization, and a preoperative time series data feature extraction module based on parallel time-frequency coupling to obtain multi-modal data features; wherein, the multi-modal preoperative data includes electronic health record data, preoperative medical image data, and preoperative time series data, and the multi-modal data features include numerical and categorical preoperative electronic health data features, preoperative image data features, and preoperative time series data features; Preoperative feature fusion module: used to perform cross-modal interaction processing, global modeling, and fusion on the multi-modal data features by using cross-modal global cross-attention to obtain complete cardiovascular risk features; Cardiovascular risk assessment module: used to receive the input cardiovascular risk features and, guided by the comprehensive quantitative assessment of the risk to be predicted, output the preoperative assessment result of the cardiovascular risk of the non-cardiac surgery of the target patient by using multi-dimensional reconstruction loss.

9. An electronic device, characterized in that, The device includes a processor and a memory coupled to the processor, wherein, The memory stores program instructions for implementing the preoperative cardiovascular risk assessment method for non-cardiac surgery according to any one of claims 1-7; The processor is used to execute the program instructions stored in the memory to control the preoperative cardiovascular risk assessment method for non-cardiac surgery.

10. A storage medium, characterized in that, Stores program instructions that can be run by a processor, and the program instructions are used to execute the preoperative cardiovascular risk assessment method for non-cardiac surgery according to any one of claims 1 to 7.

Images