Cardiac surgery risk assessment system based on artificial intelligence

Abstract

The invention relates to the field of medical treatment, and discloses a cardiac surgery risk assessment system based on artificial intelligence. The system fuses electrocardiogram, image data and physiological data of wearable equipment, and data quality is ensured through a data preprocessing module; carrying out data dimensionality reduction and nonlinear mapping by using manifold learning and the method, and capturing a relationship between complex physiological signals; the sensor weight is optimized by adopting the maximum entropy principle, and the accuracy of the evaluation result is improved; performing space-time alignment of the multi-modal data based on the Lie group and Lie algebra theories; modeling the fused data by using a machine learning algorithm to generate a risk assessment result; the real-time feedback module dynamically adjusts the evaluation model and system parameters, and the output module displays an evaluation result through a visual interface to assist doctors in accurate decision making. The system improves the accuracy and real-time performance of assessment, and provides scientific and comprehensive risk assessment support for cardiac surgery.

Description

Technical Field

[0001] The present invention relates to the technical field of medical field, and in particular to a cardiac surgery risk assessment system based on artificial intelligence. Background Art

[0002] Cardiac surgery is a key means of treating cardiovascular diseases, but its high risk and complexity require accurate assessment of surgical risks. Traditional assessment relies on physician experience and basic physiological parameters, but has limitations, especially when dealing with complex physiological data.

[0003] Multimodal data fusion technology is becoming increasingly important in the medical field, but technical challenges include data type and format differences, noise processing, and spatiotemporal alignment. Current methods mostly rely on simple weighted averages and fail to fully utilize the nonlinear relationship between data.

[0004] As the amount and dimension of data increase, traditional linear modeling methods cannot meet the needs of high-dimensional data analysis. Although nonlinear modeling technology has made progress, it has high computational complexity and lacks optimization for risk assessment of cardiac surgery.

[0005] The physiological state of patients during cardiac surgery varies in time and space, and real-time synchronization and processing of multi-sensor data is a technical challenge. Existing technologies are difficult to meet the real-time and accuracy requirements.

[0006] The present invention proposes an artificial intelligence-based cardiac surgery risk assessment system, aiming to address the deficiencies of the prior art. Summary of the invention

[0007] In view of the shortcomings of the existing technology in cardiac surgery risk assessment, the present invention provides a cardiac surgery risk assessment system based on artificial intelligence. The system can comprehensively and accurately assess the risk of cardiac surgery by integrating multimodal data acquisition, data preprocessing, nonlinear modeling, data fusion, spatiotemporal synchronization, risk assessment and real-time feedback modules.

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: a cardiac surgery risk assessment system based on artificial intelligence, comprising: A data acquisition module, used to collect data from multiple sensors, including electrocardiogram, imaging data, and wearable device physiological data; Data preprocessing module, used to denoise, standardize and time-series align the collected data; A nonlinear modeling module, which is used to reduce the dimensionality of high-dimensional data through manifold learning and to perform nonlinear mapping on the data through this method; Data fusion module, used to fuse multimodal data and generate unified feature representation; The spatiotemporal synchronization module is used to align multimodal data in space and time through the theory of Lie groups and Lie algebras; A risk assessment module, used to generate cardiac surgery risk assessment results based on fused multimodal data; A real-time feedback module is used to adjust the evaluation model and system parameters according to real-time data during surgery; Output module for displaying cardiac surgery risk assessment results.

[0009] Preferably, the data preprocessing module further comprises a denoising unit and a timing alignment unit, and the timing alignment unit is used to synchronize data from different sensors according to timestamps.

[0010] Preferably, the nonlinear modeling module performs data dimensionality reduction by a manifold learning algorithm and performs data nonlinear mapping by a Gaussian kernel function.

[0011] Preferably, the data fusion module adopts the maximum entropy principle to perform information fusion on multimodal data and optimizes the weight of each sensor to improve the accuracy of the evaluation result.

[0012] Preferably, the space-time synchronization module performs spatial coordinate transformation on multimodal data through Lie group and Lie algebra theory to solve the space-time deviation problem between data.

[0013] Preferably, the risk assessment module uses a machine learning algorithm to model the fused data to generate a risk level for cardiac surgery.

[0014] Preferably, the real-time feedback module is based on Lyapunov stability theory and dynamically adjusts control parameters of the evaluation model according to real-time data to ensure that the system remains stable during the operation.

[0015] Preferably, the output module displays the cardiac surgery risk assessment results, including risk level, relevant physiological parameters and prediction suggestions, through a visual interface.

[0016] Preferably, a cardiac surgery risk assessment method based on artificial intelligence is characterized by comprising the following steps: Collect multimodal data from multiple sensors; De-noise, standardize and time-series align the collected data; Use manifold learning algorithm to reduce the dimensionality of high-dimensional data and use this method to perform nonlinear mapping on the data; Perform information fusion on the processed multimodal data; Use Lie group and Lie algebra theory to align data in space and time; Generate cardiac surgery risk assessment results based on fused data; Adjust evaluation models and system parameters based on real-time data; Output surgical risk assessment results.

[0017] The present invention provides a cardiac surgery risk assessment system based on artificial intelligence, which has the following beneficial effects: 1. By integrating multiple sensor data (such as electrocardiogram, blood pressure, body temperature, blood oxygen, etc.) and performing efficient data processing and fusion, the present invention can comprehensively reflect the patient's multi-faceted physiological state. This multimodal data fusion technology overcomes the limitations of a single data source, thereby significantly improving the accuracy and reliability of cardiac surgery risk assessment. By reducing the dimension and nonlinear mapping of data through manifold learning and this method, the complex features in the data can be better captured, avoiding the over-simplification problem that may occur in traditional methods.

[0018] 2. The present invention can quickly respond to the physiological changes of patients during surgery by dynamically adjusting the evaluation model and system parameters based on real-time data through a real-time feedback module. This feature can provide more personalized evaluation support for medical staff, ensuring that the evaluation results are highly matched with the patient's current state, so as to make more accurate and timely decisions. This real-time dynamic adjustment capability is particularly important during surgery and can effectively reduce the risks caused by fluctuations in physiological parameters.

[0019] 3. By using Lie group and Lie algebra theory to align multimodal data in time and space, the present invention effectively solves the problem of deviation in time and space between different sensor data, and ensures the time and space consistency of the data. This technical solution can provide medical staff with accurate surgical risk assessment results, which helps to timely discover potential risk factors during surgery, so as to take preventive measures in advance, reduce misjudgment caused by inconsistent data, and thus improve the safety of surgery.

[0020] 4. The present invention can automatically provide decision support for surgery through an intelligent risk assessment model, reducing the workload of medical staff during surgery. Through the intuitive display of risk assessment results, medical staff can quickly obtain key patient physiological data and assessment results, thereby saving time and improving decision-making efficiency. This is of great significance for complex surgical environments, especially cardiac surgery that requires rapid response. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Flowchart of the artificial intelligence-based cardiac surgery risk assessment method. DETAILED DESCRIPTION

[0022] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Please see attached Figure 1 An embodiment of the present invention provides a cardiac surgery risk assessment system based on artificial intelligence. The following describes in detail each module of the system of the present invention.

[0024] Data acquisition module, used to collect data from multiple sensors, including electrocardiogram, imaging data and wearable device physiological data This embodiment relates to a data acquisition module in an artificial intelligence-based cardiac surgery risk assessment system. The module is used to collect data from multiple sensors, including electrocardiogram (ECG), imaging data, and physiological data from wearable devices. The data acquisition module interacts with multiple sensors in real time to ensure accurate and comprehensive patient physiological data, providing necessary input for subsequent multimodal data processing and cardiac surgery risk assessment.

[0025] The design goal of the data acquisition module is to improve the real-time, accuracy and comprehensiveness of the data and ensure the time synchronization of each data source to support accurate risk assessment. Specifically, the data acquisition module includes multiple sensor interfaces, each of which is responsible for exchanging data with different types of sensors. The implementation method and technical details of the data acquisition module will be described in detail below according to the technical solution of the present invention.

[0026] Sensor data collection In general, the data acquisition module can simultaneously receive data from multiple sensors, including but not limited to electrocardiogram (ECG) sensors, imaging data acquisition equipment (such as cardiac CT or MRI scanners), and wearable devices (such as smart bracelets, blood pressure monitors, pulse oximeters, etc.). Different sensors provide different types of physiological data, which respectively reflect the electrophysiological information of the patient's heart activity, morphological changes of the heart, and real-time data of other physiological parameters.

[0027] As an option, the data acquisition module can support the electrocardiogram sensor to collect heart electrical activity data in real time. The electrocardiogram signal reflects the electrophysiological state of the heart, heart rate, heart rhythm and other indicators. This information is crucial for risk assessment of cardiac surgery.

[0028] In one possible implementation, an image data acquisition device (such as cardiac CT or MRI) can provide three-dimensional image data about the patient's heart structure, blood flow, coronary artery status, etc. After preprocessing, these image data can provide detailed structural information of the heart for fusion analysis with other sensor data.

[0029] Timing synchronization and data alignment In some embodiments, the data acquisition module also includes a timing synchronization function to ensure the consistency of data from different sensors in the time dimension. Since different sensors may have different data acquisition frequencies and timestamps, timing alignment is required to ensure that the physiological data collected by each sensor can be compared and fused within the same time window.

[0030] Specifically, the data acquisition module synchronizes the data of each sensor according to the timestamp. Assume that the measurement results from different sensors are ,in represents the sampling time, is the sensor number, Indicates Sensors at time To solve the timing difference problem, the dynamic time warping (DTW) algorithm or interpolation method can be used to align the data.

[0031] The data set after time series alignment can be expressed as:

[0032] in, represents the number of sensors, are the start and end time of data collection respectively. The goal of timing alignment is to align the data points of all sensors on a unified time scale, thereby ensuring the synchronization of the data and providing a reliable basis for subsequent data processing and analysis.

[0033] Real-time acquisition of multimodal data The data acquisition module not only needs to ensure the timing consistency of the data, but also needs to meet the real-time requirements to support real-time risk assessment. In some embodiments, the data acquisition module can process the continuous data stream from each sensor in real time and transmit it to the next processing link of the system. To this end, the data acquisition module is usually equipped with a high-speed data transmission interface, such as USB, Bluetooth, Wi-Fi and other communication protocols to ensure efficient and real-time data collection and transmission.

[0034] In particular, during the data collection process of wearable devices, the sensor module connects to the system through wireless communication technology (such as Bluetooth or Wi-Fi) to transmit physiological parameters (such as heart rate, blood oxygen saturation, etc.) in real time. For example, when using a smart bracelet to monitor the patient's heart rate data, the data can be transmitted to the data collection module through the Bluetooth interface. The system can dynamically evaluate the patient's heart health status through real-time analysis of these data.

[0035] Data denoising and preprocessing In order to improve data quality, the data acquisition module also includes a denoising function. Electrocardiogram (ECG) signals often contain interference components such as myoelectric noise and power frequency noise, which may affect subsequent data analysis and risk assessment results. Therefore, the data acquisition module usually uses filtering algorithms (such as Kalman filtering, low-pass filtering, wavelet transform, etc.) to denoise the signal.

[0036] Specifically, for the denoising of electrocardiogram signals, the following wavelet transform method can be used:

[0037] in, is the original ECG signal, is the matrix operation of wavelet transform, The signal after denoising. Wavelet transform can effectively remove high-frequency noise while maintaining the main features of the signal to ensure the clarity and accuracy of the signal.

[0038] Hardware Implementation of Data Acquisition Module In this embodiment, the data acquisition module can be composed of a set of integrated circuits (ICs) or multifunctional sensor units. In order to achieve efficient data acquisition and processing, the module uses a high-performance processor, which can perform preliminary processing on the data while collecting data, such as denoising, standardization, and time alignment.

[0039] The data acquisition module can also use the standardization module to uniformly format the data from different sensors. Specifically, the data output by different sensors may have different units and ranges, so these data need to be standardized so that the dimensions of different data sources are consistent. The standardization process can be expressed as:

[0040] in, is the original data, and represent the mean and standard deviation of the data, respectively. The standardized data is convenient for subsequent machine learning and data fusion processing, ensuring the uniformity and accuracy of model input.

[0041] Collaborative work of data acquisition modules In order to achieve accurate assessment of cardiac surgery risks, the data acquisition module not only needs to process data from a single sensor, but also needs to support the collaborative work of multiple sensors. In order to ensure the effective fusion of data from each sensor, the data acquisition module works closely with the data preprocessing module, the nonlinear modeling module, and the risk assessment module to transfer data through interfaces and protocols. During the data acquisition process, the collected data will be transmitted in real time to the next processing link, such as the data preprocessing module, for further processing and analysis.

[0042] The data acquisition module in this embodiment provides rich and accurate input data by efficiently collecting and transmitting electrocardiograms, imaging data, and physiological parameters of wearable devices in real time. The consistency and high quality of multimodal data are ensured through technical means such as time synchronization, denoising, and standardization. These data provide a reliable basis for subsequent risk assessment and can support the system's accurate assessment and real-time monitoring of patients' cardiac surgery risks.

[0043] Data preprocessing module, used to denoise, standardize and time-series align the collected data The present embodiment relates to a data preprocessing module in a cardiac surgery risk assessment system based on artificial intelligence. The main task of the data preprocessing module is to preprocess the raw data collected from various sensors to ensure the accuracy and reliability of subsequent data analysis and risk assessment. As mentioned above, the data acquisition module collects data from multimodal sensors such as electrocardiograms (ECGs), imaging devices, and wearable devices. Before these data enter the analysis stage, there are usually problems such as noise interference, timestamp misalignment, and uneven data distribution. Therefore, a series of processing is required through the data preprocessing module.

[0044] The core functions of the data preprocessing module include denoising, standardization, and time series alignment. Through these processing steps, the interference information in the data can be effectively removed, the differences between data sources can be eliminated, and it can be converted into high-quality data in a unified format for use by subsequent nonlinear modeling, data fusion, and risk assessment modules. The technical implementation of the data preprocessing module will be described in detail below.

[0045] Denoising Generally speaking, there may be various noise sources in the raw data, especially noise from electrocardiogram (ECG) signals. Common noises include baseline drift, power frequency interference, and myoelectric noise. These noises may cause errors in the analysis results, so denoising is required.

[0046] In some embodiments, the denoising process is implemented using a wavelet transform method. Wavelet transform can effectively separate the low-frequency part and high-frequency noise of the signal, thereby removing unnecessary interference. Assume that the original signal is , and the decomposition result obtained by wavelet transform is:

[0047] in, is the wavelet coefficient, is the corresponding mother wavelet function. By adjusting the wavelet coefficients , which can suppress high-frequency noise components while retaining low-frequency signals. Specifically, in the denoising stage, the noise coefficient can be reduced by a soft threshold or hard threshold method, thereby achieving signal denoising.

[0048] Another possible denoising method is Kalman filtering. Kalman filtering is a recursive estimation algorithm based on a state space model, which is suitable for noise removal in dynamic systems. In the ECG signal, the ECG can be regarded as a linear dynamic system. The signal is estimated through the Kalman filter to gradually reduce the noise interference.

[0049] Standardization In one possible implementation, data standardization is an important part of the data preprocessing module, especially unified standardization before multimodal data fusion. Data collected by different sensors often have different dimensions and ranges. For example, the unit of electrocardiogram signal is millivolt (mV), while the unit of blood oxygen saturation is percentage (%). If standardization is not performed, the dimension difference of the data will lead to deviations in subsequent analysis.

[0050] A common method of normalization is z-score normalization, which subtracts the mean from each data point and divides it by its standard deviation. The formula is:

[0051] in, is the original data, is the mean of the data series, is the standard deviation, is the standardized data. After standardization, the data will be converted into a standard normal distribution with a mean of zero and a standard deviation of one, thereby eliminating the dimensional differences in data from different sensors, making the data more uniform and convenient for subsequent analysis.

[0052] Timing Alignment and Synchronization In the application of multimodal data, various sensors may have different sampling frequencies and timestamps, which will lead to inconsistency in the time dimension of the data. In order to ensure the synchronization of data from different sensors, the data preprocessing module needs to perform time alignment processing to facilitate subsequent data analysis and fusion.

[0053] In this embodiment, the timing alignment adopts an interpolation method based on timestamps. In practical applications, the data collected by each sensor usually has a timestamp. , represents the sampling time of the data. In order to align the data from different sensors, we first need to perform time interpolation based on the timestamp. For example, suppose the data from sensor The data is , and from the sensor The data is The timestamps of the two are and , the data values ​​at the common time points of two data series can be calculated by interpolation methods. Common interpolation methods include linear interpolation and spline interpolation.

[0054] Specifically, assuming that at a certain point in time Interpolation is performed, and the interpolation result is:

[0055] in, and are adjacent timestamps, and is the corresponding sensor data value. In this way, the data of all sensors can be aligned on the same time axis, providing consistent data input for subsequent data analysis and fusion.

[0056] Joint processing of data denoising and standardization In some embodiments, denoising and standardization can be performed jointly, especially when the amount of data is large and the noise is high, the joint processing can improve the data quality. For example, the standardization process can be performed after denoising to avoid the noise in the denoising process affecting the standardization effect of the data. Specifically, the original data is first denoised to remove high-frequency noise, and then standardized, so that the final data set is clearer and has a unified scale, which is convenient for subsequent data analysis and modeling.

[0057] The data preprocessing module in this embodiment ensures the consistency of quality and format of multimodal data collected from different sensors through multiple steps such as denoising, standardization, and time alignment. These processing steps can effectively remove interference, eliminate dimensional differences in data from different sensors, and unify the data on the same timeline, ensuring the accuracy and reliability of subsequent multimodal data fusion and cardiac surgery risk assessment. Through these processes, the data preprocessing module provides high-quality input data for the subsequent risk assessment module, laying the foundation for the precise operation of the system.

[0058] Nonlinear modeling module, used to reduce the dimensionality of high-dimensional data through manifold learning, and to perform nonlinear mapping on the data through this method The present embodiment relates to a nonlinear modeling module in a cardiac surgery risk assessment system based on artificial intelligence. The main task of this module is to perform nonlinear modeling on multimodal data collected from various sensors in order to better capture the complex and nonlinear potential relationships in the data and generate low-dimensional feature representations suitable for subsequent risk assessment models. The data involved in the cardiac surgery risk assessment system are usually high-dimensional, complex and nonlinear, and these data include electrocardiogram signals, imaging data, and physiological data collected by wearable devices. In this context, the nonlinear modeling module performs data dimensionality reduction through a manifold learning algorithm, and uses a Gaussian kernel (RBF kernel) function to perform nonlinear mapping on the data, thereby reducing the data dimension while maintaining the essential information of the data, making subsequent model training and reasoning more efficient.

[0059] Generally speaking, high-dimensional data contains redundant information and may contain noise, which will affect the performance of the model. Therefore, the nonlinear modeling module uses manifold learning algorithms combined with Gaussian kernel functions to process data. These algorithms can not only effectively reduce the dimension, but also retain the nonlinear structure of the data, improving the accuracy of subsequent risk assessment. The technical implementation of the nonlinear modeling module will be described in detail below.

[0060] Manifold Learning Dimensionality Reduction In some embodiments, manifold learning is selected for data dimensionality reduction. Manifold learning is an unsupervised learning method that aims to map high-dimensional data to a low-dimensional space while preserving the local geometric structure of the data as much as possible. Since data is often distributed on a low-dimensional manifold in practical applications, manifold learning can more effectively extract the essential features of the data.

[0061] Specifically, the purpose of manifold learning is to use a mapping function (in The original high-dimensional data Mapping to low-dimensional space .in, It is The original high-dimensional features of the data points, is the corresponding low-dimensional feature.

[0062] Commonly used manifold learning methods include Local Linear Embedding (LLE) and t-SNE (t-Distributed Stochastic Neighbor Embedding). These methods can preserve the local structure of the original data as much as possible in a low-dimensional space by constructing adjacency relationships between data points.

[0063] For example, Locally Linear Embedding (LLE) performs dimensionality reduction by following the steps below: Construct the adjacency matrix: First, for each data point , find its The local linear structure of each data point is then represented by weighted average:

[0064] in, Indicates The neighbor set of a data point, is the weight, indicating that the point Point to neighbors degree of dependence.

[0065] Optimize the objective function: By minimizing the reconstruction error, we get the embedded representation in the low-dimensional space:

[0066] in, It is a data point The goal of the representation in low-dimensional space is to make the data points in the low-dimensional space maintain similar adjacency relationships as the data points in the high-dimensional space.

[0067] Nonlinear Mapping - Gaussian Method When performing manifold learning dimensionality reduction, the data often contains some complex nonlinear structures, especially in physiological data and electrocardiogram signals. In order to effectively capture these nonlinear relationships, this embodiment uses a Gaussian kernel function (RBF kernel) to perform nonlinear mapping on the data. The Gaussian kernel function is a common method that can map data from a low-dimensional space to a high-dimensional space and enhance the separability of the data.

[0068] The expression of Gaussian kernel function is: in, and is a data point, is the Euclidean distance between them, is the width parameter of the kernel function.

[0069] By using the Gaussian kernel function, the nonlinear modeling module can map the linear relationship between the original data points into a nonlinear relationship in a high-dimensional space, which can capture the complex nonlinear structure in the data. After the nonlinear mapping, the local structure of the data can be more accurately preserved, thus providing richer and more effective information for the subsequent risk assessment model.

[0070] Application of Gaussian kernel in nonlinear modeling In one possible implementation, the nonlinear modeling module first uses the manifold learning algorithm to reduce the dimension of the data and reduce the redundant information of the data. Then, the Gaussian kernel function is used to perform nonlinear mapping on the reduced-dimensional data to ensure that the nonlinear structure in the data is effectively preserved. In this process, the Gaussian kernel function used can enhance the similarity between data points, thereby further improving the representation ability of the data model.

[0071] Joint Optimization for Nonlinear Modeling In some embodiments, manifold learning and Gaussian kernel mapping can be jointly optimized to form a multi-stage nonlinear modeling process. Specifically, the data is first initially reduced in dimension by a manifold learning algorithm, and an embedding in a low-dimensional space is constructed. Then, the Gaussian kernel function is used to perform nonlinear mapping on the reduced-dimensional data, so that the structure of the data in the high-dimensional space is strengthened, thereby better capturing the nonlinear patterns in the data. This process can be achieved through iterative optimization, that is, in each round of iteration, manifold learning dimensionality reduction is first performed, and then Gaussian kernel mapping is performed until the optimization goal is reached.

[0072] Through the above-mentioned nonlinear modeling method, this embodiment can effectively process the complex nonlinear relationships in multimodal data, and perform dimensionality reduction and nonlinear mapping on high-dimensional data. Through the dimensionality reduction step of the manifold learning algorithm, the dimension of the data is reduced and redundant information is removed; at the same time, through the nonlinear mapping of the Gaussian kernel function, the nonlinear structure in the data is retained, providing a more accurate and rich feature representation for the subsequent risk assessment model. This nonlinear modeling method can effectively improve the accuracy of cardiac surgery risk assessment and ensure that the system can process multimodal data from different sensors, thereby achieving accurate assessment of cardiac surgery risks.

[0073] Data fusion module, used to fuse multimodal data and generate unified feature representation This embodiment relates to a data fusion module in an artificial intelligence-based cardiac surgery risk assessment system. The core function of the data fusion module is to effectively fuse multimodal data from different sensors to generate a unified feature representation. Since the data required for cardiac surgery risk assessment comes from various sources, including electrocardiogram (ECG), imaging data (such as CT, MRI), and physiological data collected by wearable devices (such as blood pressure, heart rate, etc.), these data usually have differences in timing, space, signal noise, etc. Therefore, when fusing multimodal data, appropriate algorithms must be used to weight, align and fuse various types of data to improve the accuracy and reliability of risk assessment results.

[0074] In general, the data fusion process involves multiple steps, including data preprocessing, alignment, weighting, and fusion. The heterogeneity of data, its temporal characteristics, and its spatial relationship make this process complicated. To address these challenges, this embodiment uses the Maximum Entropy Principle as the theoretical basis for data fusion to optimize the weight of each sensor and maximize the amount of information in the fusion result through information theory methods. This method can effectively avoid information loss during data fusion and improve the overall performance of the evaluation model.

[0075] Application of Maximum Entropy Principle in Data Fusion In this embodiment, the data fusion module performs weighted fusion of multimodal data based on the maximum entropy principle. The maximum entropy principle is an important concept in information theory. Its core idea is to select results with the greatest uncertainty (i.e., the largest amount of information) without sufficient prior knowledge. For the fusion of multimodal data, the maximum entropy principle requires that the weight of each sensor data be determined by maximizing the entropy value of the system under the constraints of each data source.

[0076] Specifically, there are multiple sensor datasets , where each data set contains feature information obtained from different sensors. Assume that the fused data is , we can define entropy for: in, The fused data result The probability distribution of .

[0077] According to the maximum entropy principle, in order to maximize the amount of information after fusion, we need to select the weights while satisfying certain constraints. , so that the entropy value is maximized. Constraints usually include factors such as the credibility of each sensor data and the quality of data collection. For example, setting the weight coefficient of each data source The following conditions are met: in, Indicates The weight of each sensor data is satisfied because the contribution of each sensor data needs to be weighted proportionally.

[0078] To calculate the weight of each sensor , we can construct an optimization problem to find the optimal Value, maximize entropy value: This optimization process can be solved by the Lagrange multiplier method to obtain the optimal weight of each sensor data.

[0079] Information fusion algorithm In a possible implementation, the data fusion module implements data weighted fusion through the following steps: Data preprocessing: First, the raw data from different sensors are preprocessed, including denoising, time alignment, and standardization. Through these preprocessing steps, different types of sensor data can be aligned in time and possible noise can be eliminated.

[0080] Weight calculation: Then, the weight of each sensor data is calculated using the maximum entropy principle. During this process, the system optimizes the weight based on factors such as the sensor's data quality, reliability, and the degree of correlation between the data. Specifically, the calculation of the weight takes into account the signal strength, accuracy, and correlation of each sensor with other sensor data.

[0081] Fusion algorithm: The fusion process is performed by weighted averaging or other weighted fusion methods. Set each sensor data Weight , then the fused data It can be expressed as: in, is the fused data, which represents the weighted sum of all sensor data. Through this weighted fusion method, the system can eliminate noise and maximize the amount of information while ensuring the characteristics of the data itself.

[0082] Result optimization: In some embodiments, the system can further optimize the fused data, such as using deep learning methods such as convolutional neural networks (CNN) or long short-term memory networks (LSTM) to further extract features and model the fusion results, thereby improving the accuracy of the system.

[0083] Specific application examples In some embodiments, it is assumed that the system obtains data such as heart rate and ECG waveform through an electrocardiogram (ECG) sensor, and obtains physiological parameters such as blood pressure and body temperature through a wearable device. In addition, imaging data such as CT or MRI images are also collected to assess the patient's heart condition. There are significant differences in the timestamps, data frequencies, dimensions, and noise characteristics of these data sources.

[0084] Under the action of the data fusion module, the system first performs time alignment on these different types of data to ensure that the data from various sensors can match in time. Then, the maximum entropy principle is used to calculate the weight of each sensor data, taking into account the quality and reliability of each data source. Finally, the system fuses the data sources through weighted averaging and other methods to generate a unified feature representation. This process ensures that the data from different sensors can complement each other, thereby improving the accuracy of risk assessment for cardiac surgery.

[0085] By using the maximum entropy principle for data fusion, this embodiment can effectively combine multimodal data from different sensors to generate high-quality fusion features. The application of the maximum entropy principle can ensure that important information is not lost during the data fusion process, and in the absence of prior knowledge, the amount of information in the fused data can be maximized by optimizing the weights of the sensors. Through this technical solution, the data fusion module can provide accurate and reliable feature representation, thereby providing support for the subsequent cardiac surgery risk assessment model.

[0086] Spatiotemporal synchronization module for spatiotemporal alignment of multimodal data using Lie group and Lie algebra theory This embodiment relates to a spatiotemporal synchronization module in an artificial intelligence-based cardiac surgery risk assessment system. The main function of the spatiotemporal synchronization module is to perform spatiotemporal alignment of multimodal data from multiple sensors to solve the spatiotemporal deviation problem caused by differences in acquisition time, sensor position, or signal quality between data. In the process of multimodal data fusion, spatiotemporal synchronization is a crucial step, especially when processing data from different types of sensors (such as electrocardiogram sensors, imaging devices, wearable devices, etc.). Ensuring that these data can be accurately aligned in time and space is the basis for ensuring the accuracy and reliability of the assessment results.

[0087] In general, there may be deviations in the timestamps and spatial positions of data collected by different sensors, which will lead to deviations and information loss during data fusion. In order to overcome this problem, this embodiment uses Lie group and Lie algebra theory to synchronize spatiotemporal data to ensure that multimodal data can be accurately aligned in both time and space. This method can not only effectively solve the problem of time misalignment between different sensor data, but also handle the spatial differences in different sensor positions, thereby ensuring that the final fused data can fully and accurately reflect the patient's physiological state.

[0088] Application Principles of Lie Groups and Lie Algebras In this embodiment, the space-time synchronization module uses Lie group and Lie algebra theory to transform the spatial coordinates of multimodal data to solve the space-time deviation problem. Lie group and Lie algebra are mathematical tools for describing space transformation and rotation, and have unique advantages in dealing with space-time alignment problems.

[0089] Lie group is a smooth manifold that represents a space with certain algebraic structure. Lie algebra is a tangent space related to Lie group, which represents the linearization of local transformations in Lie group. In the spatiotemporal synchronization module, Lie group is used to describe the spatial transformation of sensor data, while Lie algebra is used to describe the details of these transformations in local space.

[0090] Specifically, there are multimodal data sources , each data source contains its corresponding spatiotemporal coordinate information. Assume that the spatiotemporal coordinates of each data source are expressed as ,in is the timestamp, In order to align the coordinates of these different data sources, it is first necessary to transform the spatiotemporal coordinates of each data source into a unified reference coordinate system.

[0091] In the spatiotemporal synchronization module, the coordinate transformation process of the data source can be realized through Lie group operations. Assume that the Lie group describes a rotation or translation operation from one coordinate system to another, while Lie algebra It describes the local linear transformation of these operations. The spatiotemporal synchronization module performs spatiotemporal alignment between data sources through the following formula: in, represents the transformation element in the Lie group, is the new coordinate after spatial transformation. In order to ensure that the spatial and temporal coordinates of the data source are completely aligned, the transformed spatial and temporal coordinates An optimization process is performed to minimize the spatial and temporal deviations between all data sources.

[0092] Steps in the spatiotemporal synchronization process In the specific implementation, the implementation steps of the spatiotemporal synchronization module are as follows: Data collection and time synchronization: First, the data collection module collects multimodal data from different sensors. Each sensor will attach timestamp information to identify the specific time of data collection. To ensure time synchronization, the system preliminarily sorts all data sources by timestamp.

[0093] Preliminary estimation of spatial coordinate transformation: After preliminary synchronization, the spatiotemporal synchronization module performs spatial coordinate transformation on the data of different sensors through Lie group theory. Assuming that the spatial coordinate system and acquisition position of each sensor are known, the transformation matrix from one sensor coordinate system to another is calculated through linearization of Lie algebra. .

[0094] Optimization of spatiotemporal deviation: For multimodal data collected by different sensors, there may still be spatiotemporal deviations between the data due to differences in the location and orientation of the device. In order to minimize these deviations, the system uses optimization algorithms (such as least squares or Kalman filtering) to adjust the transformation elements in the Lie group. , so that the spatiotemporal data of all sensors are aligned as accurately as possible.

[0095] Data output after spatiotemporal alignment: After spatiotemporal synchronization, the data of all sensors will be aligned in a unified reference coordinate system to form a set of precisely synchronized spatiotemporal data. These aligned data will be passed as input to the data fusion module for further feature fusion and evaluation.

[0096] Mathematical model for specific implementation Assume that the data point of each sensor ,in is a three-dimensional space coordinate Then, through the transformation of Lie groups and Lie algebras, the sensor data points can be represented After the transformation of the data points for: in, is an element in the Lie group, representing the Spatial transformation to a unified reference coordinate system. The transformed data It will be adjusted during the optimization process to reduce the temporal and spatial deviations.

[0097] During the optimization process, the temporal and spatial deviations It can be calculated by the following formula: Then, the system uses an optimization algorithm to Minimize and obtain the final spatiotemporal alignment data.

[0098] By applying the theory of Lie groups and Lie algebras, the spatiotemporal synchronization module can effectively align the multimodal data from multiple sensors in the cardiac surgery risk assessment system. By combining the spatiotemporal synchronization process with the data fusion module, this embodiment can eliminate the data inconsistency caused by sensor position differences and time deviations, thereby improving the accuracy and reliability of risk assessment. The application of Lie groups and Lie algebras provides an accurate and efficient way to deal with the spatiotemporal deviation problem in multimodal data, ensuring high-quality output of data fusion.

[0099] Risk assessment module, used to generate cardiac surgery risk assessment results based on fused multimodal data This embodiment relates to a risk assessment module in an artificial intelligence-based cardiac surgery risk assessment system. The main function of the risk assessment module is to generate risk assessment results for cardiac surgery by analyzing and processing multimodal data from the data fusion module. The module combines multiple machine learning algorithms, especially models based on deep learning, to conduct in-depth mining of the fused data, thereby effectively assessing the potential risks during surgery.

[0100] Generally speaking, the risk assessment of cardiac surgery involves a comprehensive analysis of the patient's physiological state, medical history, imaging data, and other information. Since these data sources are diverse and complex, how to accurately extract effective features and predict risk levels is a core issue in this field. Therefore, this embodiment introduces a multi-level machine learning model and uses deep learning to model the nonlinear characteristics of multimodal data to provide accurate and reliable risk assessment.

[0101] Technical implementation of risk assessment module In this embodiment, the risk assessment module mainly includes the following functional steps: feature extraction, modeling, training and evaluation. Specifically, the risk assessment module of this embodiment processes the fused data from the data fusion module through a machine learning algorithm and trains it to generate the risk level of cardiac surgery.

[0102] First, the risk assessment module receives the multimodal data transmitted from the data fusion module. The data processed by data fusion not only contains information such as electrocardiogram, imaging data and wearable device physiological data, but also these data have been synchronized in time and space, eliminating the time and space deviations between different sensors. Therefore, these data can fully reflect the patient's physiological status and related risk factors.

[0103] In some embodiments, the risk assessment module performs data modeling through a deep neural network (DNN). A deep neural network can handle complex nonlinear relationships and mine potential patterns in multimodal data. Specifically, the structure of a deep neural network includes multiple hidden layers, each of which performs a nonlinear transformation on the input data through an activation function in order to extract higher-level features. Assume that the data after data preprocessing and fusion is represented as , then the data is input into the deep neural network layer, through the weight matrix and bias After the linear transformation, the activation function Get the output of the next layer: ,in, Indicates The output of the layer, is the output of the previous layer, is the activation function (such as ReLU, Sigmoid, etc.), and are the weight and bias of this layer respectively.

[0104] In the training process of the deep neural network, historical data (including the patient's physiological parameters, electrocardiogram, imaging data, etc.) are used to train and optimize the model. To this end, this embodiment uses the backpropagation algorithm to adjust the network parameters and optimize the loss function. The loss function usually uses cross entropy loss or mean square error loss, and the specific formula is: in, Indicates actual value With the predicted value The loss between is the total number of samples, is the parameter of the neural network. The loss function is minimized by the gradient descent method, so that the model can perform risk assessment most accurately.

[0105] During the training process, regularization methods such as L2 regularization may be used to prevent overfitting. The introduction of regularization terms keeps the sum of squares of weight parameters small, as shown in the formula:

[0106] in, is the regularization coefficient, For the The weight of the layer, Represents the L2 norm. Regularization can effectively suppress the overfitting of the model to the training data, thereby improving the generalization ability of the model.

[0107] Specifically, in one possible implementation, the risk assessment module uses an ensemble learning method to fuse the prediction results of multiple models. Common ensemble learning methods include Random Forest, Gradient Boosting Machine (GBM), etc. By weighted averaging or voting the results of multiple models, the stability and accuracy of the prediction can be improved.

[0108] Generation and assessment of risk levels The risk assessment module in this embodiment processes the fused data based on a deep learning model or an integrated learning model, and finally generates a risk level for cardiac surgery. Risk levels are generally divided into three categories: low risk, medium risk, and high risk. For each patient, the model outputs a probability value, indicating the probability of the patient experiencing postoperative complications or death. Represents the probability value in the risk assessment result. According to the probability value output by the model , the following classification rules can be used: when When the risk is assessed as low, when When the risk is medium, when , it is assessed as high risk.

[0109] This grading method can provide personalized risk assessment based on the patient's different circumstances, helping the medical team make better decisions.

[0110] Model evaluation and validation To ensure the accuracy and reliability of the risk assessment module, this embodiment uses a cross-validation method to evaluate the model. Specifically, the training data set and the test data set are used to evaluate the performance of the model through the k-fold cross-validation method. In the k-fold cross-validation, the data set is divided into subsets, using The remaining subset is used for testing and the training is repeated. times, and finally take the average value as the evaluation result of the model. This method can avoid overfitting and improve the stability of the model.

[0111] By introducing deep learning and ensemble learning methods, the risk assessment module of this embodiment can accurately assess the risks of cardiac surgery based on multimodal data. During the model training process, the accuracy and robustness of risk assessment are improved by optimizing the loss function, introducing regularization, and using ensemble learning. Ultimately, the risk level generated by the risk assessment module can provide a scientific basis for the surgical team to help them formulate the best surgical plan.

[0112] Real-time feedback module for adjusting evaluation models and system parameters based on real-time data during surgery This embodiment relates to a real-time feedback module in an artificial intelligence-based cardiac surgery risk assessment system. The main function of this module is to dynamically adjust the control parameters of the assessment model according to the patient's physiological state during surgery through real-time monitoring data. The real-time feedback module uses the Lyapunov stability theory to ensure that the entire assessment system remains stable and effective during surgery. The introduction of this module enables the risk assessment system to respond to any emergencies during surgery in real time, thereby optimizing risk assessment results and surgical decisions.

[0113] Generally, during cardiac surgery, the patient's physiological state will change constantly, so the output of the evaluation model may change at any time. To this end, the real-time feedback module in this embodiment needs to adjust the control parameters in real time while ensuring the stability of the system to cope with changes in the patient's state. In this case, the Lyapunov stability theory can effectively help us evaluate whether the system remains stable during disturbances or parameter adjustments, thereby avoiding drastic fluctuations in the evaluation results.

[0114] Technical implementation of real-time feedback module In this embodiment, the implementation of the real-time feedback module is based on the Lyapunov stability theory to dynamically adjust the risk assessment model. Specifically, the real-time feedback module monitors the real-time data, updates the control parameters of the calculation model, and verifies the update results through the Lyapunov stability theory to ensure that the system remains stable throughout the entire surgical process.

[0115] In this embodiment, the implementation of the real-time feedback module is based on the Lyapunov stability theory to dynamically adjust the risk assessment model. Specifically, the real-time feedback module monitors the real-time data, updates the control parameters of the calculation model, and verifies the update results through the Lyapunov stability theory to ensure that the system remains stable throughout the entire surgical process.

[0116] Application of Lyapunov Stability Theory Lyapunov stability theory is mainly used to analyze the stability of dynamic systems, especially whether the system can return to its original equilibrium state when it is disturbed or its parameters change. According to Lyapunov theorem, if the system has a Lyapunov function , and the function satisfies the following conditions: , for all (in is the state space), that is Non-negative in the entire state space; , for all ,Right now The derivative of is non-positive in the entire state space; if for If it holds, then the system is asymptotically stable.

[0117] In the real-time feedback module, the Lyapunov function It can represent the energy or error of the system, reflecting the deviation between the model parameters and the actual state. Assume that the state of the system is represented by the vector The control parameters of the evaluation model are expressed as Indicates that the Lyapunov function can be defined as:

[0118] in, is the target state, Represents the Euclidean norm, which reflects the magnitude of the state error. The real-time feedback module obtains and , calculate the current error, and dynamically adjust the control parameters according to the error .

[0119] Dynamic adjustment of control parameters During surgery, real-time data (such as electrocardiogram, wearable device data, etc.) will change over time, so the control parameters of the evaluation model need to be dynamically updated. Generally, the real-time feedback module adjusts the control parameters based on the current patient data through online learning.

[0120] In some embodiments, the real-time feedback module can use an adaptive control algorithm to adjust the control parameters. For example, assuming that the update rule of the control parameters is:

[0121] in, is the learning rate, which indicates the step size of parameter update. is the partial derivative of the Lyapunov function with respect to the control parameter, reflecting the relationship between the current error and the parameter. This update rule enables the control parameter to be adjusted as the error changes, thereby dynamically optimizing the risk assessment model during surgery.

[0122] Parameter update verification and stability assurance In order to ensure that the risk assessment system remains stable during surgery, the real-time feedback module verifies the control parameters after each update through the Lyapunov stability criterion. That is, after each control parameter update, the system checks whether the derivative of the Lyapunov function satisfies , ensuring that parameter adjustment will not cause system instability. If the stability of the system cannot be guaranteed, the real-time feedback module will take appropriate measures, such as reducing the adjustment step size or selecting a different parameter update strategy, to maintain system stability.

[0123] Real-time data processing In this embodiment, the real-time feedback module acquires and processes data from multiple sensors during the operation in real time. Since the changes in sensor data during the operation are extremely complex and have time series characteristics, the real-time feedback module needs to perform time series analysis and prediction on these data.

[0124] Specifically, the sliding window method can be used to sample real-time data and predict and process the data in each time window. Assume that the real-time data is ,in For the The sensor at the time The real-time feedback module uses the prediction model to Predict data at future times to dynamically adjust control parameters.

[0125] In some embodiments, a time series model such as a long short-term memory (LSTM) network can be used to model real-time data. By introducing a gating mechanism, the LSTM network can effectively capture long-term dependencies in time series data, thereby improving the prediction accuracy and stability of the real-time feedback module.

[0126] Real-time feedback control mechanism During surgery, when real-time data changes, the real-time feedback module will dynamically adjust the control parameters in the risk assessment model based on the data prediction results and the analysis of Lyapunov stability theory to adapt to changes in the surgical environment. Specifically, when a physiological parameter exceeds the preset threshold, the real-time feedback module will adjust the weight in the assessment model, thereby affecting the risk assessment results. Through real-time feedback, the surgical team can adjust the treatment plan in a timely manner according to the current patient status, thereby reducing surgical risks.

[0127] The real-time feedback module of this embodiment dynamically adjusts the control parameters through the Lyapunov stability theory to ensure that the system always maintains stability during the operation and responds to the patient's physiological changes in real time. Through adaptive control and real-time data prediction, the real-time feedback module can provide accurate and stable risk assessment results during the operation. The introduction of this module enhances the robustness of the risk assessment system and provides strong support for decision-making during the operation, thereby effectively improving the success rate of the operation and the patient's survival probability.

[0128] Output module for displaying cardiac surgery risk assessment results This embodiment relates to an output module in an artificial intelligence-based cardiac surgery risk assessment system. The main function of this module is to display the results of cardiac surgery risk assessment through a visual interface, including risk level, relevant physiological parameters, and surgical recommendations based on multimodal data prediction. The output module plays a key role in the system. It can not only present the complex data results processed by the system to the surgical team in a clear and understandable way, but also provide real-time support for decision-making, helping medical staff make more scientific and accurate decisions.

[0129] Technical implementation of output modules In this embodiment, the output module cooperates with other modules (such as risk assessment module, real-time feedback module, etc.) to form a complete assessment process. By visualizing the assessment results and recommended measures processed by the system, the output module provides intuitive decision support for the surgical team.

[0130] Structure and form of visual display In some embodiments, the output module displays various types of assessment information through a graphical interface. For example, the risk level can be displayed in the form of a risk chart, a risk heat map, etc., the physiological parameters can be displayed in the form of a dynamic curve chart, a bar chart, etc., and the surgical advice can be presented in the form of a text box or a warning box.

[0131] Specifically, the visualization of risk levels can be represented using a color scale, where each color scale represents a specific risk level (such as low risk, medium risk, and high risk). For example, low risk can be displayed in green, medium risk in yellow, and high risk in red. The use of such a color scale can help the surgical team quickly identify the patient's risk level and make appropriate decisions.

[0132] Physiological parameters (such as heart rate, blood oxygen saturation, blood pressure, etc.) can be presented in the form of a time series graph. The time series graph shows the changing trend of various physiological parameters over time during the operation, which is convenient for the surgical team to monitor the patient's physiological state.

[0133] For surgical recommendations based on the evaluation model, the output module can provide dynamic surgical strategy adjustment recommendations in the form of text boxes, warning boxes, or pop-up windows. For example, when certain key parameters are abnormal, the system will give recommendations such as "adjusting the depth of anesthesia" or "accelerating drug injection." Whenever the operation progresses to a critical node, the output module will provide relevant real-time recommendations to further help medical staff determine whether the treatment plan needs to be adjusted.

[0134] Works with the Risk Assessment Module and the Real-time Feedback Module The output module is closely related to the work of the risk assessment module and the real-time feedback module. After fusing and analyzing the multimodal data, the risk assessment module generates the risk level of cardiac surgery, and the output module generates a visual interface based on these levels. For example, when the risk level is "high risk", the output module will highlight it in red and remind medical staff to pay attention to potential surgical risks.

[0135] In some embodiments, the output module not only displays the patient's current status in real time, but also continuously updates the displayed content based on the dynamic adjustment results of the real-time feedback module. Specifically, the real-time feedback module adjusts the control parameters of the model based on the real-time data that changes during the operation, and these adjustments are immediately transmitted to the output module to update the displayed risk assessment results and surgical recommendations.

[0136] Personalized output content based on machine learning In order to improve the practicality and accuracy of the evaluation results, the output module can also provide personalized output display according to the specific conditions of different patients (such as age, medical history, weight, etc.). In one possible implementation, the system trains historical data through machine learning algorithms to build personalized evaluation models. These personalized models adjust the display of evaluation results based on the physiological characteristics of each patient. Specifically, the machine learning algorithm can predict which physiological indicators have the greatest impact on surgical risk assessment based on the patient's specific parameters, and give priority to displaying the changing trends of these parameters.

[0137] Association of risk assessment results with surgical recommendations In addition to displaying risk levels and physiological parameters, the output module also needs to further enhance the decision-making ability of medical staff through surgical recommendations. The generation of surgical recommendations can be achieved by combining the assessment results with a known medical knowledge base. The knowledge base contains information such as recommended treatment plans, medication plans, surgical steps, etc. under different risk levels. In some embodiments, the output module not only displays changes in physiological parameters, but also automatically recommends adjustment measures or treatment plans based on specific risk assessment results.

[0138] For example, when the system assesses that the patient's surgical risk level is high, the output module can recommend operations such as strengthening the depth of anesthesia, adjusting the amount of anesthetic drugs, and adding monitoring equipment based on the medical knowledge base. These recommendations are generated based on the patient's personalized information and real-time status, and are presented to the surgical team in a visual way.

[0139] Technical implementation of the formula part In the implementation of the output module, the generation of evaluation results not only depends on the model itself, but also involves the application of some mathematical formulas and algorithms.

[0140] Specifically, when the output module receives the output of the risk assessment module, it needs to convert the relevant risk index and physiological parameters into specific visualization content through a series of formulas. Calculated by the following formula:

[0141] in, are physiological parameters in multimodal data (such as heart rate, blood pressure, etc.), is the corresponding weight coefficient, is the evaluation model function. Risk level The value of will change according to different input data and weight coefficients, thereby generating corresponding risk presentation in the output module.

[0142] In some embodiments, the output module can also dynamically calculate the priority of the surgical suggestion according to the following formula:

[0143] in, is the risk level, are the relevant physiological parameters in the current surgical data, Generate a function for surgical recommendations. This function generates appropriate surgical recommendations based on risk level and physiological data.

[0144] Real-time update of data output To ensure that the system responds to changes during surgery in real time, the output module not only needs to regularly update the evaluation results and display of physiological parameters, but also needs to update risk assessments and surgical recommendations based on real-time data. Assuming that the system updates the evaluation results once a second, the output module will recalculate and display the relevant content at each update. Changes in real-time data directly affect the update frequency of the output content, ensuring that the surgical team always obtains the most accurate information.

[0145] The output module in this embodiment provides intuitive and easy-to-understand decision support by displaying the assessment results, risk level, physiological parameters and surgical recommendations to medical staff in a visual form. By working in collaboration with the risk assessment module and the real-time feedback module, the output module can reflect the changes in the patient's physiological state during the operation in real time and generate personalized surgical recommendations based on the assessment results. Combined with machine learning and mathematical models, the design of the output module can ensure the accuracy and practicality of the risk assessment results, thereby providing strong support for decision-making during the operation, ensuring the safety of the patient and the success rate of the operation.

[0146] The embodiment of the present invention also provides a cardiac surgery risk assessment method based on artificial intelligence, comprising the following steps: Collect multimodal data from multiple sensors; De-noise, standardize and time-series align the collected data; Use manifold learning algorithm to reduce the dimensionality of high-dimensional data and use this method to perform nonlinear mapping on the data; Perform information fusion on the processed multimodal data; Use Lie group and Lie algebra theory to align data in space and time; Generate cardiac surgery risk assessment results based on fused data; Adjust evaluation models and system parameters based on real-time data; Output surgical risk assessment results.

[0147] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based cardiac surgery risk assessment system, characterized in that: include: A data acquisition module, used to collect data from multiple sensors, including electrocardiogram, imaging data, and wearable device physiological data; Data preprocessing module, used to denoise, standardize and time-series align the collected data; A nonlinear modeling module, which is used to reduce the dimensionality of high-dimensional data through manifold learning and to perform nonlinear mapping on the data through this method; Data fusion module, used to fuse multimodal data and generate unified feature representation; The spatiotemporal synchronization module is used to align multimodal data in space and time through the theory of Lie groups and Lie algebras; A risk assessment module, used to generate cardiac surgery risk assessment results based on fused multimodal data; A real-time feedback module is used to adjust the evaluation model and system parameters according to real-time data during surgery; Output module for displaying cardiac surgery risk assessment results.

2. The artificial intelligence-based cardiac surgery risk assessment system according to claim 1, characterized in that: The data preprocessing module further includes a denoising unit and a timing alignment unit, wherein the timing alignment unit is used to synchronize data from different sensors according to timestamps.

3. The artificial intelligence-based cardiac surgery risk assessment system according to claim 1, characterized in that: The nonlinear modeling module performs data dimension reduction through a manifold learning algorithm and uses a Gaussian kernel function to perform data nonlinear mapping.

4. The artificial intelligence-based cardiac surgery risk assessment system according to claim 1, characterized in that: The data fusion module adopts the maximum entropy principle to perform information fusion on multimodal data and optimizes the weight of each sensor to improve the accuracy of the evaluation result.

5. The artificial intelligence-based cardiac surgery risk assessment system according to claim 1, characterized in that: The space-time synchronization module transforms the spatial coordinates of multimodal data through Lie group and Lie algebra theory to solve the space-time deviation problem between data.

6. The artificial intelligence-based cardiac surgery risk assessment system according to claim 1, characterized in that: The risk assessment module uses a machine learning algorithm to model the fused data to generate a risk level for cardiac surgery.

7. The artificial intelligence-based cardiac surgery risk assessment system according to claim 1, characterized in that: The real-time feedback module is based on Lyapunov stability theory and dynamically adjusts the control parameters of the evaluation model according to real-time data to ensure that the system remains stable during the operation.

8. The artificial intelligence-based cardiac surgery risk assessment system according to claim 1, characterized in that: The output module displays the cardiac surgery risk assessment results, including risk level, relevant physiological parameters and prediction suggestions, through a visual interface.

9. A cardiac surgery risk assessment method based on artificial intelligence, according to the risk assessment system according to any one of claims 1-8, characterized in that: The following steps are involved: Collect multimodal data from multiple sensors; De-noise, standardize and time-series align the collected data; Use manifold learning algorithm to reduce the dimensionality of high-dimensional data and use this method to perform nonlinear mapping on the data; Perform information fusion on the processed multimodal data; Use Lie group and Lie algebra theory to align data in space and time; Generate cardiac surgery risk assessment results based on fused data; Adjust evaluation models and system parameters based on real-time data; Output surgical risk assessment results.

Images