Atrial fibrillation prediction method and system based on deep learning and multi-gene score

Abstract

The application provides an atrial fibrillation prediction method based on deep learning and multi-gene scoring, which combines electrocardiogram (ECG) signals, clinical feature data and multi-gene risk score (PRS) based on multi-modal data fusion. By integrating deep learning models, clinical risk scores and multi-gene risk scores, the accuracy and stability of AF prediction are significantly improved. Its technical value lies in: multi-source data fusion: for the first time, ECG signals, clinical data and genetic information are combined in a single system to provide comprehensive individualized risk assessment; real-time analysis capability: based on the real-time data processing capability of ECG, it can be used for dynamic risk assessment; clinical decision support: providing scientific and quantitative AF prediction reports for medical staff to assist in developing intervention strategies.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and in particular to a method for predicting atrial fibrillation based on deep learning and polygenic scoring, an atrial fibrillation prediction system based on deep learning and polygenic scoring, an electronic device, and a computer-readable storage medium. Background Art

[0002] Atrial fibrillation (AF) is a common cardiac arrhythmia that affects millions of patients worldwide and has a high morbidity and disability rate. Early detection and intervention of AF are important means to reduce the occurrence of complications (such as stroke and heart failure). Traditional AF symptom detection methods, such as the CHARGE-AF scoring system, can quickly detect AF symptoms based on the patient's clinical indicators (such as heart rate); or predict AF based on genetic data. However, current prediction methods have the following limitations:

[0003] 1. Limitations of traditional risk scores: Scoring systems such as CHARGE-AF rely on limited clinical indicators, making it difficult to capture complex individual characteristics and resulting in low predictive accuracy.

[0004] 2. The single nature of the genetic risk score (PRS): Predictions based solely on genetic data fail to fully consider environmental and physiological factors and have limited applicability;

[0005] 3. Isolated application of deep learning models: Existing deep learning models mostly focus on ECG data but lack the ability to integrate and analyze multi-source data. Summary of the Invention

[0006] In order to solve the technical problems existing in the prior art, the present invention provides the following technical solutions:

[0007] In one aspect, a method for predicting atrial fibrillation based on deep learning and polygenic scoring is provided, the method being implemented by an electronic device and comprising:

[0008] S1. Collect clinical data of AF patients: ECG data, clinical data, and genetic data;

[0009] S2. Invoking a pre-deployed AF prediction model to identify multimodal data features in the clinical data of the AF patient, and predicting an AF risk label corresponding to the multimodal data features;

[0010] S3. Writing the clinical data and the AF risk score in the AF risk tag into the electronic medical record of the AF patient to generate an AF risk score report.

[0011] Preferably, the method for generating the AF prediction model includes:

[0012] Construct a multimodal dataset including ECG data, clinical data, and genetic data collected from several AF patients at their historical diagnosis.

[0013] Extracting multimodal data features from the multimodal dataset includes:

[0014] extracting an ECG signal diagnosed as AF from the ECG data; and

[0015] Extracting basic information of AF patients, medical history information of risk factors affecting AF, drug information of clinical medications used for AF, and clinical examination results when AF symptoms appeared from the clinical data; and

[0016] Extracting SNP (single nucleotide polymorphism) information related to atrial fibrillation from the genetic data;

[0017] Feature annotation of multimodal data features of each AF patient, including:

[0018] Annotating ECG signals diagnosed as AF as AF-positive and ECG signals not diagnosed as AF as AF-negative; and

[0019] Labeling of the clinical data: for discrete data, labeling according to the corresponding specific values; for categorical data, labeling according to the classification results; otherwise, labeling as "missing"; and

[0020] Annotation of the SNP (single nucleotide polymorphism) information: using the PRS (polygenic risk score) rule, weighted calculation of the corresponding PRS value, and annotation of the corresponding risk group according to the PRS value;

[0021] fusing the feature annotation results of the ECG data, clinical data, and genetic data of the AF patient to generate a final AF risk label, and adding it to the multimodal data features of each AF patient;

[0022] Counting the multimodal data features of each AF patient with the AF risk label added, forming a data set for training the AF prediction model, and dividing the data set into a training set, a validation set, and a test set according to a preset ratio;

[0023] Using the training set as input to the multimodal training model, performing model training according to a preset training strategy to generate the AF prediction model;

[0024] The prediction performance of the AF prediction model was verified using the validation set:

[0025] If the verification passes, the processing performance of the AF prediction model is tested using the test set: if the test passes, the AF prediction model is deployed and applied, otherwise the training strategy is adjusted;

[0026] If the verification fails, the multimodal dataset is reorganized and retrained.

[0027] Preferably, when extracting the basic information of the AF patient from the clinical data, the method further includes:

[0028] The dimensional data in the clinical data are standardized and transformed by z-score standardization processing to obtain corresponding dimensionless data.

[0029] Preferably, when marking the SNP (single nucleotide polymorphism) information, it also includes:

[0030] Based on known AF-related genes, the status of SNP (single nucleotide polymorphism) information related to atrial fibrillation in the gene data of AF patients is annotated, including:

[0031] The mark “0” indicates no mutation;

[0032] The label “1” indicates heterozygote;

[0033] The label "2" indicates homozygote.

[0034] Preferably, the weighted calculation of the corresponding PRS value using the PRS (polygenic risk score) rule includes:

[0035] make:

[0036] ,

[0037] in:

[0038] The number of selected AF-related gene markers;

[0039] is the weight of the i-th genetic marker, determined based on the strength of the association between the marker and AF risk (effect size or statistical significance); determined through genome-wide association studies (GWAS) or similar genetic studies, which reveal the strength of the association between a specific genetic marker and AF risk. The weight reflects the contribution of the marker to AF risk, usually measured by effect size (logarithm of odds ratio) or statistical significance;

[0040] Count the genotype or allele of the i-th gene marker in the patient: 0, 1, or 2, indicating the number of patients carrying the risk allele.

[0041] Preferably, the Calculation method:

[0042] make:

[0043]

[0044] in:

[0045] is the odds ratio of the i-th gene marker, indicating the strength of association between the marker and AF risk;

[0046] is the P value of the i-th gene marker, indicating the statistical significance of the association;

[0047] is the logarithmic transformation of odds ratio, used to measure effect size;

[0048] It is the negative logarithmic transformation of the P value, which is used to measure statistical significance. Generally, the smaller the P value, the larger the value after negative logarithmic transformation, indicating higher significance.

[0049] is the sign of the logarithm of the odds ratio, that is, the direction of the association (positive or negative);

[0050] is the absolute value of the logarithm of the odds ratio, which is used to ensure a positive contribution to the effect size.

[0051] On the other hand, a system for predicting atrial fibrillation based on deep learning and polygenic scoring is provided. The system for predicting atrial fibrillation based on deep learning and polygenic scoring is used to implement the above-mentioned method for predicting atrial fibrillation based on deep learning and polygenic scoring. The system comprises:

[0052] Data acquisition module, used to collect clinical data of AF patients: ECG data, clinical data and genetic data;

[0053] an AF prediction module, configured to call a pre-deployed AF prediction model, identify multimodal data features in the clinical data of AF patients, and predict an AF risk signature corresponding to the multimodal data features;

[0054] The risk management module is used to write the clinical data and the AF risk score in the AF risk tag into the electronic medical record of the AF patient and generate an AF risk score report.

[0055] On the other hand, an electronic device is provided, comprising: a processor; and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the above-mentioned methods for predicting atrial fibrillation based on deep learning and polygenic scoring is implemented.

[0056] In another aspect, a computer-readable storage medium is provided, the storage medium having stored therein at least one instruction, the at least one instruction being loaded and executed by a processor to implement any one of the above atrial fibrillation prediction methods based on deep learning and multi-gene scoring.

[0057] The technical scheme provided by the embodiment of the present application brings at least the following beneficial effects:

[0058] The present application integrates deep learning model, clinical risk score and multi-gene risk score, and combines electrocardiogram (ECG) signal, clinical feature data and multi-gene risk score (PRS) based on multi-modal data fusion. The deep learning algorithm is used to extract features from time series data (ECG), and statistical methods are used to process clinical features and gene data. Finally, an individualized AF risk score is generated through a fusion model. Therefore, the accuracy and stability of AF prediction can be significantly improved. The technical value is reflected in:

[0059] 1. Multi-source data fusion: for the first time, ECG signal, clinical data and gene information are combined in a single system to provide comprehensive individualized risk assessment.

[0060] 2. Real-time analysis capability: based on real-time ECG data processing capability, it can be used for dynamic risk assessment.

[0061] 3. Clinical decision support: provides scientific and quantitative AF prediction reports for medical staff to assist in developing intervention strategies. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0063] Figure 1 is a flow chart of an atrial fibrillation prediction method based on deep learning and multi-gene scoring provided by an embodiment of the present application;

[0064] Figure 2 is a training process schematic diagram of an AF prediction model provided by an embodiment of the present application;

[0065] Figure 3 is a block diagram of an atrial fibrillation prediction system based on deep learning and multi-gene scoring provided by an embodiment of the present application;

[0066] Figure 4 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0067] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0068] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0069] In the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same. The terms "of," "corresponding," and "corresponding" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same.

[0070] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0071] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0072] The embodiment of the present invention provides a method for predicting atrial fibrillation based on deep learning and polygenic scoring. The method can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The flowchart of the atrial fibrillation prediction method based on deep learning and polygenic scoring is shown. The processing flow of the method may include the following steps:

[0073] S1. Collect clinical data of AF patients: ECG data, clinical data, and genetic data;

[0074] S2. Invoking a pre-deployed AF prediction model to identify multimodal data features in the clinical data of the AF patient, and predicting an AF risk label corresponding to the multimodal data features;

[0075] S3. Writing the clinical data and the AF risk score in the AF risk tag into the electronic medical record of the AF patient to generate an AF risk score report.

[0076] The present invention aims to:

[0077] 1. Improve the accuracy of AF prediction and reduce false positives and false negatives.

[0078] 2. Build a novel AF prediction system that can process and learn in real time to achieve efficient fusion analysis of multi-source data.

[0079] 3. Provide a flexible prediction verification mechanism to ensure the robustness of the prediction model in different data scenarios.

[0080] In practice, the approach combines electrocardiogram (ECG) signals, clinical features, and a polygenic risk score (PRS) based on multimodal data fusion. Deep learning algorithms extract features from the time series data (ECG), while statistical methods are used to process clinical features and genetic data. Finally, a fusion model is used to generate a personalized AF risk score.

[0081] The system's collection of relevant AF data from AF patients can be implemented in conjunction with the subsequent description. The HIS system can build and manage electronic medical records of AF patients and generate AF risk score reports in a preset medical report format.

[0082] The following mainly describes the model construction.

[0083] like Figure 2 As shown, preferably, the method for generating the AF prediction model includes:

[0084] Construct a multimodal dataset including ECG data, clinical data, and genetic data collected from several AF patients at their historical diagnosis.

[0085] Extracting multimodal data features from the multimodal dataset includes:

[0086] extracting an ECG signal diagnosed as AF from the ECG data; and

[0087] Extracting basic information of AF patients, medical history information of risk factors affecting AF, drug information of clinical medications used for AF, and clinical examination results when AF symptoms appeared from the clinical data; and

[0088] Extracting SNP (single nucleotide polymorphism) information related to atrial fibrillation from the genetic data;

[0089] Feature annotation of multimodal data features of each AF patient, including:

[0090] Annotating ECG signals diagnosed as AF as AF-positive and ECG signals not diagnosed as AF as AF-negative; and

[0091] Labeling of the clinical data: for discrete data, labeling according to the corresponding specific values; for categorical data, labeling according to the classification results; otherwise, labeling as "missing"; and

[0092] Annotation of the SNP (single nucleotide polymorphism) information: using the PRS (polygenic risk score) rule, weighted calculation of the corresponding PRS value, and annotation of the corresponding risk group according to the PRS value;

[0093] fusing the feature annotation results of the ECG data, clinical data, and genetic data of the AF patient to generate a final AF risk label, and adding it to the multimodal data features of each AF patient;

[0094] Counting the multimodal data features of each AF patient with the AF risk label added, forming a data set for training the AF prediction model, and dividing the data set into a training set, a validation set, and a test set according to a preset ratio;

[0095] Using the training set as input to the multimodal training model, performing model training according to a preset training strategy to generate the AF prediction model;

[0096] The prediction performance of the AF prediction model was verified using the validation set:

[0097] If the verification passes, the processing performance of the AF prediction model is tested using the test set: if the test passes, the AF prediction model is deployed and applied, otherwise the training strategy is adjusted;

[0098] If the verification fails, the multimodal dataset is reorganized and retrained.

[0099] The specific construction and application principles of the model are as follows:

[0100] 1. Technical Principle

[0101] This system, based on multimodal data fusion, combines electrocardiogram (ECG) signals, clinical features, and a polygenic risk score (PRS). A deep learning algorithm extracts features from the time series data (ECG), while a combination of statistical methods processes the clinical features and genetic data. Finally, a fusion model generates a personalized AF risk score.

[0102] 2. Dataset Construction

[0103] 1) ECG data collection: Use wearable devices or medical-grade electrocardiogram equipment to collect 12-lead ECG data. The data samples cover healthy people and diagnosed AF patients.

[0104] 2) Clinical data collection: including patient age, gender, BMI, whether or not they have hypertension, diabetes, and other clinical indicators.

[0105] 3) Genetic data acquisition: Gene sequencing technology is used to obtain the patient's multi-gene locus information, and the PRS calculation method published in the literature is used to generate a risk score.

[0106] 4) Data annotation: Cardiologists annotate ECG data to identify AF-positive or -negative samples, and genetic and clinical data are labeled according to known risks.

[0107] 5) Dataset division: In order to evaluate the generalization ability of the model, the dataset needs to be divided into training set, validation set and test set. For example:

[0108] Training set: 70% of the dataset is used for model training. The training set data undergoes data augmentation (such as time shifting and amplitude scaling).

[0109] Validation set: 15% of the dataset is used for model hyperparameter tuning and early stopping mechanism to avoid overfitting.

[0110] Test set: 15% of the dataset is used for final performance evaluation. The test set is data outside the model training process to ensure the objectivity of the evaluation results.

[0111] 3. Labeling scheme

[0112] In this paper, the annotation scheme focuses on the processing of electrocardiogram (ECG) data, clinical data, and genetic data. To ensure the accuracy and reliability of the prediction model, the annotation process must be rigorous and scientific. The following is a detailed description of the annotation scheme for each type of data:

[0113] 1) Electrocardiogram (ECG) data annotation

[0114] ECG signal acquisition and preprocessing:

[0115] Acquisition method: Use a high-quality 12-lead ECG instrument (such as a Holter monitor) or a wearable device (such as a smartwatch or chest strap) to ensure that the collected data is free of obvious artifacts or noise.

[0116] Preprocessing: The collected ECG signals are denoised, normalized, and segmented. Denoising involves using a low-pass filter to remove high-frequency noise, and normalization ensures that each ECG signal segment has a uniform amplitude range and time scale.

[0117] Marking standards:

[0118] Expert Diagnosis: ECG data is annotated by a board-certified electrocardiographer, who uses diagnostic criteria to determine if atrial fibrillation (AF) is present. Typically, signs of AF include irregular heart rhythms and rapid atrial rate.

[0119] Annotation Process: Each ECG signal is annotated as one of the following two categories:

[0120] AF Positive (1): If the ECG signal's diagnosis result is atrial fibrillation.

[0121] AF Negative (0): If the ECG signal's diagnosis result is normal or other arrhythmias (such as atrial premature beats, ventricular premature beats, etc.).

[0122] Annotation Format: The annotation data is annotated in time segments, for example, the ECG signal of each minute is taken as an annotation unit, and is annotated as "1" or "0".

[0123] Quality Control of Annotation:

[0124] Double Review Mechanism: Each ECG annotation result is double-checked by two independent electrocardiogram experts, who independently review and discuss to ensure the accuracy of the annotation results.

[0125] Uncertainty Annotation: If the experts have uncertainty about whether AF occurs, the "undetermined" (-1) label can be used, and further analysis can be performed.

[0126] 2) Clinical Data Annotation

[0127] Clinical data mainly includes patients' basic information, medical history, physical data, etc. In order to ensure the effectiveness of clinical data on the model, the annotation scheme requires the following data to be processed in a unified format and standardized:

[0128] Data Type and Content:

[0129] Basic Information: Age, gender, height, weight, BMI.

[0130] Medical History Information: Known risk factors for atrial fibrillation such as hypertension, diabetes, high cholesterol, etc.

[0131] Medication Information: Whether to take anticoagulant drugs, heart drugs, etc., and the length of use.

[0132] Clinical Examination Results: Such as electrocardiogram, echocardiogram, blood pressure, heart rate, blood sugar, etc.

[0133] Annotation Method:

[0134] Discrete Data: For numerical data such as age, BMI, blood pressure, etc., input the specific numerical value to ensure that no data is lost. For categorical data (such as whether or not suffering from hypertension), annotate as "yes" or "no".

[0135] Missing data processing: For missing clinical data, reasonable interpolation methods (such as mean filling, front and back interpolation, etc.) are used to fill them, or they are marked as "missing" and specially processed in subsequent training.

[0136] Data normalization and standardization:

[0137] All clinical characteristic data are standardized so that they are processed on the same scale. For example, characteristics with different dimensions such as height, weight, and blood sugar are converted into dimensionless data through z-score standardization.

[0138] 3) Genetic Data Annotation

[0139] Genetic data is the third type of input data for this invention, primarily the polygenic risk score (PRS). The annotation scheme for this part includes the following:

[0140] Gene data collection and preprocessing:

[0141] Collection method: Genetic data are obtained through high-throughput genome sequencing technology, mainly collecting SNP (single nucleotide polymorphism) information related to atrial fibrillation.

[0142] Data normalization: Normalize the collected genetic data to ensure that each SNP value is analyzed under the same standard. Generally, "0" (no mutation), "1" (heterozygous), and "2" (homozygous) are used as labels.

[0143] Marking standards:

[0144] Polygenic Risk Score (PRS): A weighted calculation is performed to generate a PRS for each patient based on known AF-related genetic markers. The higher the PRS value, the greater the patient's genetic risk for AF.

[0145] Notation: The generated PRS values ​​range from 0 to 1, with higher values ​​indicating a higher risk of AF. Statistical analysis is used to determine risk stratification, for example, categorizing PRS values ​​into low-risk (0-0.3), intermediate-risk (0.3-0.7), and high-risk (0.7-1.0) groups.

[0146] Supplementary information on genetic data:

[0147] During the annotation process, genetic data may contain some rare variants or unclear genotypes. In this case, you can use the "missing" annotation feature and use this feature to handle missing data in subsequent training.

[0148] 4 Comprehensive annotation

[0149] Annotation of fused data: For each patient, the annotations of their ECG data, clinical data, and genetic data will be merged together to form the final comprehensive dataset.

[0150] Comprehensive labeling results: Each sample will generate a final label based on the ECG signal diagnosis (AF positive or negative), clinical data markers (such as age, medical history), and genetic risk score (PRS). This label is used to train a deep learning model to predict the patient's AF risk.

[0151] 5. Model Training

[0152] 1) Model architecture: A multi-layer fusion network structure (selected by the user):

[0153] ECG feature extraction module: Based on the CNN-LSTM network, it extracts spatiotemporal features.

[0154] Clinical feature module: uses a fully connected network to classify and perform risk analysis on structured data.

[0155] Genetic risk scoring module: PRS values ​​are weighted using a linear regression model.

[0156] 2) Training strategy:

[0157] Data augmentation: ECG signals are processed by time shifting, amplitude scaling, etc. to expand the sample size.

[0158] Loss function: Use weighted cross entropy loss to balance positive and negative samples.

[0159] Optimization algorithm: Adam optimizer, the initial learning rate is set to 0.001.

[0160] 6. Algorithm performance requirements

[0161] The model prediction accuracy must reach ≥90%.

[0162] The recall rate must be ≥85% to ensure effective identification of high-risk patients.

[0163] The AUC of the ROC curve must be ≥0.95.

[0164] 7. Algorithm performance verification method

[0165] 1) Cross-validation: K-fold cross-validation is used to evaluate the generalization performance of the model.

[0166] 2) Independent test set validation: Use an independent external dataset to test the stability of the model.

[0167] 3) Clinical simulation experiment: Verify the reliability of model prediction results through retrospective data and compare with traditional CHARGE-AF score.

[0168] 4) Real-time performance test: Verify the real-time processing capability and prediction accuracy of the model in real environment.

[0169] In this embodiment, the multi-layer fusion network structure for multi-modal feature training and learning mainly includes Type-A: Standard Cross-Attention based Deep Fusion (SCDF), Type-B: Custom Layer based Deep Fusion (CLDF), and Type-C and other non-standardized structures.

[0170] For example:

[0171] ‌Type-A: Standard Cross-Attention based Deep Fusion (SCDF)‌

[0172] This structure is based on the standard cross-attention mechanism for deep fusion. In multi-modal tasks, information from different modalities is interacted and fused through cross-attention mechanism, so as to extract more rich feature representation. This structure performs well in processing multi-modal data such as image and text, and can effectively capture the associated information between different modalities‌.

[0173] ‌Type-B: Custom Layer based Deep Fusion (CLDF)‌

[0174] CLDF structure relies on custom fusion layers for deep fusion of multi-modal features. These custom layers can be designed according to specific task requirements to better capture and utilize the complementary information between multi-modal data. By introducing these custom layers, the network can learn more complex and fine multi-modal feature representations‌.

[0175] Preferably, when extracting the basic information of the AF patient in the clinical data, it further includes:

[0176] Through z-score standardization processing, the dimensional data in the clinical data is standardized and converted to obtain corresponding dimensionless data.

[0177] Z-score normalization is a commonly used data preprocessing method whose main purpose is to convert data with different dimensions or units to a unified scale for easy comparison and analysis. Z-score normalization only requires calculating the mean and standard deviation and is applicable to various types of data. The calculation process is simple and efficient, allowing analysts to quickly preprocess data. ‌Eliminating Dimensionality‌: Through normalization, unit and magnitude differences between different features can be eliminated, allowing data to be compared and analyzed on the same scale.

[0178] Preferably, when marking the SNP (single nucleotide polymorphism) information, it also includes:

[0179] Based on known AF-related genes, the status of SNP (single nucleotide polymorphism) information related to atrial fibrillation in the gene data of AF patients is annotated, including:

[0180] The mark “0” indicates no mutation;

[0181] The label “1” indicates heterozygote;

[0182] The label "2" indicates homozygote.

[0183] For specific annotations, see the above training steps.

[0184] Preferably, the weighted calculation of the corresponding PRS value using the PRS (polygenic risk score) rule includes:

[0185] make:

[0186] ,

[0187] in:

[0188] The number of selected AF-related gene markers;

[0189] is the weight of the i-th genetic marker, determined based on the strength of the association between the marker and AF risk (effect size or statistical significance); determined through genome-wide association studies (GWAS) or similar genetic studies, which reveal the strength of the association between a specific genetic marker and AF risk. The weight reflects the contribution of the marker to AF risk, usually measured by effect size (logarithm of odds ratio) or statistical significance;

[0190] Count the genotype or allele of the i-th gene marker in the patient: 0, 1, or 2, indicating the number of patients carrying the risk allele.

[0191] The above weighted calculation can bring the following technical effects:

[0192] 1. Personalized risk assessment: By calculating the patient's PRS value, their personalized genetic risk of AF can be assessed, which is helpful for clinical decision-making and early intervention.

[0193] 2. Stratified management: According to the PRS value, patients can be divided into different risk strata, so that more targeted prevention and treatment strategies can be implemented.

[0194] 3. Scientific research: PRS values ​​can be used as research variables to explore the association between AF and other diseases or biomarkers, as well as to study the role of genetic factors in the development of AF.

[0195] By integrating information from multiple genetic markers, the PRS value provides a more comprehensive genetic risk assessment than a single genetic marker.

[0196] Preferably, the Calculation method:

[0197] make:

[0198] ,

[0199] in:

[0200] is the odds ratio of the i-th gene marker, indicating the strength of association between the marker and AF risk;

[0201] is the P value of the i-th gene marker, indicating the statistical significance of the association;

[0202] is the logarithmic transformation of odds ratio, used to measure effect size;

[0203] It is the negative logarithmic transformation of the P value, which is used to measure statistical significance. Generally, the smaller the P value, the larger the value after negative logarithmic transformation, indicating higher significance.

[0204] is the sign of the logarithm of the odds ratio, that is, the direction of the association (positive or negative);

[0205] is the absolute value of the logarithm of the odds ratio, which is used to ensure a positive contribution to the effect size.

[0206] When calculating specifically:

[0207] Directionality: Pass The direction of the association is determined, with an odds ratio greater than 1 indicating that the marker increases the risk of AF (positive direction) and less than 1 indicating that it decreases the risk of AF (negative direction).

[0208] ‌Statistical significance: by The weights are adjusted to ensure that statistically more significant associations contribute more to the PRS calculation.

[0209] ‌Effect size: by Reflects the actual effect size of each marker on the risk of AF.

[0210] In practical applications, some preprocessing of the odds ratio and P-value may be required, such as truncating very small P-values to avoid excessive weights. The The formula for calculating the weight can be adjusted according to the specific research purpose and data characteristics, such as considering the use of other statistical quantities or transformation methods. The final calculated PRS value should be based on scientific research and verification to ensure its accuracy and reliability in practical applications.

[0211] Figure 3 is a block diagram of an atrial fibrillation prediction system based on deep learning and multi-gene score according to an exemplary embodiment, which is used for an atrial fibrillation prediction method based on deep learning and multi-gene score. The system comprises:

[0212] A data acquisition module for acquiring clinical data of AF patients: ECG data, clinical data and genetic data;

[0213] An AF prediction module for calling a pre-deployed AF prediction model, identifying multi-modal data features in the clinical data of AF patients, and predicting an AF risk label corresponding to the multi-modal data features;

[0214] A risk management module for writing the AF risk score in the clinical data and the AF risk label into the electronic medical record file of the AF patient, and generating an AF risk score report.

[0215] The functions and interaction steps of the above modules are understood and implemented in combination with the above method steps, and the present embodiment will not be described again.

[0216] Figure 4 is a structural schematic diagram of an electronic device provided by an embodiment of the present application, as shown in Figure 4 The electronic device can include the atrial fibrillation prediction system based on deep learning and multi-gene score shown in Figure 3 Optionally, the electronic device 410 can include a first processor 2001.

[0217] Optionally, the electronic device 410 may further include a memory 2002 and a transceiver 2003 .

[0218] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0219] The following combination Figure 4 The components of the electronic device 410 are described in detail.

[0220] The first processor 2001 is the control center of the electronic device 410 and can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs) or one or more field programmable gate arrays (FPGAs).

[0221] Optionally, the first processor 2001 can execute various functions of the electronic device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0222] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 4 CPU0 and CPU1 are shown in FIG.

[0223] In a specific implementation, as an embodiment, the electronic device 410 may also include multiple processors, such as Figure 4 1 and 2. The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). A processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0224] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled by the first processor 2001 for execution. The specific implementation method can refer to the above method embodiment and will not be repeated here.

[0225] Alternatively, the memory 2002 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including a compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001 or exist independently and accessed through the interface circuit ( Figure 4 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0226] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0227] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Figure 4 The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0228] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently and communicate with the first processor 2001 through the interface circuit ( Figure 4 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0229] It should be noted that Figure 4 The structure of the electronic device 410 shown in the figure does not constitute a limitation on the router. The actual knowledge structure recognition device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0230] In addition, the technical effects of the electronic device 410 can refer to the technical effects of the atrial fibrillation prediction method based on deep learning and polygenic scoring described in the above method embodiment, and will not be repeated here.

[0231] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, etc.

[0232] It should also be understood that the memory in the embodiments of the present invention may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0233] The above embodiments can be implemented in whole or in part via software, hardware (e.g., circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product comprises one or more computer instructions or computer programs. When loaded or executed on a computer, the processes or functions described in accordance with the embodiments of the present invention are fully or partially performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that contains a collection of one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. The semiconductor media can be a solid-state drive.

[0234] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0235] In this disclosure, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, "at least one of a, b, or c" can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0236] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0237] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0238] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the devices, systems and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0239] In several embodiments provided by the present application, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are merely schematic, for example, the division of the units is only a logical functional division, and actual implementation can have another division, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, systems or units, which can be electrical, mechanical or other forms.

[0240] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0241] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0242] If the functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or the portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical disks.

[0243] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for predicting atrial fibrillation based on deep learning and polygenic scoring, characterized in that: The method comprises: S1. Collect clinical data of AF patients: ECG data, clinical data, and genetic data; S2. Invoking a pre-deployed AF prediction model to identify multimodal data features in the clinical data of the AF patient, and predicting an AF risk label corresponding to the multimodal data features; The method for generating the AF prediction model includes: Construct a multimodal dataset including ECG data, clinical data, and genetic data collected from several AF patients at their historical diagnosis. Extracting multimodal data features from the multimodal dataset includes: extracting an ECG signal diagnosed as AF from the ECG data; and Extracting basic information of AF patients, medical history information of risk factors affecting AF, drug information of clinical medications used for AF, and clinical examination results when AF symptoms appeared from the clinical data; and Extracting SNP (single nucleotide polymorphism) information related to atrial fibrillation from the genetic data; Feature annotation of multimodal data features of each AF patient, including: Annotating ECG signals diagnosed as AF as AF-positive and ECG signals not diagnosed as AF as AF-negative; and Labeling of the clinical data: for discrete data, labeling according to the corresponding specific values; for categorical data, labeling according to the classification results; otherwise, labeling as "missing"; and Annotation of the SNP (single nucleotide polymorphism) information: using the PRS (polygenic risk score) rule, weighted calculation of the corresponding PRS value, and annotation of the corresponding risk group according to the PRS value; fusing the feature annotation results of the ECG data, clinical data, and genetic data of the AF patient to generate a final AF risk label, and adding the label to the multimodal data features of each AF patient; Counting the multimodal data features of each AF patient with the AF risk label added, forming a data set for training the AF prediction model, and dividing the data set into a training set, a validation set, and a test set according to a preset ratio; The training set is used as the input of the multimodal training model, and the model is trained according to a preset training strategy to generate the AF prediction model. The architecture of the multimodal training model includes: an ECG feature extraction module: extracting spatiotemporal features based on a CNN-LSTM network; a clinical feature module: using a fully connected network to classify and analyze structured data; and a genetic risk scoring module: using a linear regression model to weight the PRS value. The prediction performance of the AF prediction model was verified using the validation set: If the verification passes, the processing performance of the AF prediction model is tested using the test set: if the test passes, the AF prediction model is deployed and applied, otherwise the training strategy is adjusted; If the verification fails, the multimodal dataset is reorganized and retrained; S3. Writing the clinical data and the AF risk score in the AF risk tag into the electronic medical record of the AF patient to generate an AF risk score report.

2. The atrial fibrillation prediction method based on deep learning and polygenic scoring according to claim 1, characterized in that: When extracting the basic information of the AF patient from the clinical data, it also includes: The dimensional data in the clinical data are standardized and transformed by z-score standardization processing to obtain corresponding dimensionless data.

3. The atrial fibrillation prediction method based on deep learning and polygenic scoring according to claim 1, characterized in that: When annotating the SNP (single nucleotide polymorphism) information, it also includes: Based on known AF-related genes, the status of SNP (single nucleotide polymorphism) information related to atrial fibrillation in the gene data of AF patients is annotated, including: The mark "0" indicates no mutation; The label "1" indicates heterozygote; The mark "2" indicates homozygote.

4. The atrial fibrillation prediction method based on deep learning and polygenic scoring according to claim 3, characterized in that: The weighted calculation of the corresponding PRS value according to the PRS rule includes: make: ， in: The number of selected AF-related gene markers; is the weight of the i-th genetic marker, determined based on the strength of the association between the marker and AF risk; determined through genome-wide association studies (GWAS) or similar genetic studies, which reveal the strength of the association between a specific genetic marker and AF risk. The weight reflects the contribution of the marker to AF risk, which is usually measured by effect size or statistical significance. Count the genotype or allele of the i-th gene marker in the patient: 0, 1, or 2, indicating the number of patients carrying the risk allele.

5. The atrial fibrillation prediction method based on deep learning and polygenic scoring according to claim 4, characterized in that: described Calculation method: make: ， in: is the odds ratio of the i-th gene marker, indicating the strength of association between the marker and AF risk; is the P value of the i-th gene marker, indicating the statistical significance of the association; is the logarithmic transformation of odds ratio, used to measure effect size; It is the negative logarithmic transformation of the P value, which is used to measure statistical significance. Generally, the smaller the P value, the larger the value after negative logarithmic transformation, indicating higher significance. is the sign of the logarithm of the odds ratio, that is, the direction of the association; is the absolute value of the logarithm of the odds ratio, which is used to ensure a positive contribution to the effect size.

6. A deep learning and polygenic scoring-based atrial fibrillation prediction system, wherein the deep learning and polygenic scoring-based atrial fibrillation prediction system is used to implement the deep learning and polygenic scoring-based atrial fibrillation prediction method according to any one of claims 1 to 5, characterized in that: The system comprises: Data acquisition module, used to collect clinical data of AF patients: ECG data, clinical data and genetic data; an AF prediction module, configured to call a pre-deployed AF prediction model, identify multimodal data features in the clinical data of AF patients, and predict an AF risk signature corresponding to the multimodal data features; The risk management module is used to write the clinical data and the AF risk score in the AF risk tag into the electronic medical record of the AF patient and generate an AF risk score report.

7. An electronic device, characterized in that: The electronic device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the method according to any one of claims 1 to 5.

Images