CN-121987221-A - Portable real-time recording equipment and real-time generation method of electrocardiosignal three-dimensional excitation conduction dynamic diagram based on deep learning

CN121987221ACN 121987221 ACN121987221 ACN 121987221ACN-121987221-A

Abstract

A portable real-time recording device and a real-time generation method of an electrocardiosignal three-dimensional excitation conduction dynamic graph based on deep learning belong to the technical field of electrocardio physiological signal processing and visualization. The system comprises a portable acquisition box, a terminal display device and a terminal display device, wherein the portable acquisition box is used for acquiring a body surface potential mapping electrode array of an electrocardiosignal, the portable acquisition box is used for synchronously acquiring the electrocardiosignal, a deep learning reasoning module is integrated in the portable acquisition box, the electrocardiosignal is processed based on a CNN-DRNN model, potential and three-dimensional coordinate information of an electrocardiosignal excitation point are output in real time, and the terminal display device is connected with the portable acquisition box in a wireless communication mode, receives the potential and the three-dimensional coordinate information, and accordingly performs three-dimensional dynamic display to present the electrocardiosignal excitation conduction track. The method is used for electrocardiosignal analysis.

Inventors

XIANG HAO
ZHONG TAO
KE QUAN
YUAN RUI
JIANG FUCHUN
ZHU LIJUN
ZHAO YONGMING
CHEN FEI
YIN HAO
ZHU CHENLONG
WANG ZHIYUAN
ZHANG QIANG
HU YAOTIAN

Assignees

剑虎医疗科技(苏州)有限公司

Dates

Publication Date: 20260508
Application Date: 20251226

Claims (10)

1. The portable real-time recording device of the electrocardiosignal three-dimensional exciting conduction dynamic graph based on deep learning is characterized by comprising the following components: body surface potential mapping electrode array in the form of The device is used for collecting electrocardiosignals; The portable acquisition box is connected with the body surface potential mapping electrode array, synchronously acquires electrocardiosignals, is internally integrated with a deep learning reasoning module, processes the acquired electrocardiosignals based on a CNN-DRNN model and outputs potential and three-dimensional coordinate information of an electrocardio-excitation point in real time; the terminal display device is connected with the portable acquisition box in a wireless communication mode, receives the electric potential and three-dimensional coordinate information, and performs three-dimensional dynamic display according to the electric potential and the three-dimensional coordinate information so as to present an electrocardio exciting conduction track.
2. The portable real-time recording device of an electrocardiosignal three-dimensional excitation conduction dynamic diagram based on deep learning as claimed in claim 1, wherein at least eight reference electrodes are arranged in the body surface potential mapping electrode array and are respectively positioned at the top center, the bottom center and the back symmetrical position of the front surface of the array; the eight reference electrodes are used for acquiring reference potentials, and the average value of the reference potentials of the reference electrodes is used as a potential datum point which is set as the origin of the three-dimensional coordinate system.
3. The portable real-time recording device of a three-dimensional activation conduction dynamic map of cardiac signals based on deep learning of claim 1, wherein the portable acquisition box comprises: A multi-channel switch for gating each electrode channel; the multichannel analog-to-digital converter is used for acquiring electrocardiosignals with microsecond synchronous precision; The micro control unit is integrated with a neural network processing unit and is used for running the CNN-DRNN model to perform real-time reasoning; the memory is used for storing the collected original data or model parameters; the wireless communication module is used for transmitting data with the terminal display equipment; And the power supply module is used for supplying power to each component.
4. The portable real-time recording device of three-dimensional electrocardiosignal exciting conduction dynamic diagram based on deep learning as claimed in claim 3, wherein the multichannel analog-to-digital converter adopts at least two eight-channel high-precision ECG front-end ADC chips and is matched with a multi-channel switch to realize the recording of Synchronous sampling of the electrode array.
5. The portable real-time recording device of a deep learning based electrocardiographic signal three-dimensional activation conduction dynamic map of claim 1 wherein the CNN-DRNN model comprises: the first feature extraction module is formed by sequentially connecting four parallel two-dimensional convolutional neural network layers A1-A4 and four subsequent largest pooling layers B1-B4 and is used for inputting Carrying out primary space feature extraction and dimension reduction on the electrocardio data; the second feature extraction module is formed by sequentially connecting four two-dimensional convolutional neural network layers C1-C4 and four subsequent largest pooling layers D1-D4 and is used for carrying out deep space feature extraction and further dimension reduction on the primary features; The full connection layer F1 is used for fusing and mapping the deep features after dimension reduction; the time sequence fusion module is formed by sequentially connecting four DRNN hidden layers G1-G4 and is used for modeling the dependency relationship of the characteristics output by the full-connection layer on the time sequence; the output layer H1 is used for predicting quadruple information of the electrocardiograph exciting point based on the output of the time sequence fusion module; The output of each two-dimensional convolutional neural network layer is sequentially subjected to standardization processing and ReLU activation function processing.
6. The portable real-time recording device of a three-dimensional activation conduction dynamic map of cardiac signals based on deep learning of claim 5, wherein the structural parameters of the CNN-DRNN model are configured as follows: the two-dimensional convolution neural network layers A1-A4 in the first feature extraction module all adopt two-dimensional convolution kernels with the size of 4 multiplied by 24, and the convolution step length is 1, and the two-dimensional convolution neural network layers A1-A4 are opposite to each other The output dimension of the dimension input data is ; The maximum pooling layers B1-B4 reduce the length and width of the input feature map by half, and the output dimension is ; The two-dimensional convolutional neural network layers C1-C4 in the second feature extraction module adopt the same convolutional kernel size and step length as those of the layers A1-A4, the subsequent maximum pooling layers D1-D4 execute the same dimension reduction operation as those of the layers B1-B4, and the final output dimension of the module is as follows ; The number of neurons of the full connection layer F1 is ; The number of neurons of each DRNN hidden layer G1-G4 in the time sequence fusion module is ; The number of neurons of the output layer H1 is 10, each neuron outputs four-element group information of an electrocardio-activated point, and the four-element group comprises a potential value and a three-dimensional space coordinate value.
7. The method for generating the electrocardiosignal three-dimensional excitation conduction dynamic graph in real time based on deep learning is characterized by comprising the following steps of: acquisition by body surface potential mapping electrode array A wiry electrocardiosignal; Preprocessing the acquired signals, including background buckling, drift removal, zero-phase filtering and normalization; inputting the preprocessed data into a pre-trained CNN-DRNN model, and reasoning in real time to obtain the potential and three-dimensional coordinates of the electrocardiographic exciting point; and sending the reasoning result to the terminal equipment in a three-dimensional video stream mode for dynamic display.
8. The method for generating the three-dimensional activation conduction dynamic map of the electrocardiographic signal based on deep learning according to claim 7, wherein the CNN-DRNN model is trained by: Synchronously acquiring body surface potential mapping data and intra-cardiac electric signal mapping data as training samples; Taking body surface data as input, and taking an exciting point quadruple in intracardiac data as a label; And extracting spatial features by using a convolutional neural network, carrying out time feature fusion by using a DRNN, and carrying out coordinate prediction by using a full-connection layer.
9. The method for generating the three-dimensional electrocardiosignal excitation conduction dynamic map based on deep learning in real time as claimed in claim 7, wherein a grouping acquisition mode is adopted in the training process, each group of data lasts for 2 seconds, and the body surface data dimension is The label data dimension is 。
10. The method for generating the three-dimensional activation conduction dynamic map of the electrocardiosignal based on deep learning in real time according to claim 7, wherein the terminal display device is a smart phone or a tablet computer, and a built-in processor of the terminal display device is used for reconstructing and displaying three-dimensional images and has standby reasoning capability.

Description

Portable real-time recording equipment and real-time generation method of electrocardiosignal three-dimensional excitation conduction dynamic diagram based on deep learning Technical Field The invention relates to portable real-time recording equipment and a real-time generation method of an electrocardiosignal three-dimensional excitation conduction dynamic graph, and belongs to the technical field of electrocardio physiological signal processing and visualization. Background Electrocardiographic signal analysis is one of the most popular and noninvasive cardiac function assessment means in clinic, and has an important role in abnormal activation point localization of atrial and ventricular tachyarrhythmias (such as supraventricular tachycardia, atrial tachycardia, ventricular tachycardia, etc.). The traditional electrocardiograph technology has been developed into 12-lead or even 18-lead electrocardiographs since the William Earthwen invention string electrocardiograph in 1903, but the traditional electrocardiograph technology still cannot intuitively reflect the real-time conduction process of the cardiac electric excitation in the three-dimensional space, and has the limitations of separation of time domain and space information, single analysis angle and the like. In recent years, the prior art has attempted to achieve multi-dimensional display of electrocardiographic signals. For example, in 1989, the patent "method and imager for imaging three-dimensional electrocardiogram" and the patent "method and system for implementing four-dimensional electrocardiograph diagnostic apparatus" in 2011 propose a multi-dimensional electrocardiograph concept, which provides a space-time combined idea for electrocardiograph analysis, but still lacks detailed electro-anatomical mapping function of heart, and cannot implement three-dimensional dynamic visualization of activation conduction. In addition, although the CNN and LSTM based abnormal excitation point positioning method (CN 110555388 a) proposed by university of Zhejiang in 2019 introduces a deep learning technique, it performs abnormal point positioning only for 12-lead data, and does not realize three-dimensional real-time portable application of the multi-lead excitation conduction process. In the aspect of the existing product, although common Holter devices (such as Philips DigiTrak XT, GE SEER 12, michael EPM10 and the like) can realize long-term electrocardiograph recording, the device does not have an electrocardiograph excitation conduction three-dimensional dynamic display function. Although 'CardioInsight CARDIAC MAPPING SYSTEM' of the Medun force company can provide a noninvasive three-dimensional electrical signal anatomical map of the heart, the anatomical map is needed to be used in combination with CT images, and the equipment is huge (1300 mm in height, 664mm in depth, 615mm in width and 102kg in weight) and cannot meet the clinical requirements of portability and real-time, and is particularly not suitable for long-time monitoring and abnormal point pre-positioning before operation. Therefore, the prior art still lacks a device capable of realizing three-dimensional dynamic display of electrocardiosignal excitation conduction in real time and portability, and is particularly suitable for preoperative positioning and observation of conditions which are not suitable for being induced in surgery, such as paroxysmal ventricular arrhythmia. Disclosure of Invention The invention aims to solve the problems that the traditional electrocardiogram cannot intuitively display the three-dimensional dynamic conduction process of cardiac electric excitation, the existing three-dimensional mapping system is huge in size and cannot be used portably, and the three-dimensional excitation track is difficult to display in real time and continuously, and provides portable real-time recording equipment and a real-time generation method of the three-dimensional excitation conduction dynamic map of the cardiac electric signal based on deep learning. The invention relates to a portable real-time recording device of an electrocardiosignal three-dimensional exciting conduction dynamic graph based on deep learning, which comprises the following components: body surface potential mapping electrode array in the form of The device is used for collecting electrocardiosignals; The portable acquisition box is connected with the body surface potential mapping electrode array, synchronously acquires electrocardiosignals, is internally integrated with a deep learning reasoning module, processes the acquired electrocardiosignals based on a CNN-DRNN model and outputs potential and three-dimensional coordinate information of an electrocardio-excitation point in real time; the terminal display device is connected with the portable acquisition box in a wireless communication mode, receives the electric potential and three-dimensional coordinate information, and performs three-dimensio