Method and system for predicting low-voltage area of left room based on artificial intelligence

Abstract

The invention provides a left atrium low-voltage area prediction method and system based on artificial intelligence, and the method comprises the steps: collecting an electrocardiogram data set of a target group, and carrying out the preprocessing of the electrocardiogram data set, and obtaining an input image; a backbone network is selected, an original full-connection layer of the backbone network is replaced with a fusion full-connection layer, a neural network is obtained, and the input of the fusion full-connection layer is the sum of the output number of a second last layer and the number of baseline features; training the neural network through the input image to obtain a prediction model; and inputting the baseline data of the to-be-tested person after feature splicing and the electrocardiogram of the to-be-tested person into the prediction model, and performing left atrium low-voltage region prediction on the to-be-tested person through the prediction model to obtain a prediction result. According to the method, complex features related to the left atrium low-voltage region can be learned, the accuracy of prediction of the left atrium low-voltage region is remarkably improved, and the prediction efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical informatics, and in particular to a method and system for predicting a left atrial low voltage area based on artificial intelligence. Background Art

[0002] Atrial fibrillation is the most common sustained arrhythmia in clinical practice. Catheter ablation is the first-line treatment for symptomatic atrial fibrillation. However, the success rate of catheter ablation for atrial fibrillation is still not ideal. Many studies have shown that the left atrial low voltage area is an important risk factor for the recurrence of atrial fibrillation. Studies have shown that for patients with atrial fibrillation with a left atrial low voltage area, the risk of atrial fibrillation recurrence after surgery is significantly increased. If ablation of the left atrial low voltage area is added to the conventional pulmonary vein isolation during surgery, the success rate of the operation can be improved. Therefore, identifying left atrial low voltage is very important for assessing the risk of atrial fibrillation recurrence and selecting the ablation strategy for catheter ablation.

[0003] The left atrial low voltage area is obtained through electrophysiological mapping during catheter ablation. The left atrial gadolinium delay phenomenon shown by cardiac enhanced magnetic resonance imaging is well correlated with the left atrial low voltage area. In addition, some clinical indicators including age, gender, left atrial diameter, left ventricular ejection fraction, renal insufficiency, diabetes, etc. are also correlated with the left atrial low voltage area in patients with atrial fibrillation. Based on these indicators, some scores for predicting the left atrial low voltage area have been established, such as the DR-FLASH score and the ZAQ score.

[0004] However, the spatial resolution of the currently used cardiac enhanced magnetic resonance imaging is approximately 1.2-1.5 mm, while the atrial wall thickness is only 2-4 mm, so this method cannot display the atrial matrix well; in addition, these existing clinical scoring systems were developed in relatively small cohorts, and the largest study, the DR-FLASH study, only included 238 patients with atrial fibrillation, resulting in poor predictive accuracy.

[0005] It is possible to establish an atrial fibrillation prediction model through artificial intelligence technology. However, at present, artificial intelligence technology cannot be used to directly predict the recurrence of atrial fibrillation from the sinus rhythm electrocardiogram, and it cannot assist doctors in making strategic decisions on catheter ablation.

[0006] Therefore, this field needs an artificial intelligence-based prediction method to predict left atrial low voltage by analyzing the sinus rhythm electrocardiogram, so as to provide a reference for clinicians to predict the risk of atrial fibrillation recurrence and make strategic decisions on nuclear catheter ablation. Summary of the invention

[0007] The present invention provides a left atrial low voltage area prediction method and system based on artificial intelligence, so as to solve the defects of the prior art.

[0008] The present invention provides a method for predicting a left atrial low voltage area based on artificial intelligence, comprising: S1: collecting an electrocardiogram dataset of a target group and preprocessing images in the electrocardiogram dataset to obtain an input image; the electrocardiogram dataset includes a sinus rhythm electrocardiogram, baseline data of the sinus rhythm electrocardiogram, and a low voltage label of the sinus rhythm electrocardiogram; S2: Select a backbone network, replace the original fully connected layer of the backbone network with a fused fully connected layer, and obtain a neural network, wherein the input of the fused fully connected layer is the sum of the number of outputs of the next last layer and the number of baseline features; S3: training the neural network through the input image to obtain a prediction model; S4: Input the baseline data of the subject after feature splicing and the electrocardiogram image of the subject into the prediction model, and predict the left atrial low voltage area of ​​the subject through the prediction model to obtain a prediction result.

[0009] According to a left atrial low voltage zone prediction method based on artificial intelligence provided by the present invention, the target group in step S1 includes a group of atrial fibrillation patients with left atrial low voltage zone and a group of atrial fibrillation patients without left atrial low voltage zone; the subjects to be tested in step S4 are patients with arrhythmia and at risk of atrial fibrillation, patients with atrial fibrillation or patients after atrial fibrillation surgery.

[0010] According to a left atrial low voltage area prediction method based on artificial intelligence provided by the present invention, the preprocessing of the electrocardiogram data in step S1 specifically includes: distorting the image in the electrocardiogram dataset at a random ratio calculated by random sample values, and filling the boundary of the distorted image in the electrocardiogram dataset with gray pixel values; flipping the images in the electrocardiogram data set after padding according to the random sample value; Converting the color space of the image in the electrocardiogram dataset using the random sample value as a scale, wherein the random sample value obeys a standard normal distribution; The images in the electrocardiogram dataset are normalized.

[0011] According to a left atrial low voltage area prediction method based on artificial intelligence provided by the present invention, when the random sample value is greater than or equal to a preset threshold, the image in the filled electrocardiogram data set is flipped; when the random sample value is less than the preset threshold, the image in the filled electrocardiogram data set is not flipped; wherein the preset threshold is 0.5.

[0012] According to a left atrial low voltage area prediction method based on artificial intelligence provided by the present invention, the expression of the random ratio is:

[0013] in, is the random ratio, Expressing obedience A random sample value uniformly distributed over an interval, is the first preset number, is the second preset number.

[0014] According to a left atrial low voltage area prediction method based on artificial intelligence provided by the present invention, step S2 further includes: S21: Randomly initialize the network parameters of the backbone network; S22: Define the loss function between the true label and the predicted result.

[0015] According to an artificial intelligence-based left atrial low voltage area prediction method provided by the present invention, in step S21, the randomly initialized network parameters obey a normal distribution with a mean of 0 and a variance of 1.

[0016] According to a left atrial low voltage area prediction method based on artificial intelligence provided by the present invention, the loss function expression of the neural network in step S22 is:

[0017] in, is the loss function of the neural network, is the total number of input images, is the input image index value, is the first weight parameter, is the second weight parameter, For the The predicted probability of the input image.

[0018] According to the artificial intelligence-based left atrial low voltage area prediction method provided by the present invention, in step S3, when the prediction model is obtained through training, the neural network is converged by the Adam optimization algorithm.

[0019] According to a left atrial low voltage area prediction method based on artificial intelligence provided by the present invention, step S4 further comprises: S41: Input the baseline data of the subject and the electrocardiogram image of the subject; S42: performing convolution on the electrocardiogram image of the subject to be tested to obtain electrocardiogram features of the subject to be tested; S43: performing feature splicing on the baseline data of the subject and the electrocardiogram features of the subject to obtain splicing features; S44: Input the splicing features into the prediction model to obtain a prediction result.

[0020] The present invention also provides a left atrial low voltage area prediction system based on artificial intelligence, comprising: Image processing module: used for collecting the electrocardiogram data set of the target group and preprocessing the electrocardiogram data set to obtain an input image; A construction module is used to select a backbone network, replace the original fully connected layer of the backbone network with a fused fully connected layer, and obtain a neural network, wherein the input of the fused fully connected layer is the sum of the number of outputs of the next last layer and the number of baseline features; Training module: used to train the neural network obtained by the building module through the input image obtained by the image processing module to obtain a prediction model; Prediction module: used to collect the baseline data and electrocardiogram of the subject and input them into the prediction model after feature splicing, and predict the left atrial low voltage area of ​​the subject through the prediction model to obtain the prediction result.

[0021] The present invention provides an artificial intelligence-based left atrial low voltage area prediction method and system, which increases the input information dimension of the model by constructing a fusion fully connected layer based on image processing of sinus electrocardiogram pictures, so that the model can more comprehensively learn the complex features related to the left atrial low voltage area, thereby significantly improving the accuracy of the prediction; secondly, the present invention not only relies on a single electrocardiogram feature, but also integrates multiple information sources, including advanced features extracted from electrocardiogram data and baseline features, which helps the model to show stronger generalization ability when processing electrocardiogram data of different patients and under different conditions, enhances the interpretability of the model, and improves the universality and practicality of the method; in addition, the present invention is highly automated in the entire prediction process from electrocardiogram data collection and preprocessing to model training and prediction, which greatly reduces the workload of medical staff and improves work efficiency. By accurately predicting the left atrial low voltage area, it is helpful to detect potential heart diseases or abnormalities at an early stage, provide timely diagnosis and treatment suggestions for clinicians, and facilitate early intervention and rehabilitation of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 A schematic flow chart of a method for predicting a left atrial low voltage area based on artificial intelligence provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a left atrial low voltage area prediction system based on artificial intelligence provided by an embodiment of the present invention; Figure 3A schematic diagram of the prediction principle of a left atrial low voltage area prediction method based on artificial intelligence provided in an embodiment of the present invention.

[0024] Reference numerals: 100, image processing module; 200, construction module; 300, training module; 400, prediction module. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. 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. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0026] Combine the following Figures 1 to 3 Embodiments of the present invention are described.

[0027] like Figure 1 As shown, the present invention provides a left atrial low voltage area prediction method based on artificial intelligence, comprising: S1: Collecting an electrocardiogram dataset of a target group and preprocessing images in the electrocardiogram dataset to obtain an input image; the electrocardiogram dataset includes a sinus rhythm electrocardiogram, baseline data of the sinus rhythm electrocardiogram, and a low voltage label of the sinus rhythm electrocardiogram.

[0028] The target group in step S1 includes an atrial fibrillation patient group with a left atrial low voltage area and an atrial fibrillation patient group without a left atrial low voltage area.

[0029] In step S1, an electrocardiogram data set required for identifying left atrial low voltage prediction is first constructed. The data set mainly includes a sinus rhythm electrocardiogram, baseline data corresponding to the electrocardiogram, and a low voltage label.

[0030] The present invention uses sinus electrocardiogram as basic data collection because under sinus rhythm, the P wave of the electrocardiogram represents the depolarization process of the atrium, and its shape and amplitude are relatively stable, which is convenient for observation and measurement. Under atrial fibrillation, the P wave disappears and is replaced by f waves of varying sizes, shapes, and spacings. This makes the analysis of the sinus electrocardiogram more concise than the analysis of the atrial fibrillation electrocardiogram. In addition, under sinus rhythm, parameters such as the amplitude and interval of the P wave are significantly correlated with the left atrial low voltage area. For example, a decrease in the P wave amplitude and a prolongation of the P wave interval can predict left atrial low voltage, thereby improving accuracy.

[0031] In addition, sinus rhythm is the normal rhythm of the heart, so it is more common in clinical practice. Using the electrocardiogram parameters under sinus rhythm to predict the low voltage area of ​​the left atrium can be applicable to a wider range of patient groups; while the electrocardiogram analysis under atrial fibrillation rhythm is relatively complicated, and atrial fibrillation patients themselves may have abnormalities in heart structure and function, which further increases the difficulty of analysis.

[0032] Specifically, to obtain sinus rhythm ECG images, it is necessary to obtain paper ECGs through clinical practice, and use scanning equipment to scan the paper ECGs into image format and store them; and to obtain baseline data and low-voltage labels, the baseline data and low-voltage labels corresponding to the sinus rhythm ECGs are recorded by searching the medical history.

[0033] After obtaining the above data, the data set is produced. Specifically, the paper ECG images, baseline data and low voltage labels are integrated to obtain the left atrial low voltage prediction data set, which contains sinus rhythm paper ECG and table files containing baselines and labels; then the training set, validation set and test set are divided according to the ratio of 8:1:1, and the array is shuffled by the random.shuffle function in numpy, and the data are selected one by one.

[0034] The preprocessing of the electrocardiogram data in step S1 specifically includes: distorting the image in the electrocardiogram dataset at a random ratio calculated by random sample values, and filling the boundary of the distorted image in the electrocardiogram dataset with gray pixel values; flipping the images in the electrocardiogram data set after padding according to the random sample value; Converting the color space of the image in the electrocardiogram dataset using the random sample value as a scale, wherein the random sample value obeys a standard normal distribution; The images in the electrocardiogram dataset are normalized.

[0035] Specifically, the obtained data set is distorted at a random ratio and the upper, lower, left and right edges are filled with gray pixel values ​​to ensure that the image will not be distorted after being sampled to a fixed size. The image is normalized so that the pixel value of the image is scaled to between 0 and 1 in order to convert it into a tensor.

[0036] When the random sample value is greater than or equal to a preset threshold, the image in the filled electrocardiogram data set is flipped; when the random sample value is less than the preset threshold, the image in the filled electrocardiogram data set is not flipped; wherein the preset threshold is 0.5.

[0037] Furthermore, the generated random number is uniformly distributed between 0 and 1, so in this stage the preset threshold is set to the median 0.5, and the random number is compared with 0.5. If the random number is less than or equal to 0.5, a flipping operation is performed, and if it is greater than 0.5, no flipping is performed. The purpose of the flipping operation is usually to increase the diversity of the data, thereby helping to avoid overfitting when training the model and improving the generalization ability of the model. In the context of image processing and data enhancement, flipping is a common operation because it can simulate images from different perspectives, thereby helping the model learn more robust features.

[0038] Wherein, the expression of the random ratio is:

[0039] in, is the random ratio, Expressing obedience A random sample value uniformly distributed over an interval, is the first preset number, is the second preset number.

[0040] In the above formula, rand(0,1) can obtain a random sample value that obeys a uniform distribution of 0-1, and a and b are pre-set values. This formula can be used to obtain a random number between ab.

[0041] S2: Select a backbone network, replace the original fully connected layer of the backbone network with a fused fully connected layer, and obtain a neural network, wherein the input of the fused fully connected layer is the sum of the number of outputs of the next-to-last layer and the number of baseline features.

[0042] Preferably, the present invention selects the ResNet model as the backbone network, which can obtain higher accuracy when computing resources are limited, and obtain better accuracy benefits when the network depth is increased.

[0043] When building a network, first determine the input and output of the neural network. The input of the deep neural network is the image data after preprocessing in step S1. , the output of the deep neural network is the predicted probability, where is the batch size fed into the network, is the number of channels of the image, and are the length and width of the preprocessed image.

[0044] Further, in step S2, when building the model, a fused fully connected layer is set, which not only receives the output features from the penultimate layer, but also integrates the baseline features, which enables the network to comprehensively consider more information sources, improves the robustness and generalization ability of the model, and by integrating features at different levels, the model can learn a richer and more comprehensive feature representation; by introducing baseline features, the fused fully connected layer can also better adapt to the characteristics of different data sets, and the fused fully connected layer can flexibly adjust its weights and bias items to adapt to these changes; in addition, the fully connected layer itself usually contains a large number of parameters, and the fused fully connected layer set by the present invention can reduce parameter redundancy to a certain extent, so that the model can achieve better performance with fewer parameters, and reducing parameter redundancy can not only reduce the complexity of the model, but also improve computational efficiency. During the training and reasoning process, the amount of data and the amount of calculation that the model needs to process will be reduced accordingly, thereby speeding up the processing speed and reducing resource consumption.

[0045] Wherein, step S2 further includes: S21: Randomly initialize the network parameters of the backbone network.

[0046] In step S21, the randomly initialized network parameters obey a normal distribution with a mean of 0 and a variance of 1.

[0047] S22: Define the loss function between the true label and the predicted result.

[0048] The loss function expression of the neural network in step S22 is:

[0049] in, is the loss function of the neural network, is the total number of input images, is the input image index value, is the first weight parameter, is the second weight parameter, For the The predicted probability of the input image.

[0050] S3: Training the neural network through the input image to obtain a prediction model.

[0051] Wherein, in step S3, when training to obtain the prediction model, the neural network is converged by using the Adam optimization algorithm.

[0052] The specific training process is as follows: first, the forward propagation algorithm is used to extract image features to predict the input and obtain the prediction probability; then the difference between the prediction result and the true label is obtained through the loss function defined in step S2; secondly, the back propagation algorithm based on the chain rule and the Adam optimization algorithm are combined to perform gradient updates on the weight parameters in the deep neural network, and then the deep neural network with updated weight parameters is used to repeat the above steps until the deep neural network converges; in the process of updating the weight parameters, the change curve of the loss value is recorded, and the weight parameters are saved by presetting the corresponding threshold for the loss value to obtain the final deep neural network.

[0053] S4: Input the baseline data of the subject after feature splicing and the electrocardiogram image of the subject into the prediction model, and predict the left atrial low voltage area of ​​the subject through the prediction model to obtain a prediction result.

[0054] The subject to be tested in step S4 is a patient with arrhythmia and a risk of atrial fibrillation, a patient with atrial fibrillation, or a patient after atrial fibrillation surgery.

[0055] A prediction model can be obtained through steps S1 to S3, and then the prediction model can be used for the electrocardiogram to be predicted. The prediction result gives the probability of whether it is a low-voltage area. The prediction probability can be used to determine whether there is a low-voltage area in the patient's left atrium. This form of result is the most intuitive and easy to understand.

[0056] Wherein, step S4 further comprises: S41: Input the baseline data of the subject and the electrocardiogram image of the subject; S42: performing convolution on the electrocardiogram image of the subject to be tested to obtain electrocardiogram features of the subject to be tested; S43: performing feature splicing on the baseline data of the subject and the electrocardiogram features of the subject to obtain splicing features; S44: Input the splicing features into the prediction model to obtain a prediction result.

[0057] like Figure 3 As shown, when the prediction model obtained by the above steps is used for prediction, the baseline data of the subject, including gender, age, height, etc., are first input as the first aspect, and the electrocardiogram image is input as the second aspect, and then Figure 3 The ECG image feature extraction model shown in the convolution operation performs convolution on the input ECG image to extract features, and can obtain 35-dimensional features of baseline data and 2048-dimensional features of ECG images respectively. Then, the features are concatenated to obtain 2083-dimensional features, and the concatenated features are then input into the prediction model, that is, Figure 3 The fully connected classification model shown in can obtain the final prediction results, including two categories of non-left atrial low voltage area and left atrial low voltage area, and their respective probabilities.

[0058] Furthermore, the convolutional network structure of the above-mentioned ECG image feature extraction model for convolution of the subject's ECG is as follows: the first layer of the network uses a 7x7 convolution kernel, 64 filters, and a step size of 2 to perform a convolution operation on the input image, and the output size is 112x112. The second layer of the network first passes through a 3x3 maximum pooling layer with a step size of 2 to further reduce the size of the feature map. Then, the second to fifth layers of the network pass through a combination of 1x1, 3x3, and 1x1 convolution layers, respectively. As the number of layers increases, the number of convolution kernels (i.e., the number of channels) gradually increases, and higher-level features are gradually extracted. Finally, average pooling is performed through the average layer of the softmax function. The total number of floating-point operations is .

[0059] In the prediction step, the present invention predicts the left atrial low-voltage electrical area through the baseline data and electrocardiogram images of the subject, which can not only comprehensively consider multiple factors, but also more comprehensively understand the health status of the subject through multi-dimensional data fusion, reduce the deviation that may be caused by a single data source, and thus improve the accuracy of the prediction of the left atrial low-voltage electrical area.

[0060] like Figure 2 As shown, the present invention also provides a left atrial low voltage area prediction system based on artificial intelligence, comprising: Image processing module 100: used for collecting an electrocardiogram dataset of a target group and preprocessing images in the electrocardiogram dataset to obtain an input image; Construction module 200: for selecting a backbone network, replacing the original fully connected layer of the backbone network with a fused fully connected layer, and obtaining a neural network, wherein the input of the fused fully connected layer is the sum of the number of outputs of the next last layer and the number of baseline features; Training module 300: used to train the neural network obtained by the construction module 200 through the input image obtained by the image processing module 100 to obtain a prediction model; Prediction module 400: used to collect baseline data and electrocardiogram of the subject and input them into the prediction model after feature splicing, and predict the left atrial low voltage area of ​​the subject through the prediction model to obtain the prediction result.

[0061] In a specific embodiment, the present invention analyzes 12-lead electrocardiograms corresponding to 1,133 patients with atrial fibrillation who underwent catheter ablation. The final training set contains 906 images, the validation set contains 113 images, and the test set contains 114 images. In the final test results, taking the electrocardiogram as an example, the area under the ROC curve (AUROC) of the convolutional neural network (CNN) model is 0.704, which is better than the DR-FLASH score (AUROC=0.601) and the APPLE score (AUROC=0.589), and two multimodal artificial intelligence models that integrate electrocardiogram images and clinical baseline features show higher diagnostic accuracy (CNN-Multimodal and CNN-Random The AUROCs of the Forest-Multimodal model were 0.816 and 0.796 respectively), and the AUROCs of the model of the present invention in the single ECG model, CNN-multimodal model and CNN random forest multimodal model were 0.711, 0.785 and 0.879 respectively. It can be seen that the model proposed in the present invention performed better than the above models in the multi-center validation data set.

[0062] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0063] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0064] The present invention provides an artificial intelligence-based left atrial low voltage zone prediction method and system, which uses sinus electrocardiogram as a data set, so that the electrocardiogram features are clearer, the prediction accuracy is higher, the application is wider, and it is convenient for preoperative evaluation and surgical planning. In further improvements, the prediction model can also mark the specific location and range of the low voltage zone on the electrocardiogram image, providing doctors with more detailed reference information, and may also output some quantitative indicators, such as the area and voltage level of the low voltage zone, so that doctors can conduct more in-depth analysis and evaluation. The present invention can serve as an important tool for predicting left atrial low voltage zones.

[0065] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting left atrial low voltage area based on artificial intelligence, characterized in that: include: S1: collecting an electrocardiogram dataset of a target group and preprocessing images in the electrocardiogram dataset to obtain an input image; The electrocardiogram data set includes a sinus rhythm electrocardiogram, baseline data of the sinus rhythm electrocardiogram, and a low voltage label of the sinus rhythm electrocardiogram; S2: Select a backbone network, replace the original fully connected layer of the backbone network with a fused fully connected layer, and obtain a neural network, wherein the input of the fused fully connected layer is the sum of the number of outputs of the next last layer and the number of baseline features; S3: training the neural network through the input image to obtain a prediction model; S4: Input the baseline data of the subject after feature splicing and the electrocardiogram image of the subject into the prediction model, and predict the left atrial low voltage area of ​​the subject through the prediction model to obtain a prediction result.

2. The method for predicting left atrial low voltage area based on artificial intelligence according to claim 1, characterized in that: The target group in step S1 includes a group of atrial fibrillation patients with left atrial low voltage area and a group of atrial fibrillation patients without left atrial low voltage area; the subjects in step S4 are patients with arrhythmia and atrial fibrillation risk, atrial fibrillation patients or patients after atrial fibrillation surgery.

3. The method for predicting left atrial low voltage area based on artificial intelligence according to claim 1, characterized in that: The preprocessing of the electrocardiogram data set in step S1 specifically includes: distorting the image in the electrocardiogram dataset at a random ratio calculated by random sample values, and filling the boundary of the distorted image in the electrocardiogram dataset with gray pixel values; flipping the images in the electrocardiogram data set after padding according to the random sample value; Converting the color space of the image in the electrocardiogram dataset using the random sample value as a scale, wherein the random sample value obeys a standard normal distribution; The images in the electrocardiogram dataset are normalized.

4. The method for predicting left atrial low voltage area based on artificial intelligence according to claim 3, characterized in that: When the random sample value is greater than or equal to a preset threshold, the image in the filled electrocardiogram data set is flipped; when the random sample value is less than the preset threshold, the image in the filled electrocardiogram data set is not flipped; wherein the preset threshold is 0.

5.

5. The method for predicting left atrial low voltage area based on artificial intelligence according to claim 3, characterized in that: The expression of the random ratio is: in, is the random ratio, Expressing obedience Random sample values ​​uniformly distributed in the interval, is the first preset number, is the second preset number.

6. The method for predicting left atrial low voltage area based on artificial intelligence according to claim 1, characterized in that: Step S2 also includes: S21: Randomly initialize the network parameters of the backbone network; S22: Define the loss function between the true label and the predicted result.

7. The method for predicting left atrial low voltage area based on artificial intelligence according to claim 6, characterized in that: In step S21, the randomly initialized network parameters obey a normal distribution with a mean of 0 and a variance of 1.

8. The method for predicting left atrial low voltage area based on artificial intelligence according to claim 6, characterized in that: The loss function expression of the neural network in step S22 is: in, is the loss function of the neural network, is the total number of input images, is the input image index value, is the first weight parameter, is the second weight parameter, For the The predicted probability of the input image.

9. The method for predicting left atrial low voltage area based on artificial intelligence according to claim 1, characterized in that: Step S4 further comprises: S41: Input the baseline data of the subject and the electrocardiogram image of the subject; S42: performing convolution on the electrocardiogram image of the subject to be tested to obtain electrocardiogram features of the subject to be tested; S43: performing feature splicing on the baseline data of the subject and the electrocardiogram features of the subject to obtain splicing features; S44: Input the splicing features into the prediction model to obtain a prediction result.

10. A left atrial low voltage area prediction system based on artificial intelligence, characterized in that: include: Image processing module: used for collecting the electrocardiogram data set of the target group and preprocessing the images in the electrocardiogram data set to obtain input images; Construction module: used for selecting a backbone network, replacing the original fully connected layer of the backbone network with a fused fully connected layer, and obtaining a neural network, wherein the input of the fused fully connected layer is the sum of the number of outputs of the next last layer and the number of baseline features; Training module: used to train the neural network obtained by the building module through the input image obtained by the image processing module to obtain a prediction model; Prediction module: used to collect baseline data and electrocardiogram images of the subject and input them into the prediction model after feature splicing, and predict the left atrial low voltage area of ​​the subject through the prediction model to obtain the prediction result.

Images