CN-122025090-A - Heart failure classification auxiliary diagnosis system and method based on multi-feature fusion

CN122025090ACN 122025090 ACN122025090 ACN 122025090ACN-122025090-A

Abstract

The invention relates to the technical field of biomedical engineering, and discloses a heart failure classification auxiliary diagnosis system and a heart failure classification auxiliary diagnosis method based on multi-feature fusion, wherein the system comprises an electrocardio data preprocessing module, a data preprocessing module and a data processing module, wherein the electrocardio data preprocessing module is used for acquiring electrocardio data and demographic data of a patient and preprocessing the data; the multi-feature electrocardio data extraction module is used for extracting heart rate variability features and deep learning features from the preprocessed electrocardio data, wherein the heart rate variability features comprise time domain, frequency domain and nonlinear domain features; the heart failure recognition module is used for inputting the multi-modal fusion features into the classifier to perform heart failure three-class recognition to obtain heart failure classification results. The invention can obviously improve the accuracy of heart failure identification, promote early lesion discovery, assist clinical treatment and reduce the workload of doctors.

Inventors

WANG YANJING
REN ZHENFENG
ZHANG JUN

Assignees

苏州至心医疗科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (10)

1. A heart failure classification auxiliary diagnostic system based on multi-feature fusion, comprising: The electrocardio data preprocessing module is used for acquiring electrocardio data and demographic data of a patient and preprocessing the data; The multi-feature electrocardio data extraction module is used for extracting heart rate variability features and deep learning features from the preprocessed electrocardio data, wherein the heart rate variability features comprise time domain, frequency domain and nonlinear domain features; The attention mechanism fusion module is used for dynamically fusing heart rate variability characteristics, deep learning characteristics and patient demographic data through a multi-head attention mechanism to obtain multi-modal fusion characteristics; And the heart failure identification module is used for inputting the multi-mode fusion characteristics into the classifier to carry out heart failure three-classification identification so as to obtain a heart failure classification result.
2. The heart failure classification aid-diagnosis system based on multi-feature fusion according to claim 1, wherein the data preprocessing of the electrocardiographic data comprises: Resampling electrocardio data to 128 Hz frequency, adopting a Butterworth band-pass filter and a wavelet transformation multilevel denoising strategy to perform refined denoising treatment on electrocardiosignals, performing heart beat segmentation on the denoised electrocardio data, applying dynamic window adjustment to adapt to heart rate variability, and finally normalizing the electrocardio data.
3. The heart failure classification aid-diagnosis system based on multi-feature fusion according to claim 2, wherein the de-noised electrocardiographic data is subjected to heart beat segmentation and dynamic window adjustment is applied to adapt to heart rate variability, further comprising: Abnormal heart beat detection is carried out by adopting an outlier removing algorithm based on Mahalanobis distance.
4. The heart failure classification aid-diagnosis system based on multi-feature fusion of claim 1, wherein extracting heart rate variability features from the preprocessed electrocardiographic data comprises: And positioning an R peak of the electrocardiograph data by adopting a Pan-Tompkins algorithm, calculating an R-R interval, and extracting heart rate variability characteristics of the electrocardiograph data.
5. The heart failure classification aid diagnosis system based on multi-feature fusion according to claim 4, wherein the temporal features include mean, standard deviation, variance, skewness, kurtosis; the frequency domain features include total power, very low frequency power, high frequency power, very low frequency power normalization value, high frequency power normalization value, low high frequency power ratio, low frequency peak frequency and high frequency peak frequency; the nonlinear domain features include sample entropy, approximate entropy, fractal dimension, poincare graph index, detrending fluctuation analysis, multi-scale entropy and related dimension.
6. The heart failure classification aid-diagnosis system based on multi-feature fusion of claim 1, wherein extracting deep learning features from the preprocessed electrocardiographic data comprises: And converting the electrocardiographic data into a frequency domain diagram by adopting a constant Q conversion algorithm, and extracting multi-scale frequency spectrum characteristics of the frequency domain diagram by combining Vision Transformer models with ResNet.
7. The heart failure classification aid diagnosis system based on multi-feature fusion of claim 1, wherein the attention mechanism fusion module further comprises: and screening the deep learning features by adopting a recursive feature elimination algorithm, and reducing the dimension by combining a principal component analysis method.
8. The heart failure classification aid diagnosis system based on multi-feature fusion of claim 1, further comprising: And the model interpretation module is used for interpreting the heart failure classification result output by the heart failure identification module by using the SHAP method and outputting a characteristic importance thermodynamic diagram and a visual decision path.
9. The heart failure classification aid diagnosis system based on multi-feature fusion of claim 1, further comprising: The risk prediction module is used for predicting the heart failure progress risk by using the Cox proportion risk model based on the multimodal fusion characteristics output by the attention mechanism fusion module and the heart failure classification result output by the heart failure identification module, and outputting risk scores, kaplan-Meier survival curves and prevention suggestions.
10. A heart failure classification auxiliary diagnosis method based on multi-feature fusion, which is applied to the heart failure classification auxiliary diagnosis system based on multi-feature fusion as claimed in any one of claims 1 to 9, and is characterized by comprising the following steps: acquiring electrocardiographic data and demographic data of a patient by using an electrocardiographic data preprocessing module, and preprocessing the data; Extracting heart rate variability features and deep learning features from the preprocessed electrocardiographic data by utilizing a multi-feature electrocardiographic data extraction module, wherein the heart rate variability features comprise time domain, frequency domain and nonlinear domain features; dynamically fusing heart rate variability characteristics, deep learning characteristics and patient demographic data through a multi-head attention mechanism by using an attention mechanism fusion module to obtain multi-mode fusion characteristics; And inputting the multimodal fusion characteristics into a classifier by using a heart failure identification module to perform heart failure three-classification identification, so as to obtain a heart failure classification result.

Description

Heart failure classification auxiliary diagnosis system and method based on multi-feature fusion Technical Field The invention relates to the technical field of biomedical engineering, in particular to a heart failure classification auxiliary diagnosis system and method based on multi-feature fusion. Background According to World Health Organization (WHO) data, about 31% of deaths worldwide are shown to be associated with heart disease. Congestive Heart Failure (CHF) is a serious heart disease and is also a leading cause of high global mortality. It is counted that more than 2600 tens of thousands of adults worldwide have CHF, with up to 360 tens of thousands of new patients annually. However, CHF was found to be critical early in improving treatment options and preventing disease progression. Traditionally, accurate diagnosis of congestive heart failure has generally relied on Electrocardiogram (ECG) as the primary tool. However, existing approaches have significant limitations in multi-modal feature fusion. For Heart Rate Variability (HRV) features, it is difficult to fully capture time domain, frequency domain and nonlinear dynamic features, resulting in insufficient sensitivity to early heart rate variability anomalies (e.g., fine fluctuations due to autonomic dysfunction), and thus easy omission of potential lesion signals. For deep learning characteristics, when a one-dimensional ECG signal is directly input into a model, a spectrum mode and a multi-scale space structure are difficult to effectively reveal, so that the generalization capability of the model under a non-stationary signal is weak, and a potential deep abnormal mode is difficult to extract. Meanwhile, the existing model often ignores the influence of demographic characteristics (such as age, sex, body mass index and the like) on heart failure risk, so that diagnosis results are easily interfered by crowd difference, and particularly in old people or obese people, the risk assessment accuracy is low. These problems together limit the overall accuracy and robustness of conventional diagnostic systems, and are difficult to meet the requirements of early clinical interventions. Since the exact diagnosis usually relies on experienced cardiologists, which are time consuming and laborious, it is particularly necessary to develop computer-aided automatic detection and identification methods. Disclosure of Invention Aiming at the defects in the prior art, the invention provides a heart failure classification auxiliary diagnosis system and a heart failure classification auxiliary diagnosis method based on multi-feature fusion. The system firstly carries out HRV feature extraction on electrocardiosignals, including feature extraction of time domain, frequency domain and nonlinear domain. Meanwhile, a CQT algorithm is utilized to convert one-dimensional electrocardiosignals into a frequency domain map, a Vision Transformer model is combined to conduct deep learning feature extraction, and then HRV features, deep learning features and patient demographic information are fused, and information dynamic weighting is achieved through a multi-head attention mechanism. The fused multi-feature data are input into XGBoost classifiers, and model generalization performance is improved through regularization and residual iteration. Finally, the system classifies heart failure samples into three classes, and provides interpretable decision basis and individuation prognosis evaluation through a model interpretation and risk prediction module. The system can remarkably improve the accuracy of heart failure identification, promote early lesion discovery, assist clinical treatment and reduce the workload of doctors. In order to achieve the aim of the invention, the invention adopts the following technical scheme: In a first aspect, the present invention proposes a heart failure classification auxiliary diagnosis system based on multi-feature fusion, comprising: The electrocardio data preprocessing module is used for acquiring electrocardio data and demographic data of a patient and preprocessing the data; The multi-feature electrocardio data extraction module is used for extracting heart rate variability features and deep learning features from the preprocessed electrocardio data, wherein the heart rate variability features comprise time domain, frequency domain and nonlinear domain features; The attention mechanism fusion module is used for dynamically fusing heart rate variability characteristics, deep learning characteristics and patient demographic data through a multi-head attention mechanism to obtain multi-modal fusion characteristics; And the heart failure identification module is used for inputting the multi-mode fusion characteristics into the classifier to carry out heart failure three-classification identification so as to obtain a heart failure classification result. Preferably, the data preprocessing of the electrocardiographic data includes: Resamplin