Method and system for predicting risk of allogeneic blood transfusion during perioperative cardiac surgery

Abstract

The application provides a method and system for predicting the risk of allogeneic blood transfusion during cardiac surgery. The method obtains static clinical data and intraoperative real-time monitoring data of the target patient, extracts static features and intraoperative dynamic features based on a sliding window, and concatenates the features in the input layer of a pre-trained allogeneic blood transfusion risk prediction model according to the feature dimensions to obtain a mixed feature vector corresponding to each sliding window. Each mixed feature vector is input into the allogeneic blood transfusion risk prediction model in turn to obtain an allogeneic blood transfusion risk probability updated with the sliding window. The risk level and risk warning state are determined according to the continuously updated allogeneic blood transfusion risk probability and a preset risk threshold, and the risk level and the risk warning in the trigger state are output to the clinical anesthesia information system. The method can continuously update the allogeneic blood transfusion risk and provide dynamic warning as the intraoperative monitoring data changes.

Description

Technical Field

[0001] This application relates to the fields of medical big data and artificial intelligence technology, and more specifically, to a method and system for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery. Background Technology

[0002] Cardiac surgery is highly invasive and involves prolonged cardiopulmonary bypass, inherently carrying a high risk of perioperative bleeding, often necessitating allogeneic blood transfusions in clinical practice. However, allogeneic blood transfusions can lead to a series of problems, including transfusion reactions, immunosuppression, increased risk of infection, and higher medical costs. Therefore, it is essential to assess the risks of allogeneic blood transfusions in patients during the perioperative period to provide a basis for preoperative blood preparation and intraoperative transfusion management.

[0003] Currently, clinical assessments primarily rely on physicians' personal experience or employ simplified scoring systems based on a limited number of clinical indicators and statistical methods such as logistic regression. These tools are mostly in the form of standalone calculators or paper forms, and assessments are concentrated in the pre-operative or post-operative stages, with low integration with clinical anesthesia information systems.

[0004] Existing methods rely on relatively singular predictive dimensions, making it difficult to comprehensively consider both static baseline patient information and continuous changes in intraoperative physiological parameters. Traditional statistical methods such as logistic regression have limited capabilities when handling nonlinear relationships and high-dimensional data. Furthermore, assessments are often limited to preoperative or postoperative periods, failing to dynamically update risk levels based on real-time changes in the patient's condition during surgery, thus lacking intraoperative early warning capabilities. The use of standalone tools further increases the workload of medical staff. In summary, existing methods do not adequately address the key needs of multidimensional data utilization, dynamic intraoperative early warning, and clinical system integration.

[0005] Therefore, this application provides a method and system for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery, in order to solve one of the above-mentioned technical problems. Summary of the Invention

[0006] The purpose of this application is to provide a method and system for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery, which can solve at least one of the aforementioned technical problems. The specific solution is as follows: According to a specific embodiment of this application, in a first aspect, this application provides a method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery, comprising: Acquire static clinical data and intraoperative real-time monitoring data of the target patient; extract static features from the static clinical data and extract intraoperative dynamic features from the intraoperative real-time monitoring data based on a sliding window; in the input layer of a pre-trained allogeneic transfusion risk prediction model, concatenate the static features and the intraoperative dynamic features according to the feature dimensions to obtain a mixed feature vector corresponding to each sliding window; input each mixed feature vector into the allogeneic transfusion risk prediction model in sequence to obtain the allogeneic transfusion risk probability updated with the sliding window; determine the risk level and risk warning status based on the continuously updated allogeneic transfusion risk probability and a preset risk threshold, and output the risk level and the risk warning in the triggered state to the clinical anesthesia information system.

[0007] According to a specific embodiment of this application, in a second aspect, this application provides a system for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery, comprising: The data acquisition module is used to acquire the patient's static clinical data and intraoperative real-time monitoring data, and output the static clinical data and the intraoperative real-time monitoring data to the artificial intelligence prediction module. The artificial intelligence prediction module is used to extract static features from the static clinical data, extract intraoperative dynamic features from the intraoperative real-time monitoring data based on a sliding window, concatenate the static features and the intraoperative dynamic features according to the feature dimensions to obtain a mixed feature vector corresponding to each sliding window, input each mixed feature vector sequentially into a pre-trained allogeneic transfusion risk prediction model to obtain the allogeneic transfusion risk probability updated with the sliding window, and output the allogeneic transfusion risk probability to the clinical early warning and decision recommendation module. The clinical early warning and decision recommendation module is used to determine the risk level based on the allogeneic transfusion risk probability and a preset risk threshold, trigger a risk warning when the allogeneic transfusion risk probability corresponding to multiple consecutive sliding windows exceeds the preset risk threshold, generate a processing suggestion corresponding to the risk level in combination with a medical knowledge base, and output the risk level, the risk warning, and the processing suggestion to the clinical anesthesia information system.

[0008] Compared with the prior art, the above-described solutions of this application have at least the following beneficial effects: This application provides a method for predicting the risk of perioperative allogeneic transfusion in cardiac surgery. This method extracts static features from static clinical data and dynamic features from real-time intraoperative monitoring data using a sliding window. These two types of features are then concatenated dimensionally at the model input layer to form a hybrid feature vector corresponding to each sliding window. Based on this, the pre-trained allogeneic transfusion risk prediction model can simultaneously consider the patient's static baseline condition and the constantly changing dynamic information during surgery, thus making a more accurate judgment on the current risk. The hybrid feature vectors of each sliding window are sequentially input into the model, and the risk probability is continuously updated as the window slides. Based on this, the risk level can be determined and a corresponding warning state can be triggered according to the continuously updated risk probability and a preset threshold. Finally, the results are output to the clinical anesthesia information system. Thus, continuous dynamic assessment and real-time intraoperative warning of perioperative allogeneic transfusion risk can be achieved, effectively overcoming the risk identification lag problem caused by relying solely on static assessment in the past. Attached Figure Description

[0009] Figure 1 A flowchart for a method to predict the risk of perioperative allogeneic blood transfusion in cardiac surgery; Figure 2 A flowchart for sliding window updates and intraoperative dynamic feature extraction; Figure 3 Flowchart for pre-training a risk prediction model for allogeneic blood transfusion; Figure 4 A flowchart for constructing training sample labels; Figure 5 A flowchart for triggering and canceling risk warnings; Figure 6 A flowchart for feature importance analysis and key factor output; Figure 7 Structure and data flow diagram of a system for predicting perioperative allogeneic blood transfusion risk in cardiac surgery; Figure 8 This is a schematic diagram of the dynamic risk dashboard interface. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0012] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0013] It should be understood that although the terms first, second, third, etc., may be used in the embodiments of this application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, first may also be referred to as second without departing from the scope of the embodiments of this application, and similarly, second may also be referred to as first.

[0014] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0015] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes said element.

[0016] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.

[0017] The optional embodiments of this application are described in detail below with reference to the accompanying drawings.

[0018] The embodiments provided in this application are embodiments of a method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery.

[0019] The following is combined with Figure 1 The embodiments of this application will be described in detail.

[0020] Figure 1 A flowchart for a method to predict the risk of perioperative allogeneic blood transfusion in cardiac surgery.

[0021] Reference Figure 1 In some embodiments, methods for predicting the risk of perioperative allogeneic blood transfusion during cardiac surgery include: S101a. Obtain static clinical data of the target patients; S101b: Acquire real-time intraoperative monitoring data of the target patient; S102a. Extracting static features from static clinical data; S102b: Extracting intraoperative dynamic features from real-time intraoperative monitoring data based on a sliding window; Among them, static features are used to characterize the relatively stable clinical state of the target patient, while dynamic features are used to characterize the changes in the target patient during the operation.

[0022] S103. In the input layer of the pre-trained allogeneic blood transfusion risk prediction model, static features and intraoperative dynamic features are concatenated according to the feature dimensions to obtain the hybrid feature vector corresponding to each sliding window. S104. Input each mixed feature vector into the allogeneic transfusion risk prediction model in sequence to obtain the allogeneic transfusion risk probability updated with the sliding window. S105. Based on the continuously updated allogeneic blood transfusion risk probability and preset risk threshold, determine the risk level and risk warning status, and output the risk level and the risk warning in the triggered state to the clinical anesthesia information system.

[0023] For example, when extracting intraoperative dynamic features from intraoperative real-time monitoring data based on a sliding window, the data range used for this feature extraction is limited by the current sliding window, so that each sliding window corresponds to a set of intraoperative dynamic features.

[0024] In this embodiment, at the input layer of the pre-trained allogeneic transfusion risk prediction model, static features and intraoperative dynamic features corresponding to the current sliding window are concatenated according to the feature dimension, so that the two feature vectors form a mixed feature vector in the feature dimension. For different sliding windows, the static features are concatenated with the intraoperative dynamic features of the corresponding sliding window to obtain the mixed feature vector corresponding to each sliding window.

[0025] In some specific embodiments, the mixed feature vectors are sequentially input into the allogeneic transfusion risk prediction model according to the update order of the sliding window.

[0026] The allogeneic transfusion risk prediction model calculates the risk of each hybrid feature vector based on the model parameters learned during the training phase, and outputs the allogeneic transfusion risk probability for the corresponding sliding window.

[0027] For example, during continuous monitoring, as new intraoperative real-time monitoring data enters, the sliding window continues to update, and intraoperative dynamic features, mixed feature vectors, and allogeneic transfusion risk probabilities are regenerated accordingly.

[0028] In some embodiments, the risk level and risk warning status are determined based on the relationship between continuously updated allogeneic blood transfusion risk probability and a preset risk threshold.

[0029] In some specific embodiments, when the risk warning state is triggered, the risk level and risk warning are output to the clinical anesthesia information system, enabling the clinical anesthesia information system to receive the corresponding risk information.

[0030] In this embodiment, by extracting static features from static clinical data and extracting intraoperative dynamic features from real-time intraoperative monitoring data based on a sliding window, and then concatenating the two types of features according to feature dimensions and performing risk calculations sequentially, it is possible to simultaneously utilize the target patient's baseline condition and monitoring information that changes during the surgical process. Furthermore, the probability of allogeneic blood transfusion risk is updated with the sliding window and further used to determine the risk level and risk warning status, which helps to reduce the risk identification lag caused by relying solely on static assessments or single-point-of-time monitoring data.

[0031] In some embodiments, the static clinical data of the target patient are obtained from the hospital information system, the laboratory information system, and the clinical anesthesia information system, respectively.

[0032] Among them, the hospital information system and laboratory information system provide preoperative baseline data for the target patient, while the clinical anesthesia information system provides static clinical data related to the current surgery.

[0033] In some embodiments, intraoperative real-time monitoring data of the target patient are acquired from a patient monitor, anesthesia machine, cardiopulmonary bypass machine, and ventilator, respectively.

[0034] Different devices continuously generate corresponding monitoring data during the operation. After obtaining the corresponding data, this application uses the corresponding data as input for dynamic feature extraction of the sliding window.

[0035] Figure 2 This is a flowchart for sliding window updates and intraoperative dynamic feature extraction.

[0036] Reference Figure 2 This includes the following steps: S201. Determine the sliding window with a fixed time length; S202. In response to the detection of new intraoperative real-time monitoring data, the sliding window is moved along the time series and the intraoperative dynamic characteristics within the current sliding window are recalculated. S203. Calculate the mean, maximum, minimum, median, standard deviation, variance, and quartiles within the current sliding window to obtain statistical characteristics; determine the maximum rate of increase, maximum rate of decrease, slope of change, cumulative change, and direction of change within the current sliding window to obtain trend characteristics; calculate the coefficient of variation, mean of continuous differences, fluctuation amplitude, and number of abnormal fluctuations within the current sliding window to obtain volatility characteristics; statistically analyze the duration of hypotension, duration of hypoxia, and duration of body temperature below a preset body temperature threshold within the current sliding window to obtain duration characteristics; extract changes in anesthetic drug infusion rate, time of initiation of vasoactive drug use, number of vasoactive drug dose adjustments, infusion rate, and time of occurrence of important surgical events within the current sliding window to obtain treatment intervention characteristics; In S203, each feature acquisition method is executed separately, and this application does not specify the execution order.

[0037] S204. Based on statistical characteristics, trend characteristics, volatility characteristics, duration characteristics, and treatment intervention characteristics, the intraoperative dynamic characteristics corresponding to the current sliding window are formed.

[0038] In some specific embodiments, the range of intraoperative real-time monitoring data used for each dynamic feature calculation is limited by a fixed time length. Once the sliding window is determined, the current sliding window covers a continuous segment of data in the time series. When new intraoperative real-time monitoring data is detected, the sliding window is moved along the time series so that the current sliding window includes the updated intraoperative real-time monitoring data, and the intraoperative dynamic features corresponding to the current sliding window are recalculated.

[0039] As one specific implementation method, sampling data can be generated every thirty seconds, the sliding window can be moved once every minute and a model calculation can be performed once. Each model calculation uses data from the ten minutes prior to the current time point.

[0040] As a specific implementation method, for infant heart surgery, the sliding window duration can be configured to five minutes.

[0041] In this embodiment, statistical features are used to describe the overall level and fluctuation of the corresponding data within the current sliding window, trend features are used to reflect the trend of the corresponding data within the current sliding window, volatility features are used to evaluate the stability of the corresponding data within the current sliding window, duration features are used to characterize the persistence of the corresponding abnormal state, and treatment intervention features are used to describe the treatment intervention process within the time range covered by the current sliding window.

[0042] In this embodiment, by taking a common current sliding window as input, statistical features, trend features, volatility features, duration features, and treatment intervention features are calculated or extracted in parallel. This allows the patient's status corresponding to the current sliding window to be described from multiple aspects, including the overall level, trend, volatility stability, degree of abnormality, and treatment intervention process.

[0043] In some embodiments, the allogeneic transfusion risk prediction model is pre-trained using the following process before acquiring static clinical data and intraoperative real-time monitoring data of the target patient. Figure 3 A flowchart for pre-training a risk prediction model for allogeneic blood transfusion.

[0044] Reference Figure 3 This includes the following steps: S301. Obtain historical perioperative case data; Among them, historical perioperative case data includes static clinical data of historical patients, intraoperative real-time monitoring data, and the results of allogeneic blood transfusion; S302. Process historical perioperative case data using the same static feature extraction, sliding window dynamic feature extraction, and feature splicing methods as the target patients, and construct training samples using allogeneic blood transfusion results as supervised learning labels. S303. Divide the training samples into training set, validation set and test set, use the training set to learn the model parameters, use the validation set to adjust the model hyperparameters, and use the test set to evaluate the model performance. S304. Save the trained model parameters to the prediction server to obtain the allogeneic transfusion risk prediction model.

[0045] In this embodiment, the training phase and the online prediction phase use the same static feature extraction, sliding window dynamic feature extraction, and feature concatenation methods, so that the model inputs formed in the two phases have the same feature composition relationship.

[0046] For example, during the model training phase, the perioperative case data of historical patients obtained includes static clinical data, intraoperative real-time monitoring data, and the results of allogeneic blood transfusion for each historical perioperative case.

[0047] For example, after the allogeneic transfusion risk prediction model is trained, the trained model parameters are saved to the prediction server to form an allogeneic transfusion risk prediction model that can be called in the online prediction stage.

[0048] For example, after the target patient enters the perioperative period, the online prediction process calls the model parameters stored in the prediction server and uses the same feature processing method as in the training phase to calculate the probability of allogeneic blood transfusion risk.

[0049] In some specific embodiments, the allogeneic transfusion risk prediction model may employ an extreme gradient boosting (XGBoost) model, a categorical gradient boosting (CatBoost) model, a random forest model, or a long short-term memory network (LSTM) model.

[0050] In this embodiment, the training set, validation set, and test set are responsible for model parameter learning, model hyperparameter adjustment, and model performance evaluation, respectively. This ensures that model performance evaluation does not directly use all the data used in model parameter learning, which helps to reduce the situation where the allogeneic transfusion risk prediction model can only be used for training data. It also enables the trained allogeneic transfusion risk prediction model to be used for online risk calculation of target patients.

[0051] Figure 4 A flowchart for constructing labels for training samples.

[0052] Reference Figure 4 This includes the following steps: S401. Obtain historical allogeneic transfusion event data from the blood transfusion management system and nursing records; For example, allogeneic transfusion event data includes whether an allogeneic transfusion occurred and the transfusion start time. The allogeneic transfusion status is used to generate supervised learning labels, and the transfusion start time is used to determine the position of the allogeneic transfusion event within the historical case time series.

[0053] S402. Use whether or not allogeneic blood transfusion occurred in the allogeneic blood transfusion event data as the supervised learning label, and determine the supervised learning label corresponding to each sliding window by backtracking according to the blood transfusion start time. S403. Combine the mixed feature vectors corresponding to each sliding window with the corresponding supervised learning labels to obtain training samples.

[0054] In this embodiment, whether or not allogeneic blood transfusion occurs is used as the supervised learning label, and the supervised learning label corresponding to each sliding window is determined by backtracking based on the transfusion start time, so that each sliding window has a corresponding supervised learning label.

[0055] For example, after obtaining the supervised learning labels corresponding to each sliding window, the mixed feature vector corresponding to each sliding window is combined with the corresponding supervised learning label to obtain the training samples.

[0056] In this embodiment, the transfusion start time is used only for backtracking during the training phase to determine the supervised learning label. During the online prediction phase for the target patient, the transfusion start time and transfusion volume are not used as input features of the allogeneic transfusion risk prediction model.

[0057] Figure 5 This is a flowchart for triggering and canceling risk warnings.

[0058] Reference Figure 5 This includes the following steps: S501. Determine the probability of allogeneic blood transfusion risk corresponding to multiple consecutive sliding windows; S502a. When the probability of allogeneic blood transfusion risk corresponding to multiple consecutive sliding windows exceeds the preset risk threshold, the risk warning state is determined as the trigger state and the risk warning is output. S502b: When the vital signs of the target patient stabilize and the probability of allogeneic blood transfusion risk decreases, the risk warning status is set to canceled and the risk warning is cancelled.

[0059] In this embodiment, the risk probability of allogeneic blood transfusion corresponding to multiple consecutive sliding windows is comprehensively judged, rather than triggering a formal risk warning based solely on the risk probability of allogeneic blood transfusion corresponding to a single sliding window.

[0060] As one specific implementation, when the probability of allogeneic blood transfusion risk corresponding to multiple consecutive sliding windows all exceeds a preset risk threshold, the consecutive window confirmation condition is further determined. If the consecutive window confirmation condition is met, the risk warning state can be determined as the triggered state, and a risk warning can be output to the clinical anesthesia information system.

[0061] Among them, the preset risk threshold is used to determine whether the risk probability corresponding to each sliding window reaches the risk warning condition, and the continuous window confirmation condition is used to determine whether the risk warning condition is continuously met in multiple consecutive sliding windows.

[0062] For example, when the probability of allogeneic blood transfusion risk corresponding to three consecutive sliding windows exceeds the preset risk threshold, a formal risk warning is triggered.

[0063] As one specific implementation method, the preset risk threshold can be set to 0.8.

[0064] For example, after a risk warning has been triggered, the probability of allogeneic blood transfusion risk can continue to be calculated based on subsequent sliding windows. When the target patient's vital signs stabilize and the probability of allogeneic blood transfusion risk decreases, the risk warning status is set to canceled, and the previously issued risk warning is cancelled. Therefore, warning triggering and warning cancellation can occur at different time stages in the perioperative period, responding to changes in risk probability and the target patient's condition, respectively.

[0065] In this embodiment, by comprehensively judging the allogeneic blood transfusion risk probability corresponding to multiple consecutive sliding windows, and triggering a risk warning only when multiple consecutive sliding windows meet the preset risk threshold condition, the possibility of false alarms caused by instantaneous monitoring errors or equipment noise can be reduced. Furthermore, the risk warning can be canceled as the target patient's vital signs stabilize and the allogeneic blood transfusion risk probability decreases, allowing the risk warning status to be updated according to changes in the patient's condition.

[0066] Figure 6 This is a flowchart for feature importance analysis and key factor output.

[0067] Reference Figure 6 This includes the following steps: S601. Perform feature importance analysis on the risk probability of allogeneic blood transfusion to determine the contribution of each feature in static and intraoperative dynamic features to the risk probability of allogeneic blood transfusion. S602. Screen key factors affecting the risk of allogeneic blood transfusion based on the degree of contribution; S603. Output the key factors and their contribution levels to the clinical anesthesia information system.

[0068] In this embodiment, after the allogeneic transfusion risk prediction model outputs the allogeneic transfusion risk probability, a feature importance analysis is performed on the allogeneic transfusion risk probability. The feature importance analysis processes both static features that form the corresponding mixed feature vector and intraoperative dynamic features. The analysis results are used to characterize the contribution of each feature to the corresponding allogeneic transfusion risk probability.

[0069] For example, based on the contribution of various static features and intraoperative dynamic features, key factors affecting the risk of allogeneic blood transfusion are screened from the corresponding features.

[0070] In this embodiment of the application, by providing key factors and their contribution levels in addition to the probability of allogeneic blood transfusion risk, the model output can be correlated with the static features and intraoperative dynamic features involved in the risk calculation. This approach is beneficial to improving the interpretability of allogeneic blood transfusion risk prediction results.

[0071] Figure 7 This is a structural diagram of a system for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery.

[0072] Reference Figure 7 The system includes a data acquisition module 701, an artificial intelligence prediction module 702, and a clinical early warning and decision recommendation module 703; The data acquisition module 701 is used to acquire the patient’s static clinical data and intraoperative real-time monitoring data, and output the static clinical data and intraoperative real-time monitoring data to the artificial intelligence prediction module 702. The artificial intelligence prediction module 702 is used to extract static features from static clinical data, extract intraoperative dynamic features from intraoperative real-time monitoring data based on sliding windows, and concatenate static features and intraoperative dynamic features according to feature dimensions to obtain the mixed feature vectors corresponding to each sliding window. The mixed feature vectors are then sequentially input into the pre-trained allogeneic transfusion risk prediction model to obtain the allogeneic transfusion risk probability updated with the sliding window. Finally, the allogeneic transfusion risk probability is output to the clinical early warning and decision recommendation module 703. The clinical early warning and decision recommendation module 703 is used to determine the risk level based on the risk probability of allogeneic blood transfusion and the preset risk threshold. When the risk probability of allogeneic blood transfusion corresponding to multiple consecutive sliding windows exceeds the preset risk threshold, a risk warning is triggered. The module combines the medical knowledge base to generate treatment suggestions corresponding to the risk level and outputs the risk level, risk warning and treatment suggestions to the clinical anesthesia information system.

[0073] In some embodiments, the allogeneic blood transfusion risk prediction model in the artificial intelligence prediction module 702 is encapsulated as a microservice, and the microservice is embedded into the clinical anesthesia information system through a medical data standard protocol. The artificial intelligence prediction module 702 is also used to: perform feature importance analysis on the risk probability of allogeneic blood transfusion, determine the contribution of each feature in static features and intraoperative dynamic features to the risk probability of allogeneic blood transfusion, screen key factors affecting the risk of allogeneic blood transfusion, and output the key factors and their contribution to the clinical early warning and decision recommendation module 703. The Clinical Early Warning and Decision Recommendation Module 703 outputs risk levels, risk warnings, and treatment suggestions in the following manner: A dynamic risk dashboard is generated on the interface of the clinical anesthesia information system, displaying the probability of allogeneic blood transfusion risk, the changing trend of the probability of allogeneic blood transfusion risk, the risk level, key factors, the contribution degree of key factors, and treatment suggestions, and pop-up reminders are generated for risk warnings.

[0074] Figure 8 This is a schematic diagram of the dynamic risk dashboard interface.

[0075] Reference Figure 8 The clinical early warning and decision recommendation module generates a dynamic risk dashboard on the interface of the clinical anesthesia information system. This dynamic risk dashboard includes a display area for allogeneic blood transfusion risk probability, a display area for risk probability change trends, a display area for risk levels, a display area for key factors and their contribution, and a display area for treatment suggestions.

[0076] For example, the allogeneic transfusion risk probability display area is used to display the allogeneic transfusion risk probability corresponding to the current sliding window. The risk probability change trend display area is used to display the change trend of the allogeneic transfusion risk probability according to the update order of the sliding windows, and the risk level display area is used to display the risk level determined based on the allogeneic transfusion risk probability and a preset risk threshold.

[0077] For example, the key factors and contribution level display area shows the key factors selected by the artificial intelligence prediction module and the corresponding contribution level of each key factor. The processing suggestion display area shows the processing suggestions generated by the clinical early warning and decision recommendation module in conjunction with the medical knowledge base.

[0078] For example, when a risk warning is triggered, the clinical warning and decision recommendation module generates a pop-up reminder on the interface of the clinical anesthesia information system. This pop-up reminder is presented together with the allogeneic transfusion risk probability, risk probability trend, risk level, key factors, contribution level, and treatment suggestions in the dynamic risk dashboard. This allows risk calculation results, risk change information, model interpretation results, and treatment suggestions to be output on the same interface, reducing information fragmentation caused by the scattered display of different risk information.

[0079] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0080] Although the operations are described in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all of the operations shown to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0081] The methods and systems of this application can be implemented using standard programming techniques, utilizing rule-based logic or other logic to implement various method steps. It should also be noted that the terms "system" and "module" as used herein and in the claims are intended to include implementations using one or more lines of software code and / or hardware implementations and / or devices for receiving input.

[0082] Any step, operation, or procedure described herein may be performed or implemented using one or more hardware or software modules, either alone or in combination with other devices. In one embodiment, the software module is implemented using a computer program product comprising a computer-readable medium containing computer program code, which is executable by a computer processor to perform any or all of the described steps, operations, or procedures.

[0083] The foregoing description of implementations of this application has been provided for illustrative and descriptive purposes. The foregoing description is not exhaustive and is not intended to limit this application to the exact forms disclosed. Various modifications and variations may exist in accordance with the foregoing teachings, or may arise from practice of this application. These embodiments were chosen and described to illustrate the principles of this application and its practical application, enabling those skilled in the art to utilize this application in various implementations and modifications to suit the specific purpose of the concept.

[0084] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0085] It can be further understood that, unless otherwise specified, "connection" includes both connections where no other components exist between them and indirect connections where other components exist between them.

[0086] It is further understood that although the operations are described in a specific order in the accompanying drawings in the embodiments of this application, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all the operations shown to be performed to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0087] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the field of this application that are not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0088] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

[0089] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery, characterized in that, include: Acquire static clinical data and intraoperative real-time monitoring data of the target patient; Static features are extracted from the static clinical data, and intraoperative dynamic features are extracted from the intraoperative real-time monitoring data based on a sliding window; In the input layer of the pre-trained allogeneic blood transfusion risk prediction model, the static features and the intraoperative dynamic features are concatenated according to the feature dimensions to obtain the hybrid feature vector corresponding to each sliding window. Each hybrid feature vector is sequentially input into the allogeneic transfusion risk prediction model to obtain the allogeneic transfusion risk probability updated with the sliding window; Based on the continuously updated allogeneic blood transfusion risk probability and preset risk threshold, the risk level and risk warning status are determined, and the risk level and the risk warning in the triggered state are output to the clinical anesthesia information system.

2. The method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery according to claim 1, characterized in that, The acquisition of static clinical data and intraoperative real-time monitoring data of the target patient includes: Static clinical data of the target patients are obtained from the hospital information system, laboratory information system, and clinical anesthesia information system. Intraoperative real-time monitoring data of the target patient are obtained from monitors, anesthesia machines, cardiopulmonary bypass machines, and ventilators.

3. The method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery according to claim 1, characterized in that, The extraction of intraoperative dynamic features from the intraoperative real-time monitoring data based on a sliding window includes: The sliding window is defined by a fixed time length; In response to the detection of new intraoperative real-time monitoring data, the sliding window is moved along the time series, and the intraoperative dynamic features within the current sliding window are recalculated.

4. The method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery according to claim 3, characterized in that, The recalculation of intraoperative dynamic features within the current sliding window includes: Calculate the mean, maximum, minimum, median, standard deviation, variance, and quartiles within the current sliding window to obtain statistical characteristics; Determine the maximum upward speed, maximum downward speed, slope of change, cumulative change, and direction of change within the current sliding window to obtain the trend characteristics of the change; Calculate the coefficient of variation, mean of consecutive differences, fluctuation amplitude, and number of abnormal fluctuations within the current sliding window to obtain the volatility characteristics; The duration of hypotension, hypoxia, and body temperature below a preset threshold within the current sliding window are statistically analyzed to obtain duration characteristics. The changes in anesthetic drug infusion rate, the start time of vasoactive drug administration, the number of vasoactive drug dose adjustments, the infusion rate, and the occurrence time of important surgical events within the current sliding window are extracted to obtain treatment intervention characteristics. Based on the statistical characteristics, the trend characteristics, the volatility characteristics, the duration characteristics, and the treatment intervention characteristics, the intraoperative dynamic characteristics corresponding to the current sliding window are formed.

5. The method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery according to claim 1, characterized in that, Before acquiring the static clinical data and intraoperative real-time monitoring data of the target patient, the allogeneic transfusion risk prediction model was pre-trained in the following manner: Acquire historical perioperative case data; wherein, the historical perioperative case data includes static clinical data of historical patients, intraoperative real-time monitoring data, and results of allogeneic blood transfusion; The historical perioperative case data were processed using the same static feature extraction, sliding window dynamic feature extraction, and feature stitching methods as the target patient, and training samples were constructed using the allogeneic blood transfusion results as supervised learning labels. The training samples are divided into a training set, a validation set, and a test set. The training set is used to learn the model parameters, the validation set is used to adjust the model hyperparameters, and the test set is used to evaluate the model performance. The trained model parameters are saved to the prediction server to obtain the allogeneic transfusion risk prediction model.

6. The method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery according to claim 5, characterized in that, The historical perioperative case data is processed using the same static feature extraction, sliding window dynamic feature extraction, and feature concatenation methods as the target patient, and training samples are constructed using the allogeneic blood transfusion results as supervised learning labels, including: Historical allogeneic transfusion event data of patients are obtained from the blood transfusion management system and nursing records; wherein, the allogeneic transfusion event data includes whether allogeneic transfusion occurred and the transfusion start time; The presence or absence of allogeneic blood transfusion in the allogeneic blood transfusion event data is used as the supervised learning label, and the supervised learning label corresponding to each sliding window is determined by backtracking based on the blood transfusion start time; The training samples are obtained by combining the hybrid feature vectors corresponding to each sliding window with the supervised learning labels.

7. The method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery according to claim 1, characterized in that, The step of determining the risk level and risk warning status based on the continuously updated allogeneic blood transfusion risk probability and preset risk threshold includes: The probability of allogeneic blood transfusion risk corresponding to multiple consecutive sliding windows is determined. When the probability of allogeneic blood transfusion risk corresponding to multiple consecutive sliding windows exceeds the preset risk threshold, the risk warning state is determined as the trigger state and the risk warning is output. When the vital signs of the target patient stabilize and the probability of allogeneic blood transfusion risk decreases, the risk warning status is set to canceled and the risk warning is cancelled.

8. The method for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery according to claim 1, characterized in that, The method further includes: Feature importance analysis was performed on the risk probability of allogeneic blood transfusion to determine the contribution of each feature among the static features and the intraoperative dynamic features to the risk probability of allogeneic blood transfusion. Based on the degree of contribution, key factors affecting the risk of allogeneic blood transfusion were screened; The key factors and their contribution levels are output to the clinical anesthesia information system.

9. A system for predicting the risk of perioperative allogeneic blood transfusion in cardiac surgery, characterized in that, include: The data acquisition module is used to acquire the patient's static clinical data and intraoperative real-time monitoring data, and output the static clinical data and the intraoperative real-time monitoring data to the artificial intelligence prediction module. The artificial intelligence prediction module is used to extract static features from the static clinical data, extract intraoperative dynamic features from the intraoperative real-time monitoring data based on the sliding window, concatenate the static features and the intraoperative dynamic features according to the feature dimension to obtain the mixed feature vector corresponding to each sliding window, input each mixed feature vector into the pre-trained allogeneic transfusion risk prediction model in sequence to obtain the allogeneic transfusion risk probability updated with the sliding window, and output the allogeneic transfusion risk probability to the clinical early warning and decision recommendation module. The clinical early warning and decision recommendation module is used to determine the risk level based on the allogeneic blood transfusion risk probability and a preset risk threshold. When the allogeneic blood transfusion risk probability corresponding to multiple consecutive sliding windows exceeds the preset risk threshold, a risk warning is triggered. The module then generates a processing suggestion corresponding to the risk level by combining the medical knowledge base, and outputs the risk level, the risk warning, and the processing suggestion to the clinical anesthesia information system.

10. The prediction system for perioperative allogeneic blood transfusion risk in cardiac surgery according to claim 9, characterized in that, The allogeneic blood transfusion risk prediction model in the artificial intelligence prediction module is encapsulated as a microservice, and the microservice is embedded into the clinical anesthesia information system through a medical data standard protocol. The artificial intelligence prediction module is also used to: perform feature importance analysis on the risk probability of allogeneic blood transfusion, determine the contribution of each feature in the static features and the intraoperative dynamic features to the risk probability of allogeneic blood transfusion, screen key factors affecting the risk of allogeneic blood transfusion, and output the key factors and the contribution of the key factors to the clinical early warning and decision recommendation module. The clinical early warning and decision recommendation module outputs the risk level, the risk warning, and the treatment suggestion in the following manner: a dynamic risk dashboard is generated on the interface of the clinical anesthesia information system, which displays the probability of allogeneic blood transfusion risk, the trend of the probability of allogeneic blood transfusion risk, the risk level, the key factors, the contribution degree of the key factors, and the treatment suggestion, and generates a pop-up reminder for the risk warning.

Images