一种新的动态心电数据的神经网络压缩方法

A NEW METHOD OF AMBULATED ECG COMPREESION USING NEURAL NETWORKS

本文提出了一种新的、性能更加稳定的动态心电数据的神经网络压缩算法。该方法采用一种不全联接的三层前馈神经网络，将一个ＥＣＧ心搏表示为三个主要的波即Ｐ波，ＱＲＳ波和Ｔ波。三个波的输入与输出只通过少量的隐层单元相联接，并通过各波的隐层单元将相邻波的边缘联系起来。这种方法的优点是在不增加计算量的情况下提高算法对波形的重现能力、较为有效地避免因为Ｐ波和Ｔ波受到干扰，波形变异或其它因素的影响而导致的波形重现失败，提高神经网络压缩方法的鲁棒性和实用性。

In the method, a threelayer feedforward neural network structure incompletely connected was employed. Each heart beat was divided into three major waves, i.e. P, QRS complex and T waves. The inputs and outputs of each wave were connected by two hidden units respectively. The inputs and outputs in conjunction of the neighbouring waves were also connected by the hidden units. The advantage of this method was that it could enhance the capability of recovering the waveforms without increasing computation burden, it could also efficiently avoid the failure of redisplaying the waveforms when the P and T waves were unstable in waveform or highly noised, thus strengthening the robustness and feasibility of ECG compression techniques using neural networks.