A method for constructing an anesthesia complication prediction model based on deep learning

Abstract

The present invention relates to the field of medical health, and specifically provides a method for constructing an anesthesia complication prediction model based on deep learning. The method comprises the following steps: obtaining a historical multidimensional data set, preprocessing the historical multidimensional data set and annotating it; the historical multidimensional data set includes a preoperative data set, an intraoperative data set and a postoperative data set; extracting key features that can characterize anesthesia complications from the preoperative data set, the intraoperative data set and the postoperative data set respectively; constructing a time series prediction network, and training the time series prediction network using the extracted key features to obtain an anesthesia complication prediction model; and testing and verifying the trained anesthesia complication prediction model; inputting preoperative data, intraoperative data or postoperative data into the trained anesthesia complication prediction model to predict anesthesia complications at different stages. The present invention can effectively predict anesthesia complications that may occur to patients during surgery.

Description

Technical Field

[0001] The present invention relates to the field of medical information technology, and in particular to a method for constructing an anesthesia complication prediction model based on deep learning. Background Art

[0002] Anesthesia is an indispensable part of the surgical process. However, complications that may occur during anesthesia, such as hypotension and arrhythmia, will not only affect the surgical effect, but may also threaten the patient's life safety. Therefore, how to accurately predict and effectively prevent the occurrence of anesthesia complications has always been a research hotspot in the medical field.

[0003] Traditional prediction methods for anesthesia complications mainly rely on the doctor's experience and the patient's medical history information. This method has problems such as strong subjectivity and low prediction accuracy. In recent years, with the development of deep learning technology, its application in the medical field has also achieved remarkable results. Deep learning can automatically extract features from a large amount of data and build models to achieve accurate prediction of anesthesia complications. However, existing prediction methods based on deep learning still have problems such as complex model structure, large amount of calculation, and low prediction accuracy.

[0004] For example, the application number is CN116312958B, which is an anesthesia risk warning system, an emergency management system and a method. The anesthesia risk warning system includes: a monitoring terminal; a monitoring terminal; wherein the monitoring terminal obtains the monitoring data of the patient during anesthesia, stores the monitoring data, and transmits it to the monitoring terminal in encrypted form; the monitoring terminal receives and verifies the monitoring data, predicts the monitoring data of the patient at the target time based on the monitoring data at the current time, and performs an anesthesia risk warning based on the monitoring data at the current time and the monitoring data at the target time. The present invention transmits the patient's monitoring data securely through the monitoring terminal and the monitoring terminal, accurately predicts the patient's anesthesia risk, and monitors the duty status of the on-duty anesthesiologist, and sequentially realizes the system design of the entire process from wireless monitoring data collection, anesthesia risk prediction to emergency treatment. Although the above invention can provide anesthesiologists with comprehensive, safe and efficient anesthesia services, the early warning system only predicts future risks based on current monitoring data, and cannot fully capture all risk factors, especially when facing the complex and changeable physiological state of patients. Therefore, the accuracy of the prediction may be limited.

[0005] Therefore, the present invention proposes a method for constructing an anesthesia complication prediction model based on deep learning, aiming to solve the above problems and improve the accuracy and efficiency of anesthesia complication prediction. Summary of the invention

[0006] The present invention provides a method for constructing an anesthesia complication prediction model based on deep learning, which is used to improve the accuracy and efficiency of anesthesia complication prediction.

[0007] The present invention provides a method for constructing an anesthesia complication prediction model based on deep learning, comprising the following steps:

[0008] Obtain a historical multidimensional data set, preprocess the historical multidimensional data set and annotate it; wherein,

[0009] The historical multidimensional data set includes a preoperative data set, an intraoperative data set and a postoperative data set;

[0010] Extracting key features capable of characterizing anesthetic complications from the preoperative data set, the intraoperative data set, and the postoperative data set respectively;

[0011] Construct a time series prediction network, and use the extracted key features to train the time series prediction network to obtain an anesthesia complication prediction model; and test and verify the trained anesthesia complication prediction model;

[0012] Preoperative data, intraoperative data or postoperative data are input into the trained anesthesia complication prediction model to predict anesthesia complications at different stages.

[0013] In an practicable manner, the steps of obtaining the historical multidimensional data set and preprocessing the historical multidimensional data set are specifically as follows:

[0014] Obtain preoperative, intraoperative and postoperative data of multiple patients;

[0015] Using the patient identifier, the preoperative data, intraoperative data and postoperative data related to the patient are associated and merged to form a multidimensional data set containing all the clinical data of the patient. The data in the multidimensional data set is arranged in chronological order; the historical multidimensional data set is composed of the multidimensional data of multiple patients;

[0016] Perform missing value processing, outlier detection and processing, and formatting on historical multidimensional data sets;

[0017] Divide the processed historical multidimensional data set according to the time sequence and define the preoperative, intraoperative and postoperative time windows respectively;

[0018] Use data monitoring mechanisms to regularly check the data quality of historical multidimensional datasets.

[0019] In an practicable manner, the step of extracting key features capable of characterizing anesthesia complications from the preoperative data set, the intraoperative data set, and the postoperative data set respectively comprises:

[0020] The initial time window is set according to the surgical process and surgical time nodes; the key events that affect the risk of anesthetic complications are identified from the preoperative data, intraoperative data, and postoperative data, and the size and position of the time window are dynamically adjusted according to the identified key events; the key features are extracted using the adaptively adjusted time window.

[0021] In an implementable manner, in the step of extracting key features using the adaptively adjusted time window, features of each stage are extracted using correlation analysis;

[0022] Among them, the extracted preoperative features include features reflecting the patient's baseline health and pre-anesthetic risk factors within the preoperative time window; the extracted intraoperative features include features covering the surgical process and anesthesia management within the intraoperative time window; the extracted postoperative features include features reflecting recovery dynamics and complication risks within the postoperative time window.

[0023] In one implementable method, the preoperative characteristics include physiological and biochemical indicators, cardiopulmonary function indicators, past medical history, medication history, physical characteristics, living habits and psychological state characteristics; the intraoperative characteristics include anesthetic drugs, vasoactive drugs, surgical operations, real-time vital signs, and surgical complications; the postoperative characteristics include recovery status, signs of complications, drug reactions, physiological indicator trends, and readmission status.

[0024] In one practicable manner, the time series prediction network structure includes three prediction network layers, a fusion layer and an output layer, and each prediction network layer includes an LSTM layer, convolutional attention and at least one hidden layer.

[0025] In an implementable manner, the anesthesia complication prediction model adopts a hierarchical training method, and the steps of training the time series prediction network using the key features to be extracted to obtain the anesthesia complication prediction model include:

[0026] Determine the length of the sequence in the preoperative, intraoperative, and postoperative phases;

[0027] Using a rolling window, the key features of each stage are divided into training set, validation set and test set;

[0028] Use the training set data to train the time series prediction network; in each stage, input the time series data of the corresponding stage and adjust the weights according to the output of the model;

[0029] The sequence length was dynamically adjusted based on the feedback from the anesthetic complication prediction model.

[0030] In an practicable manner, after the step of inputting the preoperative data, intraoperative data or postoperative data into the trained anesthesia complication prediction model to predict anesthesia complications at different stages, the method further comprises:

[0031] Obtain the actual occurrence of complications and treatment effects of patients as feedback information; use the feedback information to adjust and optimize the model's prediction parameters and the generation logic of personalized recommendations, forming a closed loop of continuous improvement.

[0032] In an practicable manner, after the step of inputting the preoperative data, intraoperative data or postoperative data into the trained anesthesia complication prediction model to predict anesthesia complications at different stages, the method further comprises:

[0033] The output results of the anesthesia complication prediction model are converted into understandable risk scores, and personalized recommendation information is generated based on the risk scores of each stage and the preset recommendation generation rules.

[0034] In an implementable manner, the suggestion generation rules include threshold determination, risk level classification, and suggestion content matching.

[0035] The beneficial effects that can be achieved by the present invention are: a method for constructing an anesthesia complication prediction model based on deep learning technology of the present invention has high accuracy and practicality, can effectively predict the anesthesia complications that may occur in patients during surgery, provides an important auxiliary tool for clinical anesthesiologists, reduces medical risks, and improves treatment effects.

[0036] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0037] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0039] Figure 1 This is a flowchart of a method for constructing an anesthesia complication prediction model based on deep learning in Example 1 of the present invention;

[0040] Figure 2 This is a flow chart for extracting key features capable of characterizing anesthesia complications in Example 1 of the present invention;

[0041] Figure 3 This is a flowchart of training an anesthesia complication prediction model in Example 1 of the present invention;

[0042] Figure 4 Charts of different indicators of general anesthesia for patients during medical procedures;

[0043] Figure 5 This is a schematic diagram of a patient with no obvious abnormalities in his chest X-ray during a medical procedure;

[0044] Figure 6 This is a schematic diagram of the patient's right diaphragmatic top of the lung during medical surgery, after anesthesia induction and intubation, and after the end of the surgery. DETAILED DESCRIPTION

[0045] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0046] During medical surgeries, many surgeries require anesthesia for patients. Although anesthesia can help doctors perform surgery and help patients lose consciousness, it also has some risks. General anesthesia is divided into three stages: anesthesia induction, anesthesia maintenance, and anesthesia awakening. Under normal circumstances, blood pressure, heart rate, and bis values ​​will drop immediately after anesthesia induction, mainly due to the inhibition of circulation by anesthetics; during anesthesia, blood pressure, pulse, and heart rate must be maintained within an appropriate range, that is, to maintain hemodynamic stability. If there are drastic hemodynamic fluctuations, or blood pressure and heart rate are difficult to maintain, vasoactive drugs are needed to maintain blood pressure. The Bis value reflects the degree of sedation. Previous studies have shown that low mean arterial pressure, low bis, and low Mac (minimum effective alveolar gas concentration) are closely related to increased postoperative mortality. As shown in the attached Figure 4 As shown. During general anesthesia, the patient's physical indicators need to be tested. Routine test indicators during general anesthesia include blood pressure, pulse, heart rate, oxygen saturation, end-tidal carbon dioxide, and bispectral index (bis); critically ill patients undergo regular blood gas analysis, which includes arterial oxygen partial pressure, carbon dioxide partial pressure, blood pH, lactate, potassium, sodium, calcium and other electrolyte levels. The above indicators can be used to determine whether there are anesthetic complications. Anesthesia-related complications mainly include: cardiovascular and cerebrovascular accidents, malignant hyperthermia, restlessness during the recovery period, delayed recovery, postoperative nausea and vomiting, lung complications, etc., which are generally not closely related to the patient's resected part. The main complications that can be reflected by imaging are postoperative lung complications (pleural effusion, atelectasis, pneumothorax, etc.). Schematic diagram of the lungs after anesthesia, as shown Figure 5 As shown in the figure, 3 hours after surgery, the translucency of both lungs decreased, and flake-like ground-glass density shadows were visible, and the heart shadow was not large; 19 hours after surgery, the chest X-ray showed diffuse patchy high-density shadows in both lungs, and the hilum of both lungs was not large; 64 hours after surgery, the chest X-ray showed no obvious abnormalities in both lungs, heart, and diaphragm. During anesthesia, if there is no abnormality, such as Figure 6Shown are CT images of the 1 cm layer of the right diaphragm before anesthesia, after anesthesia induction and intubation, and after the end of surgery. The arrow points to the atelectasis area, indicating that the patient's lungs are normal.

[0047] This embodiment 1 provides a method for constructing an anesthesia complication prediction model based on deep learning, comprising the following steps:

[0048] Obtain a historical multidimensional data set, preprocess the historical multidimensional data set and annotate it; wherein,

[0049] The historical multidimensional data set includes a preoperative data set, an intraoperative data set and a postoperative data set;

[0050] Extracting key features capable of characterizing anesthetic complications from the preoperative data set, the intraoperative data set, and the postoperative data set respectively;

[0051] Construct a time series prediction network, and use the extracted key features to train the time series prediction network to obtain an anesthesia complication prediction model; and test and verify the trained anesthesia complication prediction model;

[0052] Preoperative data, intraoperative data or postoperative data are input into the trained anesthesia complication prediction model to predict anesthesia complications at different stages.

[0053] Working principle of the above technical solution: see Figure 1 , this application first obtains a historical multidimensional data set containing patient medical records, physiological indicators, drug usage and other information. The historical multidimensional data set can come from hospital databases, clinical trial records, etc. And preprocess and annotate the historical multidimensional data set. Preprocessing can improve the quality of subsequent analysis and modeling. When annotating, manual annotation or machine learning automatic annotation methods can be used to annotate the type of anesthetic complications for each sample in the data set. Then, key features are extracted from the historical multidimensional data set. For the annotated historical multidimensional data set, key features related to anesthetic complications can be extracted through cluster analysis, association rule mining and other technologies. Then, the extracted key features are combined with the time series prediction network to build an anesthetic complication prediction model. Finally, the labeled key features are input into the constructed time series prediction model, and the model is trained, and the trained anesthetic complication prediction model is tested and verified. By inputting preoperative data, intraoperative data or postoperative data into the trained anesthetic complication prediction model, anesthetic complications at different stages can be predicted.

[0054] The beneficial effects of the above technical solution: The method for constructing an anesthesia complication prediction model of the present invention can effectively predict the risk probability of anesthesia complications for patients during surgery, providing a powerful support tool for clinical anesthesiologists, helping to reduce medical risks and improve treatment effects.

[0055] Based on Example 1, the steps of obtaining the historical multidimensional data set and preprocessing the historical multidimensional data set are specifically as follows:

[0056] Obtain preoperative, intraoperative and postoperative data of multiple patients;

[0057] Using the patient identifier, the preoperative data, intraoperative data and postoperative data related to the patient are associated and merged to form a multidimensional data set containing all the clinical data of the patient. The data in the multidimensional data set is arranged in chronological order; the historical multidimensional data set is composed of the multidimensional data of multiple patients;

[0058] Perform missing value processing, outlier detection and processing, and formatting on historical multidimensional data sets;

[0059] Divide the processed historical multidimensional data set according to the time sequence, and define the preoperative, intraoperative and postoperative time windows respectively to obtain data sets suitable for different analysis stages;

[0060] Use data monitoring mechanisms to regularly check the data quality of historical multidimensional datasets.

[0061] The working principle and beneficial effects of the above technical solution are as follows: First, the preoperative data, intraoperative data and postoperative data of multiple patients are obtained; then, the preoperative data, intraoperative data and postoperative data related to the patient are associated and merged using the patient identifier, and the association between the data is established through the patient identifier. Next, the historical multidimensional data set is processed for missing values, detected and processed for outliers, and formatted. Through the above processing steps, the quality and consistency of the data are guaranteed. For missing values, strategies such as filling, deleting or replacing can be selected. For outliers, detection and processing are required to avoid affecting the model. Then, the processed historical multidimensional data set is divided according to the time sequence, and the time windows of preoperative, intraoperative and postoperative are defined respectively, so as to obtain the training set, validation set and test set suitable for different analysis stages, so that the training, validation and testing of the model have a clear time process to ensure the validity and reliability of the results. Finally, the data quality of the historical multidimensional data set is regularly checked using the data monitoring mechanism. The data monitoring mechanism can detect immediately when new data is added, so that problems can be discovered and solved in time.

[0062] Based on Example 1, the step of extracting key features capable of characterizing anesthesia complications from the preoperative data set, the intraoperative data set, and the postoperative data set respectively includes:

[0063] Set the initial time window according to the surgical process and surgical time nodes;

[0064] Identify key events that affect the risk of anesthesia complications from preoperative data, intraoperative data, and postoperative data, and dynamically adjust the size and position of the time window based on the identified key events;

[0065] The key features are extracted using the adaptively adjusted time window.

[0066] Working principle of the above technical solution: see Figure 2 , this application first sets the initial time window according to the surgical process and surgical time nodes. When setting the initial window, the start time of the operation is used as the starting point, and the estimated end time of the operation is used as the end point to obtain an initial time window. Then, the preoperative data, intraoperative data, and postoperative data are used to identify key events that affect the risk of anesthetic complications, such as certain physiological changes, the use of drugs, surgical methods, etc. Then, based on the identified key events, the size and position of the time window are dynamically adjusted. Based on the identification results of the key events, the size and position of the time window are adjusted to better capture possible changes. Finally, the adaptively adjusted time window is used to extract the key features that best represent the risk of anesthetic complications to improve the effect of the final prediction model.

[0067] Based on Example 1, in the step of extracting key features using the adaptively adjusted time window, correlation analysis is used to extract features of each stage; the extracted preoperative features include features reflecting the patient's baseline health and pre-anesthesia risk factors within the preoperative time window; the extracted intraoperative features include features covering the surgical process and anesthesia management within the intraoperative time window; the extracted postoperative features include features reflecting recovery dynamics and complication risks within the postoperative time window.

[0068] Among them, the preoperative characteristics include physiological and biochemical indicators, cardiopulmonary function indicators, past medical history, drug history, physical characteristics, living habits and psychological state characteristics; the intraoperative characteristics include anesthetic drugs, vasoactive drugs, surgical operations, real-time vital signs, and surgical complications; the postoperative characteristics include recovery status, signs of complications, drug reactions, physiological indicator trends, and readmission status.

[0069] The working principle and beneficial effects of the above technical solution: In the process of extracting features using the adaptively adjusted time window, the corresponding features will be extracted in a targeted manner according to the different characteristics of the three stages: preoperative, intraoperative and postoperative. Preoperative features include features that reflect the patient's baseline health and pre-anesthetic risk factors within the preoperative time window; intraoperative features include features that cover the surgical process and anesthesia management within the intraoperative time window; extracted postoperative features include features that reflect recovery dynamics and complication risks within the postoperative time window.

[0070] Specifically, preoperative characteristics include physiological and biochemical indicators: such as heart rate, blood pressure, respiratory rate, etc. These characteristics will directly affect the patient's anesthesia process. Past medical history: Understand the patient's health status and past medical conditions in order to predict possible complications. Drug history: The use of drugs may directly affect the patient's anesthesia process, so this is also an important feature. Living habits: such as smoking, drinking, and whether there is a history of allergies, etc., all of which will affect the anesthesia process. Psychological state characteristics: The patient's anxiety, depression and other psychological states will have an impact on the anesthesia process. As well as physical characteristics and cardiopulmonary function indicators. Intraoperative characteristics include anesthetic drugs: the type and dosage of anesthetic drugs used, which are factors that directly affect the anesthesia process. Vasoactive drugs, surgical operations: the specific process and method of the operation, etc., these are important factors in the operation process. Real-time vital signs: such as oxygen saturation, electrocardiogram, blood oxygen, etc., these are vital signs that must be closely monitored during the operation. Surgical complications: complications during the operation, which is very important for predicting postoperative complications. Postoperative characteristics include recovery: the patient's body temperature, heart rate, blood pressure and other recovery conditions, these are indicators that need to be closely observed after surgery. Signs of complications: Symptoms such as nausea, vomiting, and pain may be precursors to complications. Drug reaction: The patient's response to anesthetic drugs after surgery, which is important for adjusting the medication regimen. Physiological indicator trends: Changes in physiological indicators such as blood oxygen, heart rate, and blood pressure. Readmission: The patient's re-hospitalization after surgery, which can reflect the continuity and effectiveness of treatment.

[0071] Based on Example 1, the time series prediction network structure includes three prediction network layers, a fusion layer and an output layer, and each prediction network layer includes an LSTM layer, convolutional attention and at least one hidden layer.

[0072] Working principle of the above technical solution: This application adopts three independent LSTM (Long Short-Term Memory Network), and the data of each stage is first processed by its own independent LSTM to capture the time series characteristics. Three attention mechanism layers, followed by independent LSTM, weight the key time points of each stage. The hidden layer is located after the attention mechanism layer, and there can be one or more. These layers receive the weighted output from the attention mechanism layer and perform further feature extraction and integration. The hidden layer can be a fully connected layer, a convolutional layer, or other type of neural network layer, depending on the model design and task requirements. Fusion layer: For the intraoperative and postoperative stages, the fusion layer combines the output of the LSTM layer of the previous stage (weighted by attention) with the output of the current stage to achieve information fusion. The fusion layer can also be regarded as part of the hidden layer, especially when designing a multi-level fusion structure. Finally, the output of the fusion layer (or the last layer of the hidden layer) enters the output layer, which generates the final prediction result.

[0073] Beneficial effects of the above technical solution: The time series prediction network structure of the present application not only takes into account the characteristics of time series data, but also fully integrates information of different modalities, has strong expression ability, and can be used to predict various time series data. At the same time, the model also has good scalability and can be applied to the prediction of various anesthesia complications.

[0074] On the basis of Example 1, the anesthesia complication prediction model adopts a hierarchical training method, and the steps of training the time series prediction network using the key features to be extracted to obtain the anesthesia complication prediction model include:

[0075] Determine the length of the sequence in the preoperative, intraoperative, and postoperative phases;

[0076] Using a rolling window, the key features of each stage are divided into training set, validation set and test set;

[0077] Use the training set data to train the time series prediction network; in each stage, input the time series data of the corresponding stage and adjust the weights according to the output of the model;

[0078] The sequence length was dynamically adjusted based on the feedback from the anesthetic complication prediction model.

[0079] Working principle of the above technical solution: see Figure 3First, the sequence lengths of the preoperative, intraoperative, and postoperative stages are determined to ensure that the prediction model can adapt well to the characteristics of different stages. When determining the sequence strength, it is preferred to increase the sequence length of the postoperative stage, because the data in the postoperative stage is often richer than that in the preoperative and intraoperative stages and is more conducive to model training. Then, using a rolling window, the key feature data of each stage is divided into a training set, a validation set, and a test set. The training set is used to train the model, the validation set is used to test the performance of the model, and the test set is used to evaluate the generalization ability of the model on unknown data. Next, the time series prediction network is trained using the training set data. This process is usually carried out step by step. First, the time series data of the corresponding stage is input and the prediction results are output. Then, according to the output results of the model, the weight of the model is adjusted to improve the prediction performance of the model. Finally, the sequence length is dynamically adjusted according to the feedback of the anesthesia complication prediction model.

[0080] Beneficial effects of the above technical solution: The training process of the anesthesia complications prediction model of the present application helps to improve the prediction accuracy, generalization ability and clinical practicality of the model, and provides strong technical support for anesthesia management.

[0081] On the basis of Example 1, after the step of inputting the preoperative data, intraoperative data or postoperative data into the trained anesthesia complication prediction model to predict anesthesia complications at different stages, the method further includes:

[0082] Obtain the actual occurrence of complications and treatment effects of patients as feedback information; use the feedback information to adjust and optimize the model's prediction parameters and the generation logic of personalized recommendations, forming a closed loop of continuous improvement.

[0083] The working principle and beneficial effects of the above technical solution: First, the actual occurrence of complications and treatment effects of patients are obtained as feedback information. It can be obtained from the electronic medical record system, medical records or clinical surveys of medical institutions. The feedback information obtained can be used to determine the accuracy of the prediction model and the effectiveness of personalized suggestions. Secondly, the feedback information is used to adjust and optimize the prediction parameters of the model and the generation logic of personalized suggestions. Specifically, the prediction performance of the model can be improved by changing the architecture of the model, adjusting hyperparameters, adding new features, etc. At the same time, the generation logic of the personalized suggestions provided can also be adjusted according to the feedback information. For example, more specific drug suggestions or surgical plan recommendations may be provided based on the patient's actual condition and medication. Finally, this process needs to be continuously iterated so that the model can continue to learn and optimize. Each feedback can help discover the shortcomings of the model and then improve the model to make it more accurate and reliable.

[0084] On the basis of Example 1, after the step of inputting the preoperative data, intraoperative data or postoperative data into the trained anesthesia complication prediction model to predict anesthesia complications at different stages, the method further includes:

[0085] The output results of the anesthesia complication prediction model are converted into understandable risk scores, and personalized recommendation information is generated based on the risk scores of each stage and the preset recommendation generation rules.

[0086] Among them, the recommendation generation rules include threshold judgment, risk level classification and recommendation content matching.

[0087] The working principle of the above technical solution is: by converting the prediction results of the model into a risk score that reflects the possibility of anesthetic complications in patients at each stage. The higher the score, the greater the risk of anesthetic complications in patients at this stage. Conversely, the lower the score, the lower the risk of anesthetic complications in patients. Then, personalized advice information can be generated based on the risk score of each stage and the preset advice generation rules. For example, if a patient is at a certain point after surgery and the model's prediction results show that the patient is at a high risk of anesthetic complications in the next few days, then some additional advice can be provided to the doctor or nurse, such as avoiding certain behaviors or practices that may cause complications, or taking some special care measures to reduce the incidence of complications.

[0088] Beneficial effects of the above technical solution: This application enables doctors and nurses to make optimal decisions at different stages of the operation based on the model's prediction results and recommended information to reduce medical risks.

[0089] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for constructing an anesthesia complication prediction model based on deep learning, characterized in that: The following steps are involved: Obtain a historical multidimensional data set, preprocess the historical multidimensional data set and annotate it; wherein, The historical multidimensional data set includes a preoperative data set, an intraoperative data set and a postoperative data set; Extracting key features capable of characterizing anesthetic complications from the preoperative data set, the intraoperative data set, and the postoperative data set respectively; Construct a time series prediction network, and use the extracted key features to train the time series prediction network to obtain an anesthesia complication prediction model; and test and verify the trained anesthesia complication prediction model; Inputting preoperative data, intraoperative data or postoperative data into the trained anesthesia complication prediction model to predict anesthesia complications at different stages; The steps of extracting key features capable of characterizing anesthesia complications from the preoperative data set, the intraoperative data set, and the postoperative data set include: According to the surgical process and surgical time nodes, the initial time window is set; the key events that affect the risk of anesthesia complications are identified from preoperative data, intraoperative data, and postoperative data, and the size and position of the time window are dynamically adjusted according to the identified key events; key features are extracted using the adaptively adjusted time window; In the step of extracting key features using the adaptively adjusted time window, correlation analysis is used to extract features at each stage; Among them, the extracted preoperative features include features that reflect the patient's baseline health and pre-anesthetic risk factors within the preoperative time window; the extracted intraoperative features include features that cover the surgical process and anesthesia management within the intraoperative time window; the extracted postoperative features include features that reflect recovery dynamics and complication risks within the postoperative time window; Among them, building a time series prediction network includes: The time series prediction network structure includes three prediction network layers, a fusion layer and an output layer, each of which includes an LSTM layer, a convolutional attention layer and at least one hidden layer; Three independent LSTMs are used, and the data of each stage is first processed by their own independent LSTM to capture the time series characteristics; three attention mechanism layers are followed by independent LSTMs, and the attention mechanism layer weights the key time points of each stage; the hidden layer is located after the attention mechanism layer, and there can be one or more; these layers receive the weighted output from the attention mechanism layer and perform further feature extraction and integration; the hidden layer can be a fully connected layer, a convolutional layer, or other types of neural network layers, depending on the model design and task requirements; fusion layer: for the intraoperative and postoperative stages, the fusion layer combines the output of the LSTM layer of the previous stage with the output of the current stage to achieve information fusion; the fusion layer can also be regarded as part of the hidden layer, especially when designing a multi-level fusion structure; finally, the output of the fusion layer or the last layer of the hidden layer enters the output layer, which generates the final prediction result.

2. The method for constructing an anesthesia complication prediction model based on deep learning according to claim 1, characterized in that: The steps of obtaining the historical multidimensional data set and preprocessing the historical multidimensional data set are specifically as follows: Obtain preoperative, intraoperative and postoperative data of multiple patients; Using the patient identifier, the preoperative data, intraoperative data and postoperative data related to the patient are associated and merged to form a multidimensional data containing all the clinical data of the patient, and the data in the multidimensional data are arranged in chronological order; The historical multidimensional dataset consists of multidimensional data of multiple patients; Perform missing value processing, outlier detection and processing, and formatting on historical multidimensional data sets; Divide the processed historical multidimensional data set according to the time sequence and define the preoperative, intraoperative and postoperative time windows respectively; Use data monitoring mechanisms to regularly check the data quality of historical multidimensional datasets.

3. The method for constructing an anesthesia complication prediction model based on deep learning according to claim 1, characterized in that: The preoperative characteristics include physiological and biochemical indicators, cardiopulmonary function indicators, past medical history, medication history, physical characteristics, living habits and psychological state characteristics; the intraoperative characteristics include anesthetic drugs, vasoactive drugs, surgical operations, real-time vital signs, and surgical complications; the postoperative characteristics include recovery status, signs of complications, drug reactions, physiological indicator trends, and readmission status.

4. The method for constructing an anesthesia complication prediction model based on deep learning according to claim 1, characterized in that: The anesthesia complication prediction model adopts a hierarchical training method and uses the key features to be extracted to train the time series prediction network. The steps to obtain the anesthesia complication prediction model include: Determine the length of the sequence in the preoperative, intraoperative, and postoperative phases; Using a rolling window, the key features of each stage are divided into training set, validation set and test set; Use the training set data to train the time series prediction network; in each stage, input the time series data of the corresponding stage and adjust the weights according to the output of the model; The sequence length was dynamically adjusted based on the feedback from the anesthetic complication prediction model.

5. The method for constructing an anesthesia complication prediction model based on deep learning according to claim 4, characterized in that: After the step of inputting the preoperative data, intraoperative data or postoperative data into the trained anesthesia complication prediction model to predict anesthesia complications at different stages, the method further includes: Obtain the actual occurrence of complications and treatment effects of patients as feedback information; use the feedback information to adjust and optimize the model's prediction parameters and the generation logic of personalized recommendations, forming a closed loop of continuous improvement.

6. The method for constructing an anesthesia complication prediction model based on deep learning according to claim 1, characterized in that: After the step of inputting the preoperative data, intraoperative data or postoperative data into the trained anesthesia complication prediction model to predict anesthesia complications at different stages, the method further includes: The output results of the anesthesia complication prediction model are converted into understandable risk scores, and personalized recommendation information is generated based on the risk scores of each stage and the preset recommendation generation rules.

7. The method for constructing an anesthesia complication prediction model based on deep learning according to claim 6, characterized in that: The suggestion generation rules include threshold judgment, risk level classification and suggestion content matching.

Images