CN-121997107-A - Classification method of non-contact electrocardiosignals at night

CN121997107ACN 121997107 ACN121997107 ACN 121997107ACN-121997107-A

Abstract

The application discloses a method for classifying night non-contact electrocardiosignals, which comprises the steps of acquiring night non-contact electrocardiosignal data, preprocessing the night non-contact electrocardiosignal data, segmenting the preprocessed night non-contact electrocardiosignal data, classifying according to the characteristics of each segment of night non-contact electrocardiosignal data to obtain night non-contact electrocardiosignal data with classification labels, classifying the night non-contact electrocardiosignal data with the classification labels into a training set, a verification set and a test set, taking the night non-contact electrocardiosignal data with the classification labels in the training set as the input characteristics of a deep learning classification model, taking the classification labels of the night non-contact electrocardiosignal data as the output characteristics of the deep learning classification model, and training the deep learning classification model. According to the method, the electrocardiosignals are classified by adopting a high-precision light deep learning model, so that the utilization rate of the electrocardiosignals is improved.

Inventors

ZHENG XIONGWEN
CHEN CHEN
CHEN WEI
CUI RUI
CHEN HONGYU
LI YANG

Assignees

粤港澳大湾区精准医学研究院(广州)
上海国际人类表型组研究院

Dates

Publication Date: 20260508
Application Date: 20241101

Claims (10)

1. The training method of the deep learning classification model for classifying the non-contact electrocardiosignals at night is characterized by comprising the following steps of: (a) Acquiring night non-contact electrocardiosignal data acquired based on a capacitive coupling principle; (b) Preprocessing the night non-contact electrocardiosignal data, segmenting the preprocessed night non-contact electrocardiosignal data, and classifying according to the characteristics of each segment of the night non-contact electrocardiosignal data so as to obtain night non-contact electrocardiosignal data with classification labels; (c) The night non-contact electrocardiosignal data with the classification labels are divided into a training set, a verification set and a test set; (d) Taking the night non-contact electrocardiosignal data with the classification labels in the training set as the input characteristics of a deep learning classification model, taking the classification labels of the night non-contact electrocardiosignal data as the output characteristics of the deep learning classification model, and training the deep learning classification model so as to obtain a trained deep learning classification model; the deep learning classification model comprises a feature extraction module and a classification module which are formed by a convolutional neural network, wherein the night non-contact electrocardiosignal data with the classification labels comprises one-dimensional data of electrocardiosignals and corresponding classification labels of each section of night non-contact electrocardiosignal data, and the abscissa of the one-dimensional data of the electrocardiosignals represents time and the ordinate represents signal amplitude.
2. The method of claim 1, wherein in step (b), the nighttime non-contact electrocardiographic signal data after segmentation is divided into four classes, the classification labels comprising C1, C2, C3, C4, wherein class C1 refers to clear electrocardiographic signals useful for subsequent arrhythmia diagnosis, class C2 refers to P-wave and T-wave masked by noise, only the QRS complex can clearly identify electrocardiographic signals useful for subsequent heart rate variability calculation, class C3 refers to electrocardiographic signals useful for subsequent body movement detection that exhibit irregular large fluctuations, class C4 refers to electrocardiographic signals useful for subsequent out-of-bed identification that exhibit weak noise or are displayed directly as horizontal lines.
3. The method of claim 1, wherein the feature extraction module comprises at least five convolution blocks, each of the convolution blocks having a same structure, each of the convolution blocks being composed of a one-dimensional convolution layer, an activation function Relu layer, and a one-dimensional maximally pooled downsampling layer in that order.
4. The method of claim 1, wherein the classification module comprises a random deactivation layer, a first fully-connected layer, an activation function Relu layer, and a second fully-connected layer.
5. The method of claim 1, further comprising the step of: (e) Inputting one-dimensional data of the electrocardiosignals in the verification set into the deep learning classification model, thereby obtaining a value of a main index of classification performance of the deep learning classification model; (f) Then, sequentially and iteratively executing the steps (d) - (e), thereby obtaining an optimal deep learning classification model; (g) Inputting one-dimensional data of electrocardiosignals in the test set into the optimal deep learning classification model obtained in the step (f), thereby obtaining the value of a main index of the classification performance of the optimal deep learning classification model.
6. A method for night non-contact electrocardiographic signal classification using the deep learning classification model of any one of claims 1-5, comprising the steps of: (a) Acquiring night non-contact electrocardiosignal data acquired based on a capacitive coupling principle; (b) Preprocessing the nocturnal non-contact electrocardiosignal data, and segmenting the preprocessed nocturnal non-contact electrocardiosignal data so as to obtain a plurality of segmented nocturnal non-contact electrocardiosignal data; (c) And respectively inputting the night non-contact electrocardiosignal data of each segment into the deep learning classification model, so as to obtain a classification label of the night non-contact electrocardiosignal data of each segment.
7. A system for classifying nocturnal non-contact electrocardiographic signals, comprising A data acquisition module for acquiring the night non-contact electrocardiosignal data of a patient based on a capacitive coupling principle, The first preprocessing module is used for filtering the non-contact electrocardiosignal data at night; The second preprocessing module is used for dividing the filtered night non-contact electrocardiosignal data once every 2s-5s so as to obtain a plurality of pieces of night non-contact electrocardiosignal data; the data feature classification module is provided with a trained deep learning classification model, the trained deep learning classification model is used for classifying the multi-section night non-contact electrocardiosignal data obtained through processing by the second preprocessing module and outputting a classification label, and the trained deep learning classification model is obtained through training by the method of claim 1.
8. A device for classifying nighttime non-contact electrocardiosignals, comprising: a memory for storing computer executable instructions, and A processor coupled to the memory for implementing the steps in the method of claim 6 when executing the computer-executable instructions.
9. A computer-readable storage medium comprising, the computer-readable storage medium has stored therein computer-executable instructions, the computer executable instructions, when executed by a processor, implement the steps in the method of claim 6.
10. A computer program product comprising computer-executable instructions which, when executed by a processor, implement the steps in the method of claim 6.

Description

Classification method of non-contact electrocardiosignals at night Technical Field The application relates to the technical field of health detection, in particular to a classification method of non-contact electrocardiosignals at night. Background At present, the number of patients with cardiovascular diseases reaches 3.3 hundred million people, and 10.3 percent of residents are high-risk people with cardiovascular diseases. Cardiovascular diseases have high prevalence, leading to a high proportion of deaths. Electrocardiographic signal acquisition is one of the important methods for diagnosing cardiovascular diseases. Conventional electrocardiographic acquisition typically uses Ag/AgCl electrodes. However, long-term electrocardiographic monitoring using Ag/AgCl electrodes may irritate the skin and lead to dermatitis. Therefore, more and more scientific researchers acquire electrocardiosignals in a non-contact mode at present, and the electrocardiosignals can be acquired without directly contacting a subject. However, it is susceptible to interference of environmental noise, resulting in a large amount of signals being discarded, and the signal utilization rate is low, so it becomes very important to improve the signal utilization rate with respect to the quality classification method of the non-contact electrocardiographic signal. The current electrocardiosignal quality classification method is mainly aimed at electrocardiosignals acquired by using a traditional mode, and is mostly divided into two types of acceptable and unacceptable, and the signal utilization rate is low. And most methods have complex models and large calculation amount. Finally, the night is the high-incidence period of the acute cardiovascular event, and the current electrocardiosignal quality classification method is not specially designed for night electrocardiosignal monitoring. Therefore, there is an urgent need in the art to develop a classification method of non-contact electrocardiosignals at night, which classifies the electrocardiosignals by adopting a high-precision light deep learning model, improves the utilization rate of the electrocardiosignals, and has low computational complexity and high-efficiency classification. Disclosure of Invention The application aims to provide a night non-contact electrocardiosignal classifying method, which classifies electrocardiosignals by adopting a high-precision light deep learning model, improves the utilization rate of the electrocardiosignals, has low computational complexity and realizes high-efficiency classification. The first aspect of the application provides a training method of a deep learning classification model for classifying non-contact electrocardiosignals at night, which comprises the following steps: (a) Acquiring night non-contact electrocardiosignal data acquired based on a capacitive coupling principle; (b) Preprocessing the night non-contact electrocardiosignal data, segmenting the preprocessed night non-contact electrocardiosignal data, and classifying according to the characteristics of each segment of the night non-contact electrocardiosignal data so as to obtain night non-contact electrocardiosignal data with classification labels; (c) The night non-contact electrocardiosignal data with the classification labels are divided into a training set, a verification set and a test set; (d) Taking the night non-contact electrocardiosignal data with the classification labels in the training set as the input characteristics of a deep learning classification model, taking the classification labels of the night non-contact electrocardiosignal data as the output characteristics of the deep learning classification model, and training the deep learning classification model; the deep learning classification model comprises a feature extraction module and a classification module which are formed by a convolutional neural network, wherein the night non-contact electrocardiosignal data with the classification labels comprises one-dimensional data of electrocardiosignals and corresponding classification labels of each section of night non-contact electrocardiosignal data, and the abscissa of the one-dimensional data of the electrocardiosignals represents time and the ordinate represents signal amplitude. In another preferred embodiment, the preprocessing in step (b) includes removing baseline shift interference and power line interference in the electrocardiographic signals using a band pass filter. In another preferred embodiment, in the step (b), the preprocessed nocturnal noncontact electrocardiosignal data is divided once every 2s-5s, so as to obtain a plurality of sections of nocturnal noncontact electrocardiosignals. In another preferred embodiment, the preprocessed nocturnal noncontact electrocardiographic signal data is segmented every 3 s. In another preferred embodiment, in the step (b), a classification label is added to each piece of nocturnal noncontact electrocardi