CN-122025168-A - Deep learning-based early warning system and method for postoperative complications of heart

CN122025168ACN 122025168 ACN122025168 ACN 122025168ACN-122025168-A

Abstract

The invention discloses a heart postoperative complication early warning system and method based on deep learning, which relates to the technical field of medical artificial intelligence, the system comprises a data acquisition module, a data preprocessing module, a feature extraction module, a feature fusion module, a complication prediction module and a warning output module, wherein the data acquisition module acquires electrocardiosignals, ultrasonic cardiac images and vital sign data, the feature extraction module adopts a convolutional neural network to extract image space features, adopts a bidirectional long-short-term memory network to extract time sequence characteristics, a characteristic fusion module realizes multi-mode characteristic fusion through a concentration mechanism, a complication prediction module predicts the occurrence probability of arrhythmia, hypocenter syndrome, pericardial tamponade, postoperative infection and postoperative hemorrhage, the early warning output module carries out three-level early warning according to the probability value, and the invention realizes continuous monitoring and early warning of complications of patients after cardiac operation and improves the safety of clinical nursing.

Inventors

YE WENJUAN
JIANG XIUMING
XU YAN
MA HANXIANG
WU FANG

Assignees

安徽医科大学第一附属医院
合肥市盛文信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260123

Claims (10)

1. Deep learning-based early warning system for postoperative complications of heart, comprising: The data acquisition module is used for acquiring multi-mode physiological data of a patient after cardiac operation, wherein the multi-mode physiological data comprises electrocardiosignal data, ultrasonic cardiac image data and vital sign data; The data preprocessing module is connected with the data acquisition module and is used for carrying out denoising processing, normalization processing and data alignment processing on the multi-mode physiological data to obtain standardized physiological data; the feature extraction module is connected with the data preprocessing module and comprises a convolutional neural network unit and a cyclic neural network unit, wherein the convolutional neural network unit is used for carrying out space feature extraction on the ultrasonic cardiac image data to obtain an image feature vector, and the cyclic neural network unit is used for carrying out time sequence feature extraction on the electrocardiosignal data and vital sign data to obtain a time sequence feature vector; The feature fusion module is connected with the feature extraction module and is used for carrying out splicing fusion processing on the image feature vector and the time sequence feature vector to obtain a multi-mode fusion feature vector; the complication prediction module is connected with the feature fusion module and comprises a fully-connected neural network layer and a classification output layer, and is used for predicting occurrence probability values of heart postoperative complications according to the multi-mode fusion feature vector; And the early warning output module is connected with the complications prediction module and is used for generating early warning information according to the comparison result of the occurrence probability value and a preset threshold value.
2. The early warning system for heart postoperative complications based on deep learning of claim 1, wherein the convolutional neural network unit comprises a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer, a third convolutional layer and a global average pooling layer which are sequentially connected; The first convolution layer adopts 64 convolution kernels with the size of 3×3; the second convolution layer adopts 128 convolution kernels with the size of 3×3; the third convolution layer adopts 256 convolution kernels with the size of 3×3; the first pooling layer and the second pooling layer both adopt a maximum pooling operation with a size of 2 x 2.
3. The advanced learning-based early warning system for heart postoperative complications, as set forth in claim 1, wherein the recurrent neural network unit adopts a two-way long-short-term memory network structure, and comprises a forward long-short-term memory network layer and a reverse long-short-term memory network layer; the forward long-period memory network layer and the reverse long-period memory network layer both comprise 128 hidden units; and the output of the two-way long-short-term memory network structure is the splicing result of the forward hiding state and the reverse hiding state.
4. The advanced learning based early warning system for cardiac postoperative complications, as set forth in claim 1, wherein the type of cardiac postoperative complications predicted by the complications prediction module includes arrhythmia, hypocenter syndrome, pericardial tamponade, postoperative infection, and postoperative bleeding; and the classification output layer outputs the occurrence probability value corresponding to each complication type by adopting a Softmax activation function.
5. The advanced learning-based early warning system for cardiac postoperative complications according to claim 1, wherein the early warning output module comprises: the threshold value comparison unit is used for comparing the occurrence probability value of each complication type with a corresponding preset threshold value; the early warning grade judging unit is used for determining early warning grades according to the comparison result, wherein the early warning grades comprise a low risk grade, a medium risk grade and a high risk grade; and the early warning information generation unit is used for generating early warning information comprising the complication type, the risk level and the suggested treatment measures according to the early warning level.
6. The early warning method for the postoperative complications of the heart based on deep learning is characterized by being suitable for the early warning system for the postoperative complications of the heart based on deep learning, and comprises the following steps of: S1, acquiring multi-mode physiological data of a patient after heart operation, wherein the multi-mode physiological data comprises electrocardiosignal data, echocardiographic image data and vital sign data; S2, denoising, normalizing and data alignment are carried out on the multi-mode physiological data to obtain standardized physiological data; S3, performing spatial feature extraction on the ultrasonic cardiac image data by using a convolutional neural network to obtain an image feature vector; S4, extracting time sequence features of the electrocardiosignal data and vital sign data by using a cyclic neural network to obtain time sequence feature vectors; s5, performing splicing and fusion processing on the image feature vector and the time sequence feature vector to obtain a multi-mode fusion feature vector; s6, processing the multi-mode fusion feature vector by using a fully-connected neural network, and predicting occurrence probability values of complications after cardiac surgery; and S7, generating early warning information according to a comparison result of the occurrence probability value and a preset threshold value.
7. The early warning method for heart postoperative complications based on deep learning according to claim 6, wherein in step S2, the denoising process adopts a wavelet transform denoising method; the normalization processing adopts a Z-score normalization method to convert data into standard normal distribution with the mean value of 0 and the standard deviation of 1; and the data alignment processing adopts a time stamp alignment mode, and unifies the data with different acquisition frequencies to the same time reference.
8. The method for early warning of post-operative complications of heart based on deep learning of claim 6, wherein in step S3, the training process of the convolutional neural network comprises: Calculating the difference between the predicted value and the real label by adopting a cross entropy loss function; Adopting an Adam optimizer to update network parameters, and setting the learning rate to be 0.001; The Dropout regularization method was used to prevent overfitting, and the Dropout ratio was set to 0.5.
9. The advanced learning-based early warning method for cardiac postoperative complications according to claim 6, wherein in step S5, the specific process of the stitching fusion process comprises: Vector stitching is carried out on the image feature vector and the time sequence feature vector along the feature dimension to obtain an initial fusion vector; And inputting the initial fusion vector into an attention mechanism layer for weighting processing to obtain a multi-mode fusion feature vector.
10. The method for early warning of post-operative complications of heart based on deep learning of claim 6, wherein in step S7, the generation process of the early warning information comprises the following steps: when the occurrence probability value is larger than 0.7, judging that the risk level is high, and generating emergency early warning information; When the occurrence probability value is larger than 0.4 and smaller than 0.7, judging that the risk level is middle, and generating general early warning information; when the occurrence probability value is smaller than 0.4, the risk level is judged to be low, and observation prompt information is generated.

Description

Deep learning-based early warning system and method for postoperative complications of heart Technical Field The invention relates to the technical field of medical artificial intelligence, in particular to a heart postoperative complications early warning system and method based on deep learning. Background Cardiac surgery is an important means for treating coronary heart disease, heart valve disease, aortic dissection lesions and congenital heart disease, with the continuous development of cardiac surgery technology, the success rate of surgery is improved significantly, but postoperative complications still are key factors affecting the prognosis and the quality of life of patients, and common complications after cardiac surgery include arrhythmia, hypocenter syndrome, pericardial tamponade, postoperative infection and postoperative bleeding, which can seriously threaten the life safety of patients if not found and treated in time. Traditional post-operative cardiac monitoring relies mainly on the periodic observation and subjective judgment of patient vital signs by medical staff, and has the following disadvantages: Firstly, the continuity of manual monitoring is insufficient, and the continuous comprehensive monitoring for 24 hours is difficult to realize; Second, the difference in experience of healthcare workers results in uneven ability to identify early signs of complications; Thirdly, the monitoring of a single index is difficult to comprehensively evaluate the overall state of a patient, and early signals of complex complications are easy to miss; Fourth, insufficient timeliness of early warning is often found after complications have developed to some extent. In recent years, the deep learning technology has made remarkable progress in the fields of medical image analysis and physiological signal processing, provides a new technical path for early warning of heart postoperative complications, is excellent in the aspect of medical image feature extraction, and has unique advantages in the aspect of time sequence physiological signal analysis, however, the existing deep learning-based medical early warning system is mainly used for analyzing a single data mode, lacks comprehensive fusion processing capability of multi-mode physiological data, and is difficult to fully mine the associated information among different data sources. Therefore, there is a need to develop an intelligent system and method that can fuse multimodal physiological data to enable early warning of post-operative complications of the heart. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a deep learning-based early warning system and method for heart postoperative complications, which are used for realizing early accurate prediction and hierarchical early warning of heart postoperative complications by fusing multi-mode data such as electrocardiosignals, ultrasonic cardiac images, vital signs and the like and utilizing a convolutional neural network and a cyclic neural network to extract and fuse features. In order to achieve the above purpose, the invention adopts the following technical scheme: A deep learning-based early warning system for post-operative complications of the heart, comprising: The data acquisition module is used for acquiring multi-mode physiological data of a patient after cardiac operation, wherein the multi-mode physiological data comprises electrocardiosignal data, ultrasonic cardiac image data and vital sign data. The data preprocessing module is connected with the data acquisition module and is used for carrying out denoising processing, normalization processing and data alignment processing on the multi-mode physiological data to obtain standardized physiological data. The feature extraction module is connected with the data preprocessing module and comprises a convolutional neural network unit and a cyclic neural network unit, wherein the convolutional neural network unit is used for carrying out space feature extraction on the ultrasonic cardiac image data to obtain an image feature vector, and the cyclic neural network unit is used for carrying out time sequence feature extraction on the electrocardiosignal data and the vital sign data to obtain a time sequence feature vector. And the feature fusion module is connected with the feature extraction module and is used for carrying out splicing and fusion processing on the image feature vector and the time sequence feature vector to obtain a multi-mode fusion feature vector. And the complications prediction module is connected with the characteristics fusion module and comprises a fully-connected neural network layer and a classification output layer, and is used for predicting occurrence probability values of heart postoperative complications according to the multi-mode fusion characteristic vector. And the early warning output module is connected with the complications prediction module and is used for generati