Coronary artery bypass transplantation perioperative period cerebral apoplexy risk assessment method

Abstract

The invention relates to the technical field of health management, and discloses a coronary artery bypass transplantation perioperative period cerebral apoplexy risk assessment method, which comprises the following steps: acquiring multiple groups of training physiological data, each group of training physiological data belonging to the same patient and including current physiological data and historical physiological data; based on the multiple groups of training physiological data, training a risk assessment model to obtain a target risk assessment model; for each to-be-evaluated patient, data extraction is carried out on original physiological data of the to-be-evaluated patient to obtain multiple pieces of physiological index data, and the original physiological data comprises current physiological data and historical physiological data; and performing evaluation based on the physiological index data by adopting the target risk evaluation model to obtain an evaluation result. By training the risk assessment model, cerebral apoplexy risk assessment can be accurately carried out, the scientificity and reliability of the assessment result are improved, medical intervention based on the assessment result is facilitated, support is provided for perioperative period management, and the risk of cerebral apoplexy in the perioperative period is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of health management, and in particular to a method for assessing the risk of perioperative stroke during coronary artery bypass grafting. Background Art

[0002] Coronary artery bypass grafting (CABG) is one of the most effective treatments for coronary heart disease. Perioperative stroke is the most serious complication of CABG, with an incidence of 1.2% to 6%. Once it occurs, it can lead to extremely high rates of mortality and disability. Preventing this disease from occurring is crucial. Therefore, early identification of individuals at high risk for perioperative stroke is crucial, requiring precise risk assessment.

[0003] Currently, a variety of models are used to assess the risk of perioperative complications in cardiac surgery, such as the RCRI model (Revised Cardiac Risk Index), the MICA model (Myocardial Infarction or Cardiac Arrest), and the ACS-SRC model (American College of Surgeons surgical risk calculator). However, none of these models consider stroke as an independent endpoint for risk assessment, lacking specificity. Furthermore, existing assessment methods rely on real-time physiological data, but most patients with coronary artery disease have a long course of disease, resulting in data used for risk assessment that cannot accurately reflect the patient's condition, making it difficult to accurately assess the risk of complications. Summary of the Invention

[0004] In view of this, the present invention provides a method for assessing the risk of perioperative stroke during coronary artery bypass grafting to address the problem that existing assessment methods lack specificity and accuracy.

[0005] In a first aspect, the present invention provides a method for assessing the risk of perioperative stroke during coronary artery bypass grafting, the method comprising:

[0006] Acquire multiple sets of training physiological data, each set of training physiological data belongs to the same patient, including the patient's current physiological data and historical physiological data;

[0007] Based on multiple sets of training physiological data, the risk assessment model is trained to obtain a target risk assessment model;

[0008] For each patient to be evaluated, data extraction is performed on the original physiological data of the patient to be evaluated to obtain multiple physiological index data, the original physiological data including the current physiological data and historical physiological data of the patient to be evaluated, and the patient to be evaluated is in the perioperative period of coronary artery bypass grafting;

[0009] The target risk assessment model is used to perform assessment based on physiological indicator data to obtain the assessment results of the patients to be assessed.

[0010] The perioperative stroke risk assessment method for coronary artery bypass grafting provided by an embodiment of the present invention can more comprehensively reflect the patient's condition by collecting the patient's current and historical physiological data during the perioperative period, providing rich and comprehensive basic data for subsequent model training. Furthermore, training is performed based on multiple sets of training physiological data, so that the trained target risk assessment model can accurately perform stroke risk assessment, thereby improving the scientific nature and reliability of the assessment results. After extracting multiple physiological indicator data of the patient to be assessed, the target risk assessment model is used for assessment, which can generate accurate assessment results based on the specific situation of the patient to be assessed, facilitate medical intervention based on the assessment results, provide support for perioperative management, and reduce the risk of perioperative stroke.

[0011] In an optional embodiment, a risk assessment model is trained based on multiple sets of training physiological data to obtain a target risk assessment model, including:

[0012] Processing is performed based on multiple sets of training physiological data to obtain multiple sets of pre-training physiological data;

[0013] Construct test sets and validation sets based on multiple sets of pre-training physiological data;

[0014] The risk assessment model is adopted, trained based on the test set, and the training process is monitored based on the validation set to obtain the target risk assessment model.

[0015] The method for assessing perioperative stroke risk during coronary artery bypass grafting provided by an embodiment of the present invention preprocesses the original training data to generate data suitable for model training, divides the pre-trained physiological data into a test set and a validation set. The test set is used to train the model, and the validation set is used to monitor the training process in real time to prevent overfitting or underfitting, thereby helping to improve the evaluation accuracy of the model.

[0016] In an optional embodiment, processing is performed based on multiple sets of training physiological data to obtain multiple sets of pre-training physiological data, including:

[0017] Preprocessing each set of training physiological data to obtain preprocessed training physiological data;

[0018] Each set of pre-processed training physiological data is divided according to the format to obtain historical time series data, evaluation state data and static feature data;

[0019] Each group of pre-processed training physiological data is labeled to obtain pre-training physiological data.

[0020] The perioperative stroke risk assessment method for coronary artery bypass grafting provided by an embodiment of the present invention improves data reliability and consistency through data preprocessing, reduces interference with model training, and then divides each set of preprocessed training physiological data according to format. Different types of divisions enable the model to understand the patient's condition from multiple perspectives, thereby performing a more comprehensive assessment. Each set of preprocessed training physiological data is labeled for supervising model training, thereby improving the training efficiency and accuracy of the model.

[0021] In an optional embodiment, a risk assessment model is used, trained based on a test set, and the training process is monitored based on a validation set to obtain a target risk assessment model, including:

[0022] Using the long short-term memory network in the risk assessment model, feature extraction is performed based on the historical time series data in any set of pre-trained physiological data in the test set to obtain a time series feature vector. Using the deep neural network in the risk assessment model, feature extraction is performed based on the assessment state data and static feature data in the pre-trained physiological data to obtain a static feature vector.

[0023] Fuse the time series feature vector and the static feature vector to obtain the training feature vector;

[0024] The multi-task output layer in the risk assessment model is used to perform evaluation based on the training feature vector to obtain training evaluation results, which include stroke risk prediction values and risk groups.

[0025] Calculate the training loss based on the training evaluation results and annotations of the pre-trained physiological data;

[0026] Based on the training loss, the risk assessment model is optimized through error back propagation;

[0027] Using the optimized risk assessment model, model validation is performed based on the validation set to obtain training indicators;

[0028] When the training indicators do not meet the training stop conditions, return to the long short-term memory network in the risk assessment model, perform feature extraction based on the historical time series data in any set of pre-trained physiological data in the test set, and obtain a time series feature vector. Use the deep neural network in the risk assessment model to perform feature extraction based on the assessment state data and static feature data in the pre-trained physiological data to obtain a static feature vector. Retrain the model until the training indicators meet the training stop conditions, and use the risk assessment model obtained by the last optimization as the target risk assessment model.

[0029] The perioperative stroke risk assessment method for coronary artery bypass grafting provided by an embodiment of the present invention uses a long short-term memory network and a deep neural network to extract features from different types of data, obtain two feature vectors, and fuse them to generate a training feature vector, so that the model can simultaneously capture the patient's dynamic change trend and static feature information, thereby improving the ability to fully understand the stroke risk. A multi-task output layer is used to perform evaluation based on the training feature vector to obtain training evaluation results including stroke risk prediction values and risk grouping. Then, based on the training evaluation results and data annotation, the training loss is calculated to optimize the model. Through a rigorous verification and iteration mechanism, it is ensured that the target risk assessment model finally obtained has high reliability and stability.

[0030] In an optional embodiment, the training loss is calculated based on the evaluation results and annotations of the pre-trained physiological data, including:

[0031] Determine the least squares loss based on the stroke risk prediction value in the training evaluation results and the actual stroke risk prediction value in the annotation;

[0032] Determine the cross entropy loss based on the risk grouping in the training evaluation results and the actual risk grouping in the annotations;

[0033] Get the weights corresponding to the least squares loss and cross entropy loss respectively;

[0034] Determine the training loss based on the least squares loss and its weights and the cross entropy loss and its weights.

[0035] The method for assessing the risk of perioperative stroke during coronary artery bypass grafting provided by the embodiment of the present invention combines two loss functions and their weights. The model can be optimized globally, effectively reducing the error accumulation problem that may be caused by a single loss function, improving the stability of the model, and meeting the multi-dimensional needs of perioperative stroke risk assessment.

[0036] In an optional embodiment, for each patient to be evaluated, data extraction is performed on the original physiological data of the patient to be evaluated to obtain multiple physiological indicator data, including:

[0037] For the numerical indicator data in the original physiological data, a large language model is used to extract and structure the data to obtain multiple first indicator data;

[0038] For the text data in the original physiological data, a large language model is used to extract indicators and structure them to obtain multiple second indicator data;

[0039] The plurality of first indicator data and the plurality of second indicator data are used as a plurality of physiological indicator data.

[0040] The method for assessing the risk of perioperative stroke during coronary artery bypass grafting provided by an embodiment of the present invention extracts data from both numerical and textual dimensions through a large language model, ensuring comprehensive capture of the patient's physiological state and helping to improve the accuracy of model assessment.

[0041] In an optional embodiment, the method further includes:

[0042] Based on the evaluation results of the patient to be evaluated, a medical treatment recommendation for the patient to be evaluated is determined.

[0043] The method for assessing the risk of perioperative stroke during coronary artery bypass grafting provided by an embodiment of the present invention helps to carry out medical intervention, provide support for perioperative management, and reduce the risk of perioperative stroke by determining corresponding medical treatment recommendations based on the assessment results.

[0044] In a second aspect, the present invention provides a device for assessing the risk of perioperative stroke during coronary artery bypass grafting, the device comprising:

[0045] An acquisition module is used to acquire multiple sets of training physiological data, each set of training physiological data belongs to the same patient, including the patient's current physiological data and historical physiological data;

[0046] A training module is used to train the risk assessment model based on multiple sets of training physiological data to obtain a target risk assessment model;

[0047] an extraction module, configured to extract, for each patient to be evaluated, original physiological data of the patient to be evaluated to obtain a plurality of physiological indicator data, wherein the original physiological data includes current physiological data and historical physiological data of the patient to be evaluated, and the patient to be evaluated is in the perioperative period of coronary artery bypass grafting;

[0048] The evaluation module is used to adopt a target risk assessment model to perform an evaluation based on physiological indicator data to obtain an evaluation result of the patient to be evaluated.

[0049] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to thereby execute the method for assessing perioperative stroke risk during coronary artery bypass grafting according to the first aspect or any corresponding embodiment thereof.

[0050] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method for assessing perioperative stroke risk during coronary artery bypass grafting according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 is a flow chart of a method for assessing perioperative stroke risk during coronary artery bypass grafting according to an embodiment of the present invention;

[0053] Figure 2 is a schematic diagram of a calibration curve of a target risk assessment model according to an embodiment of the present invention;

[0054] Figure 3 is a schematic diagram of model performance of a target risk assessment model according to an embodiment of the present invention;

[0055] Figure 4 is a structural block diagram of a device for assessing perioperative stroke risk in coronary artery bypass grafting according to an embodiment of the present invention;

[0056] Figure 5 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0058] Existing perioperative stroke risk assessment methods lack specificity and accuracy. The perioperative stroke risk assessment method for coronary artery bypass grafting provided by the embodiments of the present invention can accurately perform stroke risk assessment by training a risk assessment model, thereby improving the scientific nature and reliability of the assessment results, facilitating medical intervention based on the assessment results, providing support for perioperative management, and reducing the risk of perioperative stroke.

[0059] According to an embodiment of the present invention, an embodiment of a method for assessing perioperative stroke risk during coronary artery bypass grafting is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0060] In this embodiment, a method for assessing the risk of stroke during the perioperative period of coronary artery bypass grafting is provided, which can be used on a terminal, such as a computer, Figure 1 FIG. 1 is a flow chart of a method for assessing perioperative stroke risk in coronary artery bypass grafting according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0061] Step S101, obtain multiple sets of training physiological data, each set of training physiological data belongs to the same patient, including the patient's current physiological data and historical physiological data. Specifically, the embodiment of the present invention is used to assess the risk of perioperative stroke, which is an imaging-confirmed cerebral ischemic infarction event that occurs during the patient's postoperative hospitalization or within 14 days after surgery, and is defined as a new large-area, focal or lacunar cerebral infarction diagnosed by head CT (Computed Tomography) or head MRI (Magnetic Resonance Imaging). Multiple sets of training physiological data are obtained using the medical records and other clinical data of the coronary artery bypass grafting patients included in the study. Each set of training physiological data comes from the same patient, including both the patient's physiological data during the evaluation and the patient's physiological data over a period of time, and is used to comprehensively represent the patient's physical condition. Among them, the training physiological data specifically include: physical examination data (such as patient gender, age, chief complaint, onset time, admission time, operation time, discharge time, length of hospital stay, past medical history, etc.), clinical manifestation data (such as body temperature, heart rate, blood pressure, height, weight, body mass index, etc.), preoperative laboratory test data (such as blood routine, blood biochemistry, myocardial enzymes, coagulation, blood gas analysis, etc.), imaging examination data (such as head CT, echocardiography, carotid artery ultrasound, vertebral artery ultrasound, etc.). Optionally, for the data that will change over time in the above data, the time corresponding to the indicator data must be obtained at the same time to form a time series. For example, if the patient's gender does not change over time, there is no need to obtain the time. The patient's body temperature may be different at different times, so the corresponding time needs to be obtained when obtaining the body temperature.

[0062] Step S102: Training the risk assessment model based on the multiple sets of training physiological data to obtain a target risk assessment model. Specifically, by obtaining multiple sets of training physiological data and using them as a basis for model training, the model learns the characteristics and patterns related to perioperative stroke, thereby obtaining a target risk assessment model that can accurately assess the risk.

[0063] In step S103, for each patient to be evaluated, the original physiological data of the patient to be evaluated is extracted to obtain a plurality of physiological indicator data. The original physiological data includes the current physiological data and historical physiological data of the patient to be evaluated, and the patient to be evaluated is in the perioperative period of coronary artery bypass grafting. Specifically, for each patient to be evaluated who is in the perioperative period of coronary artery bypass grafting, the original physiological data of the patient is obtained. The original physiological data can refer to the training physiological data, covering various information such as physical examination, clinical manifestations, preoperative laboratory tests and imaging examinations. Since the format, type and content of the above physiological data are relatively complex, key physiological indicator data need to be extracted from them and converted into a format that can be recognized and processed by the model so that the model can perform risk assessment.

[0064] In step S104, a target risk assessment model is used to evaluate the patient's physiological indicator data, generating an assessment result. Specifically, the extracted physiological indicator data of the patient is input into the target risk assessment model. The model analyzes and calculates the data based on the knowledge and patterns learned during training, and outputs an assessment result of the patient's perioperative stroke risk, providing support for medical intervention to reduce the incidence of perioperative stroke.

[0065] The perioperative stroke risk assessment method for coronary artery bypass grafting provided by an embodiment of the present invention can more comprehensively reflect the patient's condition by collecting the patient's current and historical physiological data during the perioperative period, providing rich and comprehensive basic data for subsequent model training. Furthermore, training is performed based on multiple sets of training physiological data, so that the trained target risk assessment model can accurately perform stroke risk assessment, thereby improving the scientific nature and reliability of the assessment results. After extracting multiple physiological indicator data of the patient to be assessed, the target risk assessment model is used for assessment, which can generate accurate assessment results based on the specific situation of the patient to be assessed, facilitate medical intervention based on the assessment results, provide support for perioperative management, and reduce the risk of perioperative stroke.

[0066] In this embodiment, a method for assessing the risk of perioperative stroke during coronary artery bypass grafting is provided, which can be used in the above-mentioned terminal. The method specifically includes the following steps:

[0067] Step S201: Acquire multiple sets of training physiological data. Each set of training physiological data belongs to the same patient and includes the patient's current physiological data and historical physiological data. Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0068] Step S202 : training the risk assessment model based on multiple sets of training physiological data to obtain a target risk assessment model.

[0069] Specifically, the above step S202 includes:

[0070] Step S2021 : Processing is performed based on the multiple sets of training physiological data to obtain multiple sets of pre-training physiological data.

[0071] In some optional implementations, the above step S2021 includes:

[0072] Step a1, preprocessing each set of training physiological data to obtain preprocessed training physiological data. Specifically, preprocessing operations such as data deduplication and missing value processing are performed on each set of training physiological data. If the proportion of missing data of any indicator in the training physiological data is greater than a preset threshold, the indicator is removed from the training physiological data. In addition, zero-variance variables and near-zero-variance variables are removed from the data of each indicator. Optionally, a multiple interpolation method based on random forest can be selected for missing value processing. By performing data preprocessing, reliable data is provided for model training.

[0073] Step a2, divide each set of pre-processed training physiological data according to the format to obtain historical time series data, evaluation state data and static feature data. Specifically, each set of pre-processed training physiological data is divided into three categories according to the format: historical time series data, evaluation state data and static feature data. Among them, historical time series data refers to a sequence of physiological indicator data within a period of time, such as the hemoglobin indicator value of the same patient in the routine blood test in the past year, recorded in time series; evaluation state data refers to the patient's current status indicator, such as the various examination indicator data of the patient before the evaluation; static feature data refers to the patient's data that basically does not change over time, such as gender, age, whether there is a history of surgery, whether there is a history of hypertension, etc. Optionally, historical time series data is represented by [indicator name, time, value]; evaluation state data is represented by [indicator name, value]; static feature data is represented by [indicator name, value].

[0074] In step a3, each set of preprocessed training physiological data is labeled to obtain pretrained physiological data. Specifically, the preprocessed training data is labeled manually or machine-labeled using a training model. The labeled content includes important risk factors for stroke complications, actual stroke risk prediction values, and actual risk groups, which serve as criteria for supervised model learning.

[0075] Step S2022: Construct a test set and a validation set based on multiple sets of pre-trained physiological data. Specifically, the test set and validation set are divided into a certain proportion from the multiple sets of pre-trained physiological data. The test set is used to train the model, and the validation set is used to monitor the model performance during the training process to prevent the model from overfitting and ensure that the trained model has good performance. Optionally, for the test set and validation set, positive and negative samples can also be determined, and the pre-trained physiological data without stroke is used as the positive sample, and the pre-trained physiological data with stroke is used as the negative sample to further improve the accuracy of model training.

[0076] In step S2023, the risk assessment model is adopted, trained based on the test set, and the training process is monitored based on the validation set to obtain a target risk assessment model.

[0077] In some optional implementations, step S2023 includes:

[0078] Step b1, using the long short-term memory network in the risk assessment model, based on any set of pre-trained physiological data in the test set, perform feature extraction on the historical time series data to obtain a time series feature vector, and using the deep neural network in the risk assessment model, based on the evaluation state data and static feature data in the pre-trained physiological data, perform feature extraction to obtain a static feature vector. Specifically, a risk assessment model combining a deep neural network (DNN) and a long short-term memory network (LSTM) is constructed. During the model training process, LSTM is used to process the historical time series data in the test set, capture the dynamic change pattern of the data, and output a time series feature vector. At the same time, DNN is used to process the evaluation state data and static feature data, extract high-order nonlinear features, and obtain a static feature vector, so that the model can fully mine the key information in different types of data.

[0079] In step b2, the time-series feature vector and the static feature vector are fused to obtain a training feature vector. Specifically, the time-series feature vector and the static feature vector are weighted and added together to jointly represent the overall indicator data characteristics. This allows the model to comprehensively consider the patient's physiological characteristics across different time dimensions and types, improving the accuracy of risk assessment. Optionally, the weights of the two feature vectors can be manually set at the beginning of training and dynamically adjusted during model training.

[0080] In step b3, the multi-task output layer of the risk assessment model performs an assessment based on the training feature vectors to obtain training assessment results, which include stroke risk prediction values and risk groups. Specifically, the fused training feature vectors are input into the multi-task output layer, which is divided into two output heads, outputting the results of the regression and classification tasks, respectively. These outputs represent the stroke risk prediction value and risk group, achieving a comprehensive assessment of the patient's risk. The risk groups are divided into high-risk and low-risk groups.

[0081] Step b4: Calculate the training loss based on the training evaluation results and annotations of the pre-trained physiological data.

[0082] In some optional implementations, the above step b4 includes:

[0083] Step c1: Determine the least squares error based on the stroke risk prediction values from the training evaluation results and the actual stroke risk prediction values from the annotations. Specifically, during model training, there is often a discrepancy between the predicted and actual values. By calculating the least squares error between the stroke risk prediction values obtained by the model and the actual stroke risk prediction values from the annotations, the degree of error in the model's regression task can be measured.

[0084] In step c2, the cross-entropy loss is determined based on the risk groups in the training evaluation results and the actual risk groups in the annotations. Specifically, by calculating the cross-entropy loss between the groups obtained by the model and the groups in the annotations, the degree of error of the model in the classification task can be measured.

[0085] In step c3, we obtain the weights corresponding to the least squares loss and the cross entropy loss. Specifically, during model training, the least squares loss focuses on the regression task, while the cross entropy loss focuses on the classification task. By setting the weights, we can balance the relative importance of the two during training, so that the model can achieve better performance overall.

[0086] In step c4, the training loss is determined based on the least squares loss and its weight and the cross entropy loss and its weight. Specifically, the training loss of the model is determined by the following formula (1), which can fully reflect the performance of the model in risk prediction and risk grouping, and provide a target for model optimization.

[0087]

[0088] Where L represents the training loss; α represents the weight corresponding to the least squares loss; LSE represents the least squares loss; represents the stroke risk prediction value in the training evaluation results; Y risk represents the actual stroke risk prediction value in the annotation; (1-α) represents the weight corresponding to the cross entropy loss; CrossEntropy represents the cross entropy loss; represents the risk grouping in the training evaluation results; Y group Indicates the actual risk grouping in the annotation.

[0089] In step b5, the risk assessment model is optimized using error back propagation based on the training loss. Specifically, the error back propagation algorithm is used to adjust the model parameters based on the training loss, so that the model prediction results are closer to the actual situation and the model performance is continuously improved.

[0090] Step b6: Use the optimized risk assessment model to perform model validation based on the validation set to obtain training indicators. Specifically, the optimized model is evaluated on the validation set to obtain training indicators. The training indicators can be AUC (Area Under the Curve), R 2 (coefficient of determination), etc., and can also be validation loss (see steps c1 to c4). These training indicators reflect the performance of the model on the validation set and can determine whether the model has achieved the expected training effect.

[0091] Step b7, when the training indicators do not meet the training stop conditions, return to the long short-term memory network in the risk assessment model, extract features based on the historical time series data in any group of pre-trained physiological data in the test set, and obtain the time series feature vector. Use the deep neural network in the risk assessment model to extract features based on the evaluation state data and static feature data in the pre-trained physiological data to obtain the static feature vector. Re-train the model until the training indicators meet the training stop conditions, and use the risk assessment model obtained by the last optimization as the target risk assessment model. Specifically, some checkpoints can be set during the model training process to save the training parameters and status of the current training step, and check the performance of the model at these checkpoints. When the training indicators are AUC, R 2 , it can be determined whether its performance at these checkpoints has improved. If performance continues to improve, return to step b1 to continue model training. If performance no longer improves, it is considered that the training termination condition has been met. When the training metric is validation loss, it can be determined whether it has reached the preset value. If not, return to step b1 to continue model training. If so, it is considered that the training termination condition has been met. When the training termination condition is met, the model obtained from the last optimization is used as the target risk assessment model to ensure that the model is optimized in terms of accuracy and reliability.

[0092] In some optional embodiments, Figure 2 is a schematic diagram of a calibration curve of a target risk assessment model according to an embodiment of the present invention, such as Figure 2 The figure shows the relationship between the model's predicted stroke risk and the actual stroke risk predicted in the annotation. The "Ideal State" curve represents an ideal situation where the model's predictions completely match the actual values; the "Original State" curve reflects the model's original predictions; and the "Bias-Corrected State" curve shows the predictions after bias correction. These three curves reflect the relationship between the model's predictions and actual values from different perspectives, thereby demonstrating the model's performance. Figure 3 : is a schematic diagram of the model performance of the target risk assessment model according to an embodiment of the present invention, such as Figure 3 As shown in the figure, several checkpoints during the training process were selected for performance verification. The model's areas under the ROC curve (Receiver Operating Characteristic Curve) were all above 0.7 at all three checkpoints. At checkpoint 1, the AUC value was 0.771, the specificity was 0.671, and the sensitivity was 0.738, demonstrating good discrimination, high risk assessment accuracy, and high grouping efficiency.

[0093] Step S203 : For each patient to be evaluated, extract the original physiological data of the patient to be evaluated to obtain multiple physiological index data. The original physiological data includes the current physiological data and historical physiological data of the patient to be evaluated. The patient to be evaluated is in the perioperative period of coronary artery bypass grafting.

[0094] Specifically, the above step S203 includes:

[0095] In step S2031, the large language model is used to extract and structure the numerical indicator data in the raw physiological data, obtaining a plurality of first indicator data. Specifically, the numerical indicator data in the raw physiological data, which represents the patient's various indicators using numerical values, is organized using the large language model into a standard structured data structure for model training and calculation, namely the first indicator data. For example, each specific indicator item in a routine blood test is organized into a unified medical indicator data format.

[0096] In step S2032, the large language model is used to extract and structure the textual data within the raw physiological data, obtaining multiple secondary indicator data. Specifically, for the textual data within the raw physiological data, i.e., the indicator information obtained through the textual content, the large language model is used to analyze the textual semantics of this data, locate and identify the indicator data within the text, extract the indicator values, and organize them into a standard structured data structure to obtain secondary indicator data. For example, the degree of vertebral artery stenosis can be extracted from the textual conclusions of a vertebral artery ultrasound examination. By utilizing the large language model to analyze textual semantics, various medical indicators within the raw physiological data can be automatically identified in conjunction with contextual information, reducing the workload of manual data entry and organization, significantly improving data recognition accuracy, and enhancing the level of automation.

[0097] In step S2033, the plurality of first indicator data and the plurality of second indicator data are used as a plurality of physiological indicator data. Specifically, the indicator data obtained by the large language model are integrated as physiological indicator data that comprehensively represent the physical condition of the patient to be evaluated, providing comprehensive input data for model evaluation.

[0098] In step S204, the target risk assessment model is used to perform an assessment based on the physiological indicator data to obtain an assessment result for the patient. Specifically, the physiological indicator data of the patient to be assessed is input into the trained target risk assessment model. The model analyzes and calculates the physiological indicator data based on the learned patterns and characteristic relationships, referring to steps b1 to b3 above, to obtain the assessment result for the patient to be assessed: a stroke risk prediction value and risk grouping. This allows for stroke risk assessment of perioperative patients, providing a quantitative basis for clinical decision-making.

[0099] Step S205 determines a medical treatment recommendation for the patient based on the assessment results. Specifically, the assessment results for the patient include a predicted stroke risk value and risk group. Based on the risk group to which the patient belongs, a corresponding medical treatment recommendation is provided. For patients in the high-risk group, a comprehensive preoperative examination and assessment should be conducted, including cardiovascular status, cerebrovascular status, and diabetes control. Anticoagulants or antiplatelet drugs should be used appropriately to prevent thrombosis and reduce the likelihood of stroke. Furthermore, intraoperative and postoperative monitoring of the patient's vital signs, particularly blood pressure, should be strengthened to avoid excessively high or low blood pressure fluctuations. For patients in the low-risk group, although their baseline stroke risk is relatively low, a series of measures should still be taken to reduce the likelihood of stroke. For example, optimal control of chronic conditions such as hypertension, diabetes, and hyperlipidemia should be implemented; stable blood pressure levels should be maintained during surgery to avoid large fluctuations; appropriate anesthesia techniques should be used to minimize the impact on the cardiovascular system; and postoperative monitoring of the patient's vital signs, particularly their neurological status, should be closely monitored. By determining corresponding medical treatment recommendations based on the assessment results, it helps to carry out medical intervention, provide support for perioperative management, and reduce the risk of perioperative stroke.

[0100] The perioperative stroke risk assessment method for coronary artery bypass grafting provided by an embodiment of the present invention can more comprehensively reflect the patient's condition by collecting the patient's current and historical physiological data during the perioperative period, providing rich and comprehensive basic data for subsequent model training. Furthermore, training is performed based on multiple sets of training physiological data, so that the trained target risk assessment model can accurately perform stroke risk assessment, thereby improving the scientific nature and reliability of the assessment results. After extracting multiple physiological indicator data of the patient to be assessed, the target risk assessment model is used for assessment, which can generate accurate assessment results based on the specific situation of the patient to be assessed, facilitate medical intervention based on the assessment results, provide support for perioperative management, and reduce the risk of perioperative stroke.

[0101] This embodiment also provides a device for assessing perioperative stroke risk during coronary artery bypass grafting. This device is used to implement the above-mentioned embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" may refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0102] This embodiment provides a device for assessing the risk of stroke during the perioperative period of coronary artery bypass grafting. Figure 4 As shown, including:

[0103] The acquisition module 401 is used to acquire multiple sets of training physiological data, each set of training physiological data belongs to the same patient, and includes the patient's current physiological data and historical physiological data.

[0104] The training module 402 is used to train the risk assessment model based on multiple sets of training physiological data to obtain a target risk assessment model.

[0105] The extraction module 403 is used to extract the original physiological data of each patient to be evaluated to obtain multiple physiological indicator data. The original physiological data includes the current physiological data and historical physiological data of the patient to be evaluated. The patient to be evaluated is in the perioperative period of coronary artery bypass grafting.

[0106] The evaluation module 404 is configured to use a target risk evaluation model to perform evaluation based on physiological indicator data to obtain an evaluation result of the patient to be evaluated.

[0107] In some optional implementations, the training module 402 includes:

[0108] The data processing unit is used to process multiple sets of training physiological data to obtain multiple sets of pre-training physiological data.

[0109] The construction unit is used to construct a test set and a validation set based on multiple sets of pre-training physiological data.

[0110] The training unit is used to adopt the risk assessment model, perform training based on the test set, monitor the training process based on the validation set, and obtain the target risk assessment model.

[0111] In some optional embodiments, the data processing unit includes:

[0112] The preprocessing subunit is used to preprocess each set of training physiological data to obtain preprocessed training physiological data.

[0113] The division subunit is used to divide each group of pre-processed training physiological data according to the format to obtain historical time series data, evaluation state data and static feature data.

[0114] The labeling subunit is used to label each set of pre-processed training physiological data to obtain pre-trained physiological data.

[0115] In some optional embodiments, the training unit includes:

[0116] The extraction subunit is used to use the long short-term memory network in the risk assessment model to perform feature extraction based on the historical time series data in any set of pre-trained physiological data in the test set to obtain a time series feature vector, and to use the deep neural network in the risk assessment model to perform feature extraction based on the assessment state data and static feature data in the pre-trained physiological data to obtain a static feature vector.

[0117] The fusion subunit is used to fuse the time series feature vector and the static feature vector to obtain the training feature vector.

[0118] The evaluation subunit is used to use the multi-task output layer in the risk assessment model to perform evaluation based on the training feature vector to obtain training evaluation results, which include stroke risk prediction values and risk groups.

[0119] The computing subunit is used to calculate the training loss based on the training evaluation results and annotations of the pre-trained physiological data.

[0120] The optimization subunit is used to optimize the risk assessment model through error back propagation based on the training loss.

[0121] The verification subunit is used to use the optimized risk assessment model to perform model verification based on the verification set to obtain training indicators.

[0122] The training subunit is used to return to the long short-term memory network in the risk assessment model when the training indicators do not meet the training stop conditions, perform feature extraction based on the historical time series data in any group of pre-trained physiological data in the test set, and obtain a time series feature vector, use the deep neural network in the risk assessment model, perform feature extraction based on the assessment state data and static feature data in the pre-trained physiological data, and obtain a static feature vector, re-train the model until the training indicators meet the training stop conditions, and use the risk assessment model obtained by the last optimization as the target risk assessment model.

[0123] In some optional embodiments, the computing subunit is specifically configured to determine a least squares loss based on the stroke risk prediction value in the training evaluation result and the actual stroke risk prediction value in the annotation. Determine a cross entropy loss based on the risk grouping in the training evaluation result and the actual risk grouping in the annotation. Obtain weights corresponding to the least squares loss and the cross entropy loss, respectively. Determine a training loss based on the least squares loss and its weight and the cross entropy loss and its weight.

[0124] In some optional implementations, the extraction module 403 includes:

[0125] The first extraction unit is used to extract and structure the numerical indicator data in the original physiological data using a large language model to obtain multiple first indicator data.

[0126] The second extraction unit is used to extract indicators from the text data in the original physiological data using a large language model and structure it to obtain multiple second indicator data.

[0127] The determining unit is configured to use the plurality of first indicator data and the plurality of second indicator data as a plurality of physiological indicator data.

[0128] In some optional embodiments, the device further comprises:

[0129] The determination module is used to determine a medical treatment recommendation for the patient to be evaluated based on the evaluation result of the patient to be evaluated.

[0130] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0131] The coronary artery bypass grafting perioperative stroke risk assessment device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0132] The embodiment of the present invention also provides a computer device having the above Figure 4 A device for assessing perioperative stroke risk during coronary artery bypass grafting is shown.

[0133] See also Figure 5 , Figure 5 is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 5 As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 5 A processor 10 is taken as an example.

[0134] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0135] The memory 20 stores instructions that can be executed by at least one processor 10, so as to enable at least one processor 10 to execute the method shown in the above embodiment.

[0136] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created based on the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0137] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0138] The computer device further includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 5 The bus connection is taken as an example.

[0139] The input device 30 can receive input digital or character information and generate key signal input related to user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touch pad, an indicator stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), and a tactile feedback device (e.g., a vibration motor). The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display, and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0140] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0141] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.

[0142] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for assessing the risk of perioperative stroke during coronary artery bypass grafting, characterized in that: The method comprises: Acquire multiple sets of training physiological data, each set of training physiological data belongs to the same patient, including current physiological data and historical physiological data of the patient; Training a risk assessment model based on the multiple sets of training physiological data to obtain a target risk assessment model; For each patient to be evaluated, extracting original physiological data of the patient to be evaluated to obtain a plurality of physiological indicator data, wherein the original physiological data includes current physiological data and historical physiological data of the patient to be evaluated, and the patient to be evaluated is in the perioperative period of coronary artery bypass grafting; The target risk assessment model is used to perform an assessment based on the physiological indicator data to obtain an assessment result of the patient to be assessed.

2. The method according to claim 1, characterized in that The risk assessment model is trained based on the multiple sets of training physiological data to obtain a target risk assessment model, including: Processing the multiple sets of training physiological data to obtain multiple sets of pre-training physiological data; Constructing a test set and a validation set based on the multiple sets of pre-training physiological data; The risk assessment model is adopted, trained based on the test set, and the training process is monitored based on the validation set to obtain the target risk assessment model.

3. The method according to claim 2, characterized in that The processing based on the multiple sets of training physiological data to obtain multiple sets of pre-training physiological data includes: Preprocessing each set of training physiological data to obtain preprocessed training physiological data; Each set of pre-processed training physiological data is divided according to the format to obtain historical time series data, evaluation state data and static feature data; Each group of pre-processed training physiological data is labeled to obtain pre-training physiological data.

4. The method according to claim 3, characterized in that The risk assessment model is adopted, training is performed based on the test set, and the training process is monitored based on the validation set to obtain the target risk assessment model, including: Using the long short-term memory network in the risk assessment model, feature extraction is performed based on the historical time series data in any set of pre-trained physiological data in the test set to obtain a time series feature vector; using the deep neural network in the risk assessment model, feature extraction is performed based on the assessment state data and static feature data in the pre-trained physiological data to obtain a static feature vector; fusing the temporal feature vector and the static feature vector to obtain a training feature vector; Using the multi-task output layer in the risk assessment model, an assessment is performed based on the training feature vector to obtain a training assessment result, wherein the training assessment result includes a stroke risk prediction value and a risk group; Calculating a training loss based on the training evaluation results and annotations of the pre-trained physiological data; Optimizing the risk assessment model based on the training loss through error back propagation; Using the optimized risk assessment model, model validation is performed based on the validation set to obtain training indicators; When the training indicator does not meet the training stop condition, return to the step of using the long short-term memory network in the risk assessment model, perform feature extraction based on the historical time series data in any group of pre-trained physiological data in the test set, and obtain a time series feature vector, use the deep neural network in the risk assessment model, perform feature extraction based on the assessment state data and static feature data in the pre-trained physiological data, and obtain a static feature vector, re-train the model until the training indicator meets the training stop condition, and use the risk assessment model obtained by the last optimization as the target risk assessment model.

5. The method according to claim 4, characterized in that The calculating of the training loss based on the evaluation results and annotations of the pre-trained physiological data includes: determining a least squares loss based on the stroke risk prediction value in the training evaluation result and the actual stroke risk prediction value in the annotation; Determining a cross entropy loss based on the risk grouping in the training evaluation result and the actual risk grouping in the annotation; Obtain weights corresponding to the least squares loss and the cross entropy loss respectively; The training loss is determined based on the least squares loss and its weight and the cross entropy loss and its weight.

6. The method according to claim 1, characterized in that For each patient to be evaluated, data extraction is performed on the original physiological data of the patient to be evaluated to obtain multiple physiological indicator data, including: For the numerical indicator data in the original physiological data, a large language model is used to extract and structure the data to obtain a plurality of first indicator data; For the text data in the original physiological data, extract indicators using a large language model and structure them to obtain multiple second indicator data; The plurality of first indicator data and the plurality of second indicator data are used as the plurality of physiological indicator data.

7. The method according to claim 1, characterized in that The method further comprises: Based on the evaluation result of the patient to be evaluated, a medical treatment recommendation for the patient to be evaluated is determined.

8. A device for assessing the risk of perioperative stroke during coronary artery bypass grafting, characterized in that: The device comprises: an acquisition module, configured to acquire multiple sets of training physiological data, each set of training physiological data belonging to the same patient, including current physiological data and historical physiological data of the patient; A training module, configured to train the risk assessment model based on the multiple sets of training physiological data to obtain a target risk assessment model; an extraction module, configured to extract, for each patient to be evaluated, original physiological data of the patient to be evaluated to obtain a plurality of physiological indicator data, wherein the original physiological data includes current physiological data and historical physiological data of the patient to be evaluated, and the patient to be evaluated is in the perioperative period of coronary artery bypass grafting; An evaluation module is used to use the target risk evaluation model to perform evaluation based on the physiological indicator data to obtain an evaluation result of the patient to be evaluated.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for assessing perioperative stroke risk during coronary artery bypass grafting according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method for assessing perioperative stroke risk during coronary artery bypass grafting according to any one of claims 1 to 7.

Images