Cardiothoracic surgery risk early warning method and system based on artificial intelligence

Abstract

The invention provides a cardiothoracic surgery risk early warning method and system based on artificial intelligence, and belongs to the technical field of cardiothoracic surgery. The system comprises a data input module used for collecting basic information and reports of patients, real-time monitoring signals of surgical equipment, medical image data and physiological index data of the patients and generating time-aligned standardized data streams; the feature processing module is used for receiving the standardized data stream and generating a fusion feature vector based on a unified time axis; and the risk prediction and early warning module is used for calculating complication risk probability values by applying a dual-channel neural network pre-training model in combination with the physiological indexes of the patient and the fusion features, integrating the complication risk probability values into a real-time risk probability table, and triggering early warning by comparing a risk threshold table. According to the method, the timeliness and prediction accuracy of cardiothoracic operation complication recognition are remarkably improved through real-time standardized alignment of the multi-source medical data, independent prediction of complication risks through the two-channel neural network and real-time updating and early warning of the dynamic probability table.

Description

Technical Field

[0001] The present invention relates to the technical field of cardiothoracic surgery, and in particular to an artificial intelligence-based cardiothoracic surgery risk warning method and system. Background Art

[0002] Cardiothoracic surgery relies on high-precision operations such as extracorporeal circulation and vascular anastomosis. During the operation, the patient's electrocardiogram, blood pressure, blood oxygen and organ status must be continuously monitored. Since the operation lasts for several hours and there are significant individual differences among patients, a real-time risk warning mechanism must be established to analyze and identify the risk of complications and ensure surgical safety.

[0003] The core bottlenecks of existing cardiothoracic surgery risk warning technologies are concentrated on three levels: ① The heterogeneity of multi-source medical data makes integration difficult. The raw data formats of different surgical devices vary significantly and lack a unified timeline, forcing reliance on manual calibration of timestamps and long data analysis delays. ② Traditional prediction models couple all physiological indicators into a single computing layer, making it impossible to calculate the corresponding risk probability based on different complications, making it difficult to independently predict the risk probabilities of multiple complications. ③ The existing system uses a fixed-cycle static evaluation mechanism, which makes it difficult to respond in real time to emergencies during surgery. Therefore, it is urgent to build an intelligent cardiothoracic surgery warning system that integrates real-time alignment of multi-source data, accurate prediction of independent complications, and dynamic risk assessment, so as to realize an intelligent warning system from raw data synchronization processing, feature fusion analysis to real-time warning triggering. Summary of the Invention

[0004] In order to solve the technical problems mentioned in the current background technology, the present invention proposes an artificial intelligence-based cardiothoracic surgery risk warning method and system.

[0005] To this end, the technical solution adopted in the present invention is as follows: An artificial intelligence-based cardiothoracic surgery risk warning system, characterized in that the system includes: M1, data input module, collects original data including basic patient information and reports, real-time monitoring signals of surgical equipment, medical imaging data, and patient physiological index data throughout the entire cardiothoracic surgery process; standardizes the basic patient information and reports, real-time monitoring signals of surgical equipment, and medical imaging data to generate a standardized data stream, which is then transmitted to the feature processing module and the risk prediction and early warning module; M2, a feature processing module, receives the standardized data stream, extracts key features, performs fusion processing based on a unified time axis, generates a fusion feature vector, and transmits it to the risk prediction and warning module; M3, risk prediction and warning module, receives the fused feature vector and the patient's physiological indicators, applies the pre-trained prediction model, calculates the risk probability value of the patient developing cardiothoracic surgery complications, and integrates it into a risk probability table; compares it with the risk threshold table to make judgments and warnings.

[0006] Furthermore, the standardization process includes unifying the format of the original data and correcting the acquisition time point; The patient's basic information and report and the medical imaging data are pre-stored before the cardiothoracic surgery, and the surgical equipment's real-time monitoring signals and the patient's physiological indicators are updated in real time during the cardiothoracic surgery; After processing, a standardized data stream with consistent time points is generated, which contains standardized and time-aligned patient basic information and reports, real-time monitoring signals of surgical equipment, and medical imaging data.

[0007] Furthermore, the key feature extraction process includes structured data generation, dynamic data generation and image data generation. The structured data generation process is as follows: According to the standardized data stream, identify the patient basic information and the text content in the report, and set the text content as a string set ,in, For the Article text content, Match the string set with relevant medical terms to determine whether the string contains medical terms, and define medical terms as a set ,in, For the Medical terms, The criteria for determination are: ; like Include ,but , otherwise, ; After successful matching, the associated numerical value of the corresponding medical term is extracted. The formula is as follows: ; in, For the The associated value of the item; Finally, structured data is generated based on the medical terms and associated values , .

[0008] Furthermore, the dynamic data generation process is as follows: According to the standardized data stream, the real-time monitoring signal of the surgical equipment is set to ,in, For the Real-time monitoring signals of surgical equipment, The total number of sampling points in the real-time monitoring signal of the surgical equipment, The total number of sampling points is divided into time windows according to fixed time. The formula is: ; in, is the duration of the fixed time split window, is the sampling frequency, is the number of time windows; After obtaining the number of time windows, the signal window calculation is performed, and the formula is: ; in, is the number of points contained in a single time window, Calculated; For the Real-time monitoring signals of surgical equipment; For the signal windows; Calculate the range, rate of change and window mean for each time window. The calculation formula of the range value is: ; in, For the extreme values; The calculation formula of the change rate is: ; in, For the The signal window sampling points, For the rate of change; The calculation formula of the window mean is: ; in, is the window mean; Finally, dynamic data is generated based on the range, rate of change and fluctuation range. , .

[0009] Furthermore, the image data generation process is as follows: According to the standardized data stream, the medical imaging data is set as a two-dimensional matrix , identify the organ contour line in the medical image data, and calculate the number of pixels within the organ contour line, the formula is: ; in, is the image width, is the image height, is the pixel coordinate, is the number of pixels; The physical area and average gray value of the contour line are calculated using the number of pixels. The formula for calculating the physical area is: ; in, is the width of a single pixel, is the height of a single pixel, is the number of pixels, is the physical area; The average gray value calculation formula is: ; in, is the pixel coordinate, For coordinates The gray value at is the area within the organ outline, is the average gray value; Finally, the image data is obtained by combining the physical area and the average grayscale. , .

[0010] Furthermore, the unified time axis is defined as: ; in, For the Time nodes, is the total number of time nodes, To unify the timeline collection; The fused feature vector is expressed as: ; in, For the Structured data of time nodes, For the Dynamic data of time nodes, For the Image data of time nodes, For the The fusion feature vector of each time node; Finally, the output fusion feature vector .

[0011] Furthermore, the basic architecture of the prediction model is: The input of the prediction model is the patient's physiological indicators; Before the patient's physiological indicators are input into the prediction model, they need to be time-aligned with the fusion feature vector to generate ; The prediction model adopts a dual-channel neural network structure, accesses the fusion feature vector, and independently calculates the corresponding risk probability value of each cardiothoracic surgery complication.

[0012] Furthermore, the dual-channel neural network structure has three layers: The first layer serves as the input layer, receiving the time points The patient's physiological indicators , the expression is ,in, For the Physiological indicators of patients; The second layer is the diversion layer, which includes the left channel and the right channel. The left channel uses a self-attention model to calculate the correlation matrix between all the patient's physiological indicators. The elements of the correlation matrix are used as the influence probability between the physiological indicators. The formula is: ; in, 、 is the trainable weight matrix; is the feature dimension; is the correlation matrix; The first Rank Elements of a column.

[0013] Indicates the physiological indicators right Probability of impact; The right channel adopts a routing mechanism to extract the strongly correlated physiological indicators corresponding to specific complications from the patient's physiological indicators and directly allocate them to the corresponding complication clinics on the third floor. Through a predefined medical routing matrix Complete the allocation of the strongly correlated physiological indicators, Is a binary routing flag, representing the medical routing matrix No. Rank Elements of the column; The allocation decision is made according to the following routing rules: ; when When The patient's physiological indicators are assigned to Complications clinic; The third layer is the diagnosis layer. Two independent complication clinics work simultaneously. Each clinic only studies designated complications and independently calculates the risk probability value of each different complication. For each different complication, first calculate the weighted outcome value of the complication, The time point The formula for calculating the weighted outcome value of each complication is: ; in, For the The weighted outcome value of each complication, The first List, is the curing parameter; The weighted result value is input into the sigmoid function to obtain the risk probability value of the corresponding complication: ; in, For the The risk probability of complications, is a natural constant.

[0014] Furthermore, according to the risk probability values of the corresponding complications, a risk probability table containing time nodes is integrated. According to the risk probability table and the preset risk threshold table, the risk probability values of all complications are determined in turn. If a specific complication is higher than the corresponding risk threshold, an early warning is triggered.

[0015] The present invention also proposes an artificial intelligence-based cardiothoracic surgery risk warning method, characterized in that the method comprises the following steps: S1. Data input: collecting raw data including basic patient information and reports, real-time monitoring signals from surgical equipment, medical imaging data, and patient physiological index data throughout the entire cardiothoracic surgery process; standardizing the basic patient information and reports, real-time monitoring signals from surgical equipment, and medical imaging data to generate a standardized data stream; S2, a feature processing module, receives the standardized data stream, extracts key features, performs fusion processing based on a unified time axis, and generates a fused feature vector; S3, risk prediction and warning module, receives the fused feature vector and the patient's physiological indicators, applies the pre-trained prediction model, calculates the risk probability value of the patient developing cardiothoracic surgery complications, and integrates it into a risk probability table; compares it with the risk threshold table to make a judgment and warning.

[0016] Compared with the prior art, the advantages of the present invention are: 1. The present invention unifies the raw data format of surgical equipment through standardized processing and corrects the time point, thereby solving the problem of raw data heterogeneity, generating time-aligned standardized data streams, and realizing real-time synchronous flow of raw data, avoiding conflicts caused by time delays.

[0017] 2. The present invention adopts a dual-channel neural network structure to calculate the complication risk probability value. The left channel analyzes the correlation between physiological indicators, and the right channel realizes the diversion of physiological indicators and distributes them to the corresponding complication clinics. The corresponding complication clinics calculate and output the risk probability value, realizing independent prediction of multiple complication risk probabilities and avoiding the error interference of traditional prediction models.

[0018] 3. The present invention integrates the prediction results in real time to generate a risk probability table containing time nodes, and compares the risk probability table with the risk threshold table based on real-time updates. When the risk probability value exceeds the standard, the early warning plan is immediately triggered to achieve real-time response to emergencies during surgery and improve the early warning response capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 It is a schematic diagram of the overall process of the present invention; Figure 2 This is a working diagram of the feature processing module of the present invention; Figure 3 This is a working diagram of the risk prediction and early warning module of the present invention; Figure 4 It is a medical imaging data diagram of the present invention; Figure 5 This is a real-time detection signal diagram of the surgical equipment of the present invention. DETAILED DESCRIPTION

[0021] To achieve the above objectives, the present invention is implemented through the following technical solutions. The present invention provides an artificial intelligence-based cardiothoracic surgery risk warning system, which includes: M1, data input module, collects the original data including basic patient information and reports, real-time monitoring signals of surgical equipment, medical imaging data and patient physiological index data during the whole process of cardiothoracic surgery; standardizes the basic patient information and reports, real-time monitoring signals of surgical equipment and medical imaging data, generates standardized data streams, and transmits them to the feature processing module and risk prediction and early warning module; Figure 1 .

[0022] The standardization process includes unifying the format of the original data and correcting the acquisition time point; The patient's basic information and reports and medical imaging data are pre-stored before the cardiothoracic surgery, and the surgical equipment's real-time monitoring signals and the patient's physiological indicators are updated in real time during the cardiothoracic surgery; In this embodiment, the patient's physiological indicators include Blood pressure = 120, blood oxygen = 98%, urine output = 50ml / h, ... ; After processing, a standardized data stream with consistent time points is generated, which includes standardized and time-aligned basic patient information and reports, real-time monitoring signals of surgical equipment, and medical imaging data; The standardized data stream is directly transmitted to the feature processing module and the risk prediction and early warning module as the initial output.

[0023] M2, feature processing module, receives the standardized data stream, extracts key features, performs fusion processing based on a unified time axis, generates a fusion feature vector, and transmits it to the risk prediction and warning module; Figure 2 .

[0024] The key feature extraction process is as follows: According to the standardized data stream, identify the patient basic information and the text content in the report, and set the text content as a string set ,in, For the Article text content, Match the string set with relevant medical terms to determine whether the string contains medical terms, and define medical terms as a set ,in, For the Medical terms, The criteria for determination are: ; like Include ,but , otherwise, ; After successful matching, the corresponding medical terms are extracted by associated numerical values. The formula is as follows: ; in, For the The associated value of the item; For example, The text content is = The patient's systolic blood pressure rises to 160 mmHg, corresponding to the The medical term is = systolic blood pressure increases, at this time, the Associated values ; Finally, structured data is generated based on the medical terms and associated values , ; According to the standardized data stream, the real-time monitoring signal of the surgical equipment is set to ,in, For the Real-time monitoring signals of surgical equipment, The total number of sampling points in the real-time monitoring signal of the surgical equipment, The total number of sampling points is divided into time windows according to fixed time. The formula is: ; in, is the duration of the fixed time split window, is the sampling frequency, is the number of time windows; After obtaining the number of time windows, the signal window calculation is performed, and the formula is: ; in, is the number of points contained in a single time window, Calculated, For the Real-time monitoring signals of surgical equipment, For the signal windows; Calculate the range, rate of change and window mean for each time window. The range value is used to reflect the fluctuation amplitude of the real-time monitoring signal of the surgical device within the time window, and the calculation formula is: ; in, For the extreme values; The change rate is used to reflect the average change of adjacent sampling points within the time window, and its calculation formula is: ; in, is the number of points contained in a single time window, For the The signal window sampling points, For the rate of change; The window mean value feeds back the stable state of the real-time monitoring signal of the surgical device, and its calculation formula is: ; in, is the number of points contained in a single time window, is the window mean; Finally, dynamic data is generated based on the range, rate of change and fluctuation range. , ; According to the standardized data stream, the medical imaging data is set as a two-dimensional matrix , identify the organ contour line in the medical image data, and calculate the number of pixels within the organ contour line, the formula is: ; in, is the image width, is the image height, is the pixel coordinate, is the number of pixels; The physical area and average gray value of the contour line are calculated using the number of pixels. The formula for calculating the physical area is: ; in, is the width of a single pixel, is the height of a single pixel, is the number of pixels, is the physical area; The average gray value calculation formula is: ; in, is the pixel coordinate, For coordinates The gray value at is the area within the organ outline, is the average gray value; Finally, the image data is obtained by combining the physical area and the average grayscale. , ; The unified timeline is defined as: ; in, For the Time nodes, is the total number of time nodes, To unify the timeline collection; The fused feature vector is expressed as: ; in, For the Structured data of time nodes, For the Dynamic data of time nodes, For the Image data of time nodes, For the The fusion feature vector of each time node; Finally, the output fusion feature vector .

[0025] It should be noted that if the time point The patient basic information report or surgical equipment real-time monitoring signal or medical imaging data is used at the time point The most recent corresponding data is used to fill in the gaps.

[0026] The fused feature vector is used to represent the dynamic changes in the patient's state throughout the entire cardiothoracic surgery process and is transmitted to the risk prediction and early warning module as an input basis for calculating the risk probability value.

[0027] M3, risk prediction and warning module, receives the fused feature vector and the patient's physiological indicators, applies the pre-trained prediction model, calculates the patient's risk probability value of cardiothoracic surgery complications, and integrates it into a risk probability table; compares it with the risk threshold table to make a judgment and warning; see Figure 3 .

[0028] The basic architecture of the prediction model is: The input of the prediction model is the patient's physiological indicators; Before the patient's physiological indicators are input into the prediction model, they need to be time-aligned with the fusion feature vector to generate .

[0029] The prediction model adopts a dual-channel neural network structure, accesses the fusion feature vector, and independently calculates the corresponding risk probability value of each cardiothoracic surgery complication to ensure that the output value falls within the risk probability value range of 0 to 1; The dual-channel neural network structure has three layers: The first layer serves as the input layer, receiving the time points The patient's physiological indicators , the expression is ,in, For the Physiological indicators of patients; The second layer is the diversion layer, which includes the left channel and the right channel. The left channel uses a self-attention model to calculate the correlation matrix between all the patient's physiological indicators. The correlation matrix reflects the basic rules, and the elements of the correlation matrix serve as the influence probability between the physiological indicators. The formula is: ; in, 、 is the trainable weight matrix; is the feature dimension, which is 64 in this embodiment; is the correlation matrix; The first Rank Elements of a column.

[0030] Indicates the physiological indicators right Probability of impact; For example, Indicates blood pressure value, Indicates urine volume, then when When the blood pressure drops, there is an 87% chance that the urine volume will decrease simultaneously. The right channel adopts a routing mechanism to extract the strongly correlated physiological indicators corresponding to specific complications from the patient's physiological indicators and directly allocate them to the corresponding complication clinics on the third floor. Through a predefined medical routing matrix Complete the allocation of the strongly correlated physiological indicators, Is a binary routing flag, representing the medical routing matrix No. Rank Elements of the column; The allocation decision is made according to the following routing rules: ; when When The patient's physiological indicators are assigned to Complications clinic; For example, blood pressure and urine volume data are assigned to the renal injury clinic, and blood oxygen and respiratory rate data are assigned to the respiratory clinic; The third layer is the diagnosis layer. Two independent complication clinics work simultaneously. Each clinic only studies designated complications and independently calculates the risk probability value of each different complication. For each different complication, first calculate the weighted outcome value of the complication, The time point The formula for calculating the weighted outcome value of each complication is: ; in, For the The weighted outcome value of each complication, The first List, Curing parameters; The weighted result value is input into the sigmoid function to obtain the risk probability value of the corresponding complication: ; in, For the The risk probability of complications, is a natural constant.

[0031] The training process of the prediction model includes the dataset construction phase, the learning phase and the verification phase. The training process is as follows: The process of the dataset construction phase is as follows: Real historical cardiothoracic surgery data from the past five years were collected, including preoperative examination records, intraoperative monitoring records, postoperative medication records, and postoperative complication records within 72 hours. Basic patient information and reports, real-time monitoring signals of surgical equipment, medical imaging data, and patient physiological indicator data were extracted from them.

[0032] Before extraction, invalid data caused by disconnection of surgical equipment were cleared, and the missing data were filled with the average value of adjacent time points to ensure that the data completeness rate was greater than 98%. For example, when blood pressure data is temporarily missing, the average of the blood pressure data before and after 5 minutes is used to fill the gap. Cardiothoracic surgeons mark the time points of occurrence of all complications, and combine the historical cardiothoracic surgery data to establish a data set of at least 10,000 samples; the sample labels of the training set are the risk probability values of complications, which are frequency values obtained by statistically analyzing the real historical cardiothoracic surgery data.

[0033] The dataset is divided into training set and test set in a ratio of 8:2.

[0034] In the learning phase, the prediction model is trained using the training set and predicted using the test set. When the prediction model has a test accuracy rate of ≥85% for three consecutive days, it enters the clinical validation phase. When the test accuracy does not meet the standard, the learning phase is repeated to adjust the prediction model parameters.

[0035] The clinical verification stage includes static testing, dynamic testing and manual testing. The static test is a blind test using data from 500 new cardiothoracic surgeries. When the complication prediction accuracy is greater than 85% and the static test is passed, the dynamic test will begin. The dynamic test is connected to an external cardiothoracic surgery simulation system to verify the predictive model's ability to respond to emergencies. If an early warning is triggered within 5 minutes, manual testing will begin. The manual test involves inviting cardiothoracic surgeons to compare predictions for 100 high-risk cases. The manual test is passed when the prediction model's prediction results are more than 90% consistent with the cardiothoracic surgeon's clinical judgment.

[0036] According to the risk probability values of the corresponding complications, a risk probability table with time nodes is integrated, as shown in the following table:

[0037] The risk probability table serves as a real-time updated risk probability value to capture the disease progression; According to the risk probability table and the preset risk threshold table, the risk probability values of all complications are determined in turn. If a specific complication is higher than the corresponding risk threshold, an early warning is triggered.

[0038] An example of the risk threshold table is as follows:

[0039] According to the above risk probability table and risk threshold table, if the risk probability value of acute kidney injury is higher than the corresponding risk threshold, the corresponding early warning plan will be triggered; The warning scheme includes visual warnings and acoustic prompts, The visual warning will pop up a corresponding red warning box on the surgical equipment according to the different types of complications. For example, [Warning]! High risk of acute kidney injury (85%) exceeds the threshold (60%); The acoustic prompt will emit a 3-second prompt tone every 2 minutes.

[0040] An artificial intelligence-based cardiothoracic surgery risk warning method, the method comprising the following steps: S1. Data input: collecting basic patient information and reports, real-time monitoring signals from surgical equipment, medical imaging data, and patient physiological indicator data throughout the entire cardiothoracic surgery process; standardizing the basic patient information and reports, real-time monitoring signals from surgical equipment, and medical imaging data to generate a standardized data stream; S2, feature processing, receiving the standardized data stream, extracting key features, performing fusion processing based on a unified time axis, and generating a fused feature vector; S3, risk prediction and early warning, receives the fused feature vector and the patient's physiological indicators, applies the pre-trained prediction model, calculates the risk probability values of the patient developing various cardiothoracic surgery complications, and integrates them into a risk probability table; performs risk judgment based on the risk threshold table.

[0041] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An artificial intelligence-based cardiothoracic surgery risk warning system, characterized in that: include: M1, data input module, collects original data including basic patient information and reports, real-time monitoring signals of surgical equipment, medical imaging data, and patient physiological index data throughout the entire cardiothoracic surgery process; standardizes the basic patient information and reports, real-time monitoring signals of surgical equipment, and medical imaging data to generate a standardized data stream, which is then transmitted to the feature processing module and the risk prediction and early warning module; M2, a feature processing module, receives the standardized data stream, extracts key features, performs fusion processing based on a unified time axis, generates a fusion feature vector, and transmits it to the risk prediction and warning module; M3, a risk prediction and early warning module, receives the fused feature vector and the patient's physiological indicators, applies a pre-trained prediction model, calculates the patient's risk probability value of cardiothoracic surgery complications, and integrates the values into a risk probability table; Compare the risk threshold table and make judgments and early warnings.

2. The system according to claim 1, wherein: The standardization process includes unifying the format of the original data and correcting the acquisition time point; The patient's basic information and report and the medical imaging data are pre-stored before the cardiothoracic surgery, and the surgical equipment's real-time monitoring signals and the patient's physiological indicators are updated in real time during the cardiothoracic surgery; After processing, a standardized data stream with consistent time points is generated, which contains standardized and time-aligned patient basic information and reports, real-time monitoring signals of surgical equipment, and medical imaging data.

3. The system according to claim 2, characterized in that The key feature extraction process includes structured data generation, dynamic data generation and image data generation. The structured data generation process is as follows: According to the standardized data stream, identify the patient basic information and the text content in the report, and set the text content as a string set ,in, For the Article text content, Match the string set with relevant medical terms to determine whether the string contains medical terms, and define medical terms as a set ,in, For the Medical terms, The criteria for determination are: ； like Include ,but , otherwise, ; After successful matching, the associated numerical value of the corresponding medical term is extracted. The formula is as follows: ； in, For the The associated value of the item; Finally, structured data is generated based on the medical terms and associated values , .

4. The system according to claim 3, characterized in that The dynamic data generation process is as follows: According to the standardized data stream, the real-time monitoring signal of the surgical equipment is set to ,in, Real-time monitoring signal for the i-th surgical device, The total number of sampling points in the real-time monitoring signal of the surgical equipment, The total number of sampling points is divided into time windows according to fixed time. The formula is: ； in, is the duration of the fixed time split window, is the sampling frequency, is the number of time windows; After obtaining the number of time windows, the signal window calculation is performed, and the formula is: ； in, is the number of points contained in a single time window, Calculated; Real-time monitoring signal for the i-th surgical device; is the i-th signal window; Calculate the range, rate of change and window mean based on each time window. The calculation formula of the range value is: ； in, For the extreme values; The calculation formula of the change rate is: ； in, is the qth sampling point of the i-th signal window, For the rate of change; The calculation formula of the window mean is: ； in, is the window mean; Finally, dynamic data is generated based on the range, rate of change and fluctuation range. , .

5. The system according to claim 3, wherein: The image data generation process is as follows: According to the standardized data stream, the medical imaging data is set as a two-dimensional matrix , identify the organ contour line in the medical image data, and calculate the number of pixels within the organ contour line, the formula is: ； in, is the image width, is the image height, is the pixel coordinate, is the number of pixels; The physical area and average gray value of the contour line are calculated using the number of pixels. The formula for calculating the physical area is: ； in, is the width of a single pixel, is the height of a single pixel, is the number of pixels, is the physical area; The average gray value calculation formula is: ； in, is the pixel coordinate, For coordinates The gray value at is the area within the organ outline, is the average gray value; Finally, the image data is obtained by combining the physical area and the average grayscale. , .

6. The system according to claim 5, characterized in that The unified timeline is defined as: ； in, For the Time nodes, is the total number of time nodes, To unify the timeline collection; The fused feature vector is expressed as: ； in, For the Structured data of time nodes, For the Dynamic data of time nodes, For the Image data of time nodes, For the The fusion feature vector of each time node; Finally, the output fusion feature vector .

7. The system according to claim 6, characterized in that The basic architecture of the prediction model is: The input of the prediction model is the patient's physiological indicators; Before the patient's physiological indicators are input into the prediction model, they need to be time-aligned with the fusion feature vector to generate ; The prediction model adopts a dual-channel neural network structure, accesses the fusion feature vector, and independently calculates the corresponding risk probability value of each cardiothoracic surgery complication.

8. The system according to claim 7, characterized in that The dual-channel neural network structure has three layers: The first layer serves as the input layer, receiving the patient's physiological indicators at time point p. , the expression is ,in, is the physiological index of type c patients; The second layer is the diversion layer, which includes the left channel and the right channel. The left channel uses a self-attention model to calculate the correlation matrix between all the patient's physiological indicators. The elements of the correlation matrix are used as the influence probability between the physiological indicators. The formula is: ； in, 、 is the trainable weight matrix; is the feature dimension; G is the correlation matrix; The first Rank Elements of the column; Indicates the physiological indicators right Probability of impact; The right channel adopts a routing mechanism to extract the strongly correlated physiological indicators corresponding to specific complications from the patient's physiological indicators and directly allocate them to the corresponding complication clinics on the third floor. The allocation of the strongly correlated physiological indicators is completed through the predefined medical routing matrix R. is a binary routing flag, indicating the cth row and the Elements of the column; The allocation decision is made according to the following routing rules: ； when When The patient's physiological indicators are assigned to Complications clinic; The third layer is the diagnosis layer, where Z independent complication clinics work simultaneously. Each clinic only studies designated complications and independently calculates the risk probability value of each different complication. For each different complication, first calculate the weighted outcome value of the complication, and for the first The formula for calculating the weighted outcome value of each complication is: ； in, For the The weighted outcome value of each complication, The first List, is the curing parameter; The weighted result value is input into the sigmoid function to obtain the risk probability value of the corresponding complication: ； in, For the The risk probability of complications, is a natural constant.

9. The system according to claim 8, characterized in that According to the risk probability values of the corresponding complications, a risk probability table with time nodes is integrated. According to the risk probability table and the preset risk threshold table, the risk probability values of all complications are determined in turn. If a specific complication is higher than the corresponding risk threshold, an early warning is triggered.

10. An artificial intelligence-based cardiothoracic surgery risk warning method, characterized in that: The following steps are involved: S1. Data input: collecting raw data including basic patient information and reports, real-time monitoring signals from surgical equipment, medical imaging data, and patient physiological index data throughout the entire cardiothoracic surgery process; standardizing the basic patient information and reports, real-time monitoring signals from surgical equipment, and medical imaging data to generate a standardized data stream; S2, a feature processing module, receives the standardized data stream, extracts key features, performs fusion processing based on a unified time axis, and generates a fused feature vector; S3, a risk prediction and early warning module, which receives the fused feature vector and the patient's physiological indicators, applies a pre-trained prediction model, calculates the patient's risk probability value of cardiothoracic surgery complications, and integrates the values into a risk probability table; Compare the risk threshold table and make judgments and early warnings.

Images