| CPC G16H 50/70 (2018.01) [A61B 5/7246 (2013.01); A61B 5/7267 (2013.01)] | 5 Claims |
|
1. An electrocardiogram (ECG) mapping method, comprising the following steps:
acquiring a palm ECG signal, the palm ECG signal comprising multiple complete ECG cycles;
dividing the palm ECG signal into multiple palm ECG episodes according to a preset intercepted length;
converting the palm ECG episodes into standard ECG episodes with an ECG mapping model;
connecting all standard ECG episodes according to a connection sequence of the multiple palm ECG episodes to obtain a standard ECG signal corresponding to the palm ECG signal;
the ECG mapping method further comprising: training a back propagation (BP) neural network to obtain the ECG mapping model, and comprising:
acquiring multiple historical palm ECG signals and multiple historical standard ECG signals, wherein the historical palm ECG signals are in one-to-one correspondence with the historical standard ECG signals, and the historical palm ECG signals and the historical standard ECG signals each comprise multiple complete ECG cycles;
respectively dividing the historical palm ECG signals and the historical standard ECG signals according to the preset intercepted length to obtain multiple historical palm ECG episodes and multiple historical standard ECG episodes, wherein the historical palm ECG episodes and the historical standard ECG episodes are in one-to-one correspondence;
grouping all historical palm ECG episodes according to a first preset number to obtain multiple groups of first sets;
taking a first set in a first group as a first input set;
inputting the first input set to a hidden layer in the BP neural network to obtain a standard ECG computing set; and
adjusting a weighting parameter of the hidden layer according to the standard ECG computing set and the first input set by using a first loss function, taking a first set in a next group to the first input set as a first input set in a next round, returning to the step of “inputting the first input set to a hidden layer in the BP neural network to obtain a standard ECG computing set” until a first set in a last group is input to the hidden layer, thereby completing one iteration, and returning to the step of “grouping all historical palm ECG episodes according to a first preset number to obtain multiple groups of first sets” until a number of iterations reaches a first preset number of times to obtain a well-trained BP neural network that is the ECG mapping model;
before the training a BP neural network to obtain the ECG mapping model, further comprising: determining an initial weighting parameter of the hidden layer in the BP neural network with a sparse autoencoder;
wherein the determining an initial weighting parameter of the hidden layer in the BP neural network with a sparse autoencoder comprises:
learning characteristics of the historical palm ECG episodes through the sparse autoencoder according to all historical palm ECG episodes to determine a weighting parameter of each of an encoding layer and a decoding layer in the sparse autoencoder; and
determining the initial weighting parameter of the hidden layer in the BP neural network according to the weighting parameter of the encoding layer;
wherein the learning characteristics of the historical palm ECG episodes through the sparse autoencoder according to all historical palm ECG episodes to determine a weighting parameter of each of an encoding layer and a decoding layer in the sparse autoencoder comprises:
grouping all historical palm ECG episodes according to a second preset number to obtain multiple groups of second sets;
taking a second set in a first group as a second input set;
inputting the second input set to the sparse autoencoder, and obtaining an output set through the encoding layer and the decoding layer; and
adjusting the weighting parameter of each of the encoding layer and the decoding layer according to the second input set and the output set by using a second loss function, taking a second set in a next group to the second input set as a second input set in a next round, returning to the step of “inputting the second input set to the sparse autoencoder, and obtaining an output set through the encoding layer and the decoding layer” until a second set in a last group is input to the sparse autoencoder, thereby completing one iteration, and returning to the step of grouping all historical palm ECG episodes according to a second preset number to obtain multiple groups of second sets until a number of iterations reaches a second preset number of times to determine the weighting parameter of each of the encoding layer and the decoding layer.
|