Congenital heart disease prediction method and system

Abstract

The invention provides a congenital heart disease prediction method and system. The method comprises the following steps: acquiring an echocardiogram and a genome sequence of a patient; inputting the echocardiogram of the patient into a multi-task learning classification and segmentation model to obtain a first classification prediction result and a focus region segmentation result; inputting the genome sequence of the patient into a machine learning model based on a Bayesian algorithm to obtain a second classification prediction result; and obtaining a first classification prediction result, a focus region segmentation result and a second classification prediction result according to the congenital heart disease of the patient, and obtaining a final classification prediction result and a focus region segmentation result of the congenital heart disease of the patient. The method comprises the following steps: firstly, analyzing and classifying echocardiograms of a patient by using a deep learning technology; performing regression and learning on the gene sequence of the patient by using a machine learning method; and finally, combining the classification result of the echocardiogram and the gene sequence classification result of the patient to accurately identify and predict the congenital heart disease of the patient.

Description

Technical Field

[0001] The present invention relates to the medical field, and more specifically, to a congenital heart disease prediction method and system. Background Art

[0002] At present, the common methods for identifying and predicting congenital heart disease are mainly the following: 1. Heart auscultation: Use a stethoscope to listen to the sound of the heart. Doctors can preliminarily judge possible heart problems by identifying abnormal heart sounds (such as murmurs). 2. Chest X-ray: Use an X-ray machine to image the chest and view the structure of the heart, blood vessels and surrounding lungs. 3. Electrocardiogram (ECG / EKG): Help doctors understand the health of the heart by detecting the electrical signals generated by the heart every time it beats. 4. Echocardiogram (Echo): Use ultrasound equipment to produce dynamic images of the heart and its surrounding structures. 5. Magnetic resonance imaging (MRI): Use strong magnetic fields and radio waves to produce detailed images of the internal structure of the body. 6. Computed tomography (CT): Use an X-ray machine to rotate around the patient and combine computer processing to generate detailed cross-sectional images of the heart. 7. Genetic testing: Collect samples (usually blood or saliva) from the patient and perform DNA analysis to look for genetic variations associated with congenital heart disease.

[0003] Each single method for identifying and predicting congenital heart disease has different shortcomings: cardiac auscultation can only provide a preliminary judgment, cannot confirm the diagnosis, and depends on the doctor's experience; chest X-rays have certain radiation risks, especially for children and pregnant women, and cannot provide detailed information about the internal structure of the heart; the information output by the electrocardiogram is limited, mainly providing information about electrical activity, and cannot show structural abnormalities. Abnormal electrocardiograms usually require further examination and confirmation; the results of echocardiography depend on the technical level of the operator, and in some cases, clear images may not be obtained (such as obese patients); magnetic resonance imaging has high examination costs and long examination times, and patients need to remain motionless for a long time, and some hospitals do not have MRI equipment; the cost of computer tomography is relatively high, and there is a radiation risk, especially for children and pregnant women. In addition, different fetal positions will affect the image quality; the cost of genetic testing is high, and the waiting period for testing is long. Moreover, the identification and prediction results also require professional interpretation.

[0004] In general, each recognition and prediction method has limitations: none of the recognition and prediction methods can guarantee the credibility of the recognition and prediction results, and it is usually necessary to combine multiple methods to comprehensively evaluate and recognize and predict congenital heart disease. How to effectively combine multiple recognition and prediction methods to obtain recognition and prediction results with shorter waiting time, more accurate results, lower costs, and no reliance on manual operations is an urgent problem to be solved. Summary of the invention

[0005] The present invention aims to solve the technical problem that a single method in the prior art cannot accurately predict congenital heart disease, and provides a congenital heart disease prediction method and system.

[0006] According to a first aspect of the present invention, a method for predicting congenital heart disease is provided, comprising: Obtain echocardiograms and genome sequences from patients; The patient's echocardiogram is input into the multi-task learning classification and segmentation model to obtain the first classification prediction result of the patient's congenital heart disease and the lesion area segmentation result; The patient's genomic sequence is input into a machine learning model based on the Bayesian algorithm to obtain a second classification prediction result of the patient's congenital heart disease; According to the first classification prediction result, the lesion area segmentation result and the second classification prediction result of the patient's congenital heart disease, the final classification prediction result and the lesion area segmentation result of the patient's congenital heart disease are obtained.

[0007] Based on the above technical solution, the present invention can also make the following improvements.

[0008] Optionally, the multi-task learning classification and segmentation module includes a data preprocessing module, a multi-task neural network module and a post-processing module, and the multi-task neural network module includes a feature extraction layer for feature sharing and two independent classification branch networks and segmentation branch networks; The data preprocessing module is used to preprocess the original echocardiogram of the patient. The preprocessing includes standardizing, removing noise and enhancing the image of the original echocardiogram of the patient. The feature extraction layer is used to extract the features of the preprocessed echocardiogram, and send the extracted echocardiogram features to the classification branch network and the segmentation branch network respectively; The classification branch network is used to predict the type of congenital heart disease of the patient according to the echocardiogram features by using a fully connected layer and a Softmax activation function to obtain a first classification prediction result; The segmentation branch network is used to segment the cardiac structure at the pixel level according to the echocardiographic features by using the U-Net architecture through continuous convolutional layers and upsampling layers, so as to locate and segment the lesion area of ​​the patient's congenital heart disease and obtain the lesion area segmentation result; The post-processing module is used to optimize the lesion region segmentation result, improve the segmentation result using morphological operations, and correct the segmentation error through connected region analysis.

[0009] Optionally, the training process of the multi-task neural network module includes: Acquire a first training sample data set, wherein the first training sample data set includes a plurality of first training samples, each of which includes an echocardiogram and a disease classification label and a cardiac structure segmentation label; Use the Adam optimizer to train the multi-task neural network module and set appropriate learning rate decay and early stopping strategies; Using transfer learning technology, the multi-task neural network module is first pre-trained on a similar medical image dataset, and then fine-tuned on the target echocardiography dataset to obtain the trained multi-task neural network module.

[0010] Optionally, the step of inputting the patient's genome sequence into a machine learning model based on a Bayesian algorithm to obtain a second classification prediction result of the patient's congenital heart disease includes: A multidimensional encoding strategy is used to convert the patient's genomic sequence into numerical data, and a dynamic programming algorithm is used to align the genomic sequence after numerical conversion; The aligned genome sequence is input into the Bayesian classifier. Based on the probability of nucleotide existence, the patient's gold and silver group sequences are classified by calculating the conditional probability of the gene sequence appearing in a given category to obtain the second classification prediction result.

[0011] Optionally, the multi-dimensional encoding strategy is used to convert the patient's genome sequence into numerical data, including: A multidimensional encoding strategy is adopted to assign different numerical characteristics to each nucleotide (A, T, C, G) according to the chemical properties and biological functions of the nucleotide sequence to obtain numerical data.

[0012] Optionally, the step of inputting the aligned genome sequence into a Bayesian classifier may include: Extracting features from the genome sequence to obtain multiple genome sequence features; One or more optimal genomic sequence features are selected from multiple genomic sequence features based on the recursive feature elimination method RFE and the minimum redundancy maximum relevance method mRMR.

[0013] Optionally, the training process of the Bayesian classifier includes: Acquire a second training sample data set, where the second training sample data set includes a plurality of second training samples, each of which includes an optimal genome sequence feature and a classification label of a genome sequence; The Bayesian classifier is trained based on the second training sample data set, and the parameters of the Bayesian classifier are iteratively updated using the Bayesian rule until a convergence condition is reached or a preset number of iterations is completed, thereby obtaining a trained Bayesian classifier.

[0014] Optionally, obtaining the final classification prediction result and lesion area segmentation result of the patient's congenital heart disease based on the first classification prediction result, the lesion area segmentation result and the second classification prediction result includes: When the first classification prediction result and the second classification prediction result are both normal, the final classification prediction result of the patient's congenital heart disease is normal; When one of the first classification prediction result and the second classification prediction result is abnormal, the final classification prediction result of the patient's congenital heart disease is abnormal.

[0015] According to a second aspect of the present invention, there is provided a congenital heart disease prediction system, comprising: An acquisition module for acquiring echocardiograms and genome sequences of patients; The first prediction module is used to input the patient's echocardiogram into the multi-task learning classification and segmentation model to obtain the first classification prediction result of the patient's congenital heart disease and the lesion area segmentation result; The second prediction module is used to input the patient's genome sequence into a machine learning model based on the Bayesian algorithm to obtain a second classification prediction result of the patient's congenital heart disease; The determination module is used to determine the final classification prediction result and lesion area segmentation result of the patient's congenital heart disease based on the first classification prediction result, the lesion area segmentation result and the second classification prediction result of the patient's congenital heart disease.

[0016] The present invention provides a congenital heart disease prediction method and system, which first uses deep learning technology to analyze and classify the patient's echocardiogram; then uses a machine learning method to regress and learn the patient's gene sequence; finally, combines the classification results of the echocardiogram and the classification results of the patient's gene sequence to accurately identify and predict the patient's congenital heart disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flow chart of a congenital heart disease prediction method provided by the present invention; Figure 2 Schematic diagram of the congenital heart disease prediction process; Figure 3 This is a schematic diagram of the structure of the multi-task learning classification and segmentation model; Figure 4 A block diagram of a congenital heart disease prediction system provided by the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in 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 addition, the technical features in the various embodiments or single embodiments provided by the present invention can be arbitrarily combined with each other to form a feasible technical solution. This combination is not subject to the constraints of the sequence of steps and / or the structural composition mode, but must be based on the ability of ordinary technicians in this field to achieve. When the combination of technical solutions is contradictory or cannot be achieved, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0019] Based on the defects in the background technology, the present invention provides a method for predicting congenital heart disease. Given the echocardiographic and genomic data of a patient, these data are processed from different research directions. For echocardiographic data, a 2D multi-task learning neural network model is used to perform qualitative classification of patients and quantitative measurement of lesion areas, and an identification prediction result is obtained based on the echocardiographic data. For genomic series data, a machine learning algorithm based on the Bayesian algorithm is used to perform binary classification on all gene sequences of different patients according to whether they are sick or not. Finally, the classification results of the two are combined to make accurate identification and prediction of congenital heart disease.

[0020] Figure 1 A flow chart of a congenital heart disease prediction method provided by the present invention, such as Figure 1 and Figure 2 As shown, the method includes: Step 1: Obtain the patient's echocardiogram and genome sequence.

[0021] Step 2: Input the patient's echocardiogram into the multi-task learning classification and segmentation model to obtain the first classification prediction result of the patient's congenital heart disease and the lesion area segmentation result.

[0022] Exemplarily, the multi-task learning classification and segmentation module includes a data preprocessing module, a multi-task neural network module and a post-processing module, and the multi-task neural network module includes a feature extraction layer for feature sharing and two independent classification branch networks and segmentation branch networks; The data preprocessing module is used to preprocess the original echocardiogram of the patient. The preprocessing includes standardizing, removing noise and enhancing the image of the original echocardiogram of the patient. The feature extraction layer is used to extract the features of the preprocessed echocardiogram, and send the extracted echocardiogram features to the classification branch network and the segmentation branch network respectively; The classification branch network is used to predict the type of congenital heart disease of the patient according to the echocardiogram features by using a fully connected layer and a Softmax activation function to obtain a first classification prediction result; The segmentation branch network is used to segment the cardiac structure at the pixel level according to the echocardiographic features by using the U-Net architecture through continuous convolutional layers and upsampling layers, so as to locate and segment the lesion area of ​​the patient's congenital heart disease and obtain the lesion area segmentation result; The post-processing module is used to optimize the lesion region segmentation result, improve the segmentation result using morphological operations, and correct the segmentation error through connected region analysis.

[0023] It can be understood that for the patient's echocardiogram, a multi-task learning classification and segmentation model is used to classify and segment it.

[0024] Among them, the multi-task learning classification and segmentation model is based on the pytorch deep learning framework, which mainly includes a data preprocessing module, a multi-task neural network module and a post-processing module.

[0025] Data preprocessing module: This module is responsible for extracting effective input information from the original echocardiogram, including image standardization, noise removal, and necessary image enhancement techniques: contrast adjustment and edge sharpening, to improve the accuracy of subsequent processing.

[0026] Multi-task neural network module: This module uses a deep convolutional neural network (CNN) architecture that combines two tasks: classification and segmentation. The specific structure is as follows Figure 3 As shown in the figure, the model includes a feature extraction layer for feature sharing (the feature extraction layer includes multiple dense blocks) and two independent branches (classification branch and segmentation branch). The shared feature extraction layer extracts features from the echocardiogram, and the extracted echocardiogram features are input into the classification branch and the segmentation branch respectively.

[0027] The classification branch classifies the features and obtains the classification prediction results of the patient's congenital heart disease. The segmentation branch quantitatively measures the lesion area and obtains the segmentation result of the lesion area.

[0028] Among them, the classification branch uses the fully connected layer and the Softmax activation function to judge the health of the heart or identify specific heart diseases.

[0029] The segmentation branch uses the U-Net architecture to perform pixel-level segmentation of the cardiac structure through continuous convolutional layers and upsampling layers. This branch uses the Dice coefficient as the loss function to optimize the accuracy of segmentation.

[0030] Exemplarily, the training process of the multi-task neural network module includes: Acquire a first training sample data set, wherein the first training sample data set includes a plurality of first training samples, each of which includes an echocardiogram and a disease classification label and a cardiac structure segmentation label; Use the Adam optimizer to train the multi-task neural network module and set appropriate learning rate decay and early stopping strategies; Using transfer learning technology, the multi-task neural network module is first pre-trained on a similar medical image dataset, and then fine-tuned on the target echocardiography dataset to obtain the trained multi-task neural network module.

[0031] When training the multi-task neural network model, the training dataset is first obtained: the model training uses a large-scale echocardiogram database annotated by heart disease experts. Each image includes a disease classification label and a corresponding heart structure segmentation label.

[0032] Training strategy: Use transfer learning techniques to pre-train the network on similar medical image datasets and then fine-tune it on the target echocardiography dataset. This helps the model show better generalization ability when facing real application data.

[0033] Optimization algorithm: Adam optimizer is used for model training, and appropriate learning rate decay and early stopping strategies are set to prevent overfitting.

[0034] The post-processing module is used to optimize the segmentation results inferred by the multi-task neural network module, including using morphological operations (such as corrosion and expansion) to improve the segmentation results and correcting small errors through connected region analysis.

[0035] Step 3: Input the patient's genome sequence into a machine learning model based on the Bayesian algorithm to obtain a second classification prediction result of the patient's congenital heart disease.

[0036] Exemplarily, the step of inputting the patient's genome sequence into a machine learning model based on a Bayesian algorithm to obtain a second classification prediction result of the patient's congenital heart disease includes: Step 31, using a multi-dimensional encoding strategy to convert the patient's genome sequence into numerical data, and using a dynamic programming algorithm to align the genome sequence after numerical conversion.

[0037] The patient's genomic sequence data is preprocessed and a multidimensional encoding strategy is used to convert the genomic sequence into numerical data. Specifically, each nucleotide (A, T, C, G) is assigned different numerical characteristics according to its chemical properties and biological functions, such as A=1, T=2, C=3, G=4. Such encoding not only reflects the primary structure of the sequence, but also facilitates subsequent mathematical processing.

[0038] Then sequence alignment is performed: the genomic sequences are aligned using a dynamic programming algorithm to eliminate sequence length variations caused by insertions, deletions, or substitutions. The aligned sequences will maintain consistent length and structure, which is the key to subsequent classification accuracy.

[0039] Step 32, input the aligned genome sequence into the Bayesian classifier, and classify the patient's genome sequence based on the probability of the existence of the nucleotide by calculating the conditional probability of the gene sequence appearing in a given category to obtain a second classification prediction result.

[0040] Among them, a model based on the naive Bayes classifier is constructed, which assumes that each feature is independent of each other. The model is based on the probability of the existence of nucleotides and classifies by calculating the conditional probability of the sequence appearing in a given category.

[0041] Exemplarily, after aligning the genome sequence, feature extraction is performed on the genome sequence to obtain multiple genome sequence features; one or more optimal genome sequence features are selected from the multiple genome sequence features based on the recursive feature elimination method RFE and the minimum redundancy maximum relevance method mRMR. Feature selection using recursive feature elimination (RFE) and the method based on minimum redundancy maximum relevance (mRMR) helps to identify the gene sequence features that best represent the target category, thereby improving the prediction accuracy of the model. Among them, the recursive feature elimination method RFE and the minimum redundancy maximum relevance method mRMR are conventional technical means and are not described in detail.

[0042] The selected optimal genome column features are input into the Bayesian classifier to obtain the classification prediction results.

[0043] The training process of the Bayesian classifier includes: Acquire a second training sample data set, where the second training sample data set includes a plurality of second training samples, each of which includes an optimal genome sequence feature and a classification label of a genome sequence; The Bayesian classifier is trained based on the second training sample data set, and the parameters of the Bayesian classifier are iteratively updated using the Bayesian rule until a convergence condition is reached or a preset number of iterations is completed, thereby obtaining a trained Bayesian classifier.

[0044] When training a Bayesian classifier, the classifier parameters are adjusted and optimized through the Bayesian estimation method. To process large data sets, the variational inference method is used to approximate the posterior probability, which can significantly reduce the consumption of computing resources and speed up the training of the Bayesian classifier.

[0045] During the Bayesian training process, the Bayesian rule is used to iteratively update the parameters until the convergence condition is reached or the preset number of iterations is completed. In addition, an adaptive learning rate is used to optimize the training process to ensure that the Bayesian classifier converges quickly in the early stages of training and remains stable when approaching the optimal solution. A ten-fold cross validation is implemented to evaluate the stability and generalization ability of the Bayesian classifier. Each validation randomly selects a portion of the data from the training set as the test set, and the rest is used to train the Bayesian classifier to estimate the performance of the Bayesian classifier on unseen data.

[0046] It should be noted that there are some alternative methods to the technical solution of using neural network to process cardiac electrocardiogram (ECG) and using Bayesian algorithm to process genomic data to identify and predict congenital heart disease, which can achieve similar invention purposes. These methods may adopt different technical frameworks or algorithms, including: 1. Support Vector Machine (SVM): Support vector machine is a powerful classification technique that can be used to classify cardiogram and genetic data. Compared with neural networks, SVM may perform better on some small sample data sets, and the model is easier to interpret.

[0047] 2. Random Forest: Random Forest is an ensemble learning method that improves the accuracy and stability of predictions by building multiple decision trees. This method can effectively process large amounts of biomedical data and provide intuitive explanations of the importance of features, helping to identify which genetic markers are associated with congenital heart disease.

[0048] 3. Other architectures of deep learning: In addition to traditional feedforward neural networks, convolutional neural networks (CNN) or recurrent neural networks (RNN) can also be used to process sequence data, such as ECG or DNA sequences. These networks are able to capture local features in time series data or sequence data.

[0049] 4. Ensemble learning methods: By combining the predictions of multiple machine learning models, the accuracy and robustness of recognition predictions can be improved. For example, the prediction results of neural networks and Bayesian models can be combined to finalize the recognition predictions through voting or weighting methods.

[0050] 5. Graph processing technology: For genomic data, graph theory methods can be used to model and analyze the interactions between genes. This method is particularly suitable for dealing with complex genetic networks and disease associations.

[0051] 6. Transfer learning and multi-task learning: These methods can train a model in one task (such as cardiogram analysis) and apply the learned knowledge to another task (such as genomic classification). This helps improve the performance of the model in tasks with less data and can be optimized simultaneously across multiple biomarkers.

[0052] Step 4, according to the first classification prediction result, the lesion area segmentation result and the second classification prediction result of the patient's congenital heart disease, the final classification prediction result and the lesion area segmentation result of the patient's congenital heart disease are obtained.

[0053] Among them, according to the patient's echocardiogram, the patient's congenital heart disease category is classified and the lesion area is measured based on the multi-task classification and segmentation model to obtain the first classification prediction result and the lesion area segmentation result. The patient's genome sequence is sorted based on the Bayesian classifier to obtain the second classification prediction result.

[0054] When both the first classification prediction result and the second classification prediction result are normal, the final classification prediction result of the patient's congenital heart disease is normal; when one of the first classification prediction result and the second classification prediction result is abnormal, the final classification prediction result of the patient's congenital heart disease is abnormal.

[0055] See also Figure 4 , a congenital heart disease prediction system provided by the present invention, the system comprising: An acquisition module 401 is used to acquire an echocardiogram and a genome sequence of a patient; The first prediction module 402 is used to input the patient's echocardiogram into the multi-task learning classification and segmentation model to obtain the first classification prediction result of the patient's congenital heart disease and the lesion area segmentation result; The second prediction module 403 is used to input the patient's genome sequence into a machine learning model based on a Bayesian algorithm to obtain a second classification prediction result of the patient's congenital heart disease; The determination module 404 is used to determine the final classification prediction result and the lesion area segmentation result of the patient's congenital heart disease according to the first classification prediction result, the lesion area segmentation result and the second classification prediction result of the patient's congenital heart disease.

[0056] It can be understood that the congenital heart disease prediction system provided by the present invention corresponds to the congenital heart disease prediction method provided in the aforementioned embodiments. The relevant technical features of the congenital heart disease prediction system can refer to the relevant technical features of the congenital heart disease prediction method, which will not be repeated here.

[0057] The congenital heart disease prediction method and system provided by the embodiment of the present invention have the following beneficial effects: 1. Improve prediction accuracy: By combining cardiac echocardiogram and genomic data, this technology can provide more comprehensive data support, making the identification and prediction of congenital heart disease more accurate. Neural networks can efficiently process and analyze large amounts of echocardiogram data, while Bayesian algorithms can add prior knowledge to genomic classification. The combination of the two can significantly improve the reliability of identification and prediction.

[0058] 2. Early detection and intervention: This technical solution can help doctors detect congenital heart disease in the early stages of the disease and intervene in time. This is very beneficial for improving patient prognosis and reducing treatment costs.

[0059] 3. Personalized treatment plan: By analyzing each patient's EKG characteristics and genetic information in detail, this technology can help doctors develop more personalized treatment plans. This approach can not only treat the patient's specific situation, but also avoid the use of general treatments that may be ineffective or harmful to some patients.

[0060] 4. Reduce medical costs: Accurate identification and prediction can reduce unnecessary medical examinations and trial-and-error treatment methods, thereby reducing overall medical costs. At the same time, early identification, prediction and treatment are usually more economical than dealing with advanced conditions.

[0061] 5. Promote the development of precision medicine: The implementation of this technical solution can not only improve the treatment effect of congenital heart disease, but also promote the entire medical industry to develop in a more precise and efficient direction. This will help optimize the allocation of medical resources and improve medical services.

[0062] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0063] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0064] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting congenital heart disease, characterized in that: include: Obtain echocardiograms and genome sequences from patients; The patient's echocardiogram is input into the multi-task learning classification and segmentation model to obtain the first classification prediction result of the patient's congenital heart disease and the lesion area segmentation result; The patient's genomic sequence was input into a machine learning model based on the Bayesian algorithm to obtain a second-classification prediction result of the patient's congenital heart disease; According to the first classification prediction result, the lesion area segmentation result and the second classification prediction result of the patient's congenital heart disease, the final classification prediction result and the lesion area segmentation result of the patient's congenital heart disease are obtained.

2. The congenital heart disease prediction method according to claim 1, characterized in that: The multi-task learning classification and segmentation module includes a data preprocessing module, a multi-task neural network module and a post-processing module, wherein the multi-task neural network module includes a feature extraction layer for feature sharing and two independent classification branch networks and segmentation branch networks; The data preprocessing module is used to preprocess the original echocardiogram of the patient. The preprocessing includes standardizing, removing noise and enhancing the image of the original echocardiogram of the patient. The feature extraction layer is used to extract the features of the preprocessed echocardiogram, and send the extracted echocardiogram features to the classification branch network and the segmentation branch network respectively; The classification branch network is used to predict the type of congenital heart disease of the patient according to the echocardiogram features by using a fully connected layer and a Softmax activation function to obtain a first classification prediction result; The segmentation branch network is used to segment the cardiac structure at the pixel level according to the echocardiographic features by using the U-Net architecture through continuous convolutional layers and upsampling layers, so as to locate and segment the lesion area of ​​the patient's congenital heart disease and obtain the lesion area segmentation result; The post-processing module is used to optimize the lesion region segmentation result, improve the segmentation result using morphological operations, and correct the segmentation error through connected region analysis.

3. The method for predicting congenital heart disease according to claim 2, characterized in that: The training process of the multi-task neural network module includes: Acquire a first training sample data set, wherein the first training sample data set includes a plurality of first training samples, each of which includes an echocardiogram and a disease classification label and a cardiac structure segmentation label; Use the Adam optimizer to train the multi-task neural network module and set appropriate learning rate decay and early stopping strategies; Using transfer learning technology, the multi-task neural network module is first pre-trained on a similar medical image dataset, and then fine-tuned on the target echocardiography dataset to obtain the trained multi-task neural network module.

4. The method for predicting congenital heart disease according to claim 1, characterized in that: The patient's genome sequence is input into a machine learning model based on a Bayesian algorithm to obtain a second classification prediction result of the patient's congenital heart disease, including: A multidimensional encoding strategy is used to convert the patient's genomic sequence into numerical data, and a dynamic programming algorithm is used to align the genomic sequence after numerical conversion; The aligned genome sequence is input into the Bayesian classifier. Based on the probability of nucleotide existence, the patient's gold and silver group sequences are classified by calculating the conditional probability of the gene sequence appearing in a given category to obtain the second classification prediction result.

5. The method for predicting congenital heart disease according to claim 4, characterized in that: The multi-dimensional encoding strategy is used to convert the patient's genome sequence into numerical data, including: A multidimensional encoding strategy is adopted to assign different numerical characteristics to each nucleotide (A, T, C, G) according to the chemical properties and biological functions of the nucleotide sequence to obtain numerical data.

6. The method for predicting congenital heart disease according to claim 4, characterized in that: The aligned genome sequence is input into the Bayesian classifier, which includes: Extracting features from the genome sequence to obtain multiple genome sequence features; One or more optimal genomic sequence features are selected from multiple genomic sequence features based on the recursive feature elimination method RFE and the minimum redundancy maximum relevance method mRMR.

7. The method for predicting congenital heart disease according to claim 6, characterized in that: The training process of the Bayesian classifier includes: Acquire a second training sample data set, where the second training sample data set includes a plurality of second training samples, each of which includes an optimal genome sequence feature and a classification label of a genome sequence; The Bayesian classifier is trained based on the second training sample data set, and the parameters of the Bayesian classifier are iteratively updated using the Bayesian rule until a convergence condition is reached or a preset number of iterations is completed, thereby obtaining a trained Bayesian classifier.

8. The method for predicting congenital heart disease according to claim 1, characterized in that: The method of obtaining the final classification prediction result and the lesion area segmentation result of the patient's congenital heart disease based on the first classification prediction result, the lesion area segmentation result and the second classification prediction result of the patient's congenital heart disease comprises: When the first classification prediction result and the second classification prediction result are both normal, the final classification prediction result of the patient's congenital heart disease is normal; When one of the first classification prediction result and the second classification prediction result is abnormal, the final classification prediction result of the patient's congenital heart disease is abnormal.

9. A congenital heart disease prediction system, characterized in that: include: An acquisition module for acquiring echocardiograms and genome sequences of patients; The first prediction module is used to input the patient's echocardiogram into the multi-task learning classification and segmentation model to obtain the first classification prediction result of the patient's congenital heart disease and the lesion area segmentation result; The second prediction module is used to input the patient's genome sequence into a machine learning model based on the Bayesian algorithm to obtain a second classification prediction result of the patient's congenital heart disease; The determination module is used to determine the final classification prediction result and lesion area segmentation result of the patient's congenital heart disease based on the first classification prediction result, the lesion area segmentation result and the second classification prediction result of the patient's congenital heart disease.

Images