基于BP神经网络的肺音识别与诊断研究

Research of lung sound recognition and diagnosis based on BP neural network

通过大量研究表明肺音蕴含丰富的肺器官生理病态信息,呼吸道空气的吸入,呼出经由支气管管道,其因肺部及其呼吸道病变而产生通道狭窄,部分堵塞以及分泌物的堆积,在听诊过程中将在肺部噪音的振动频率,声波增幅以及幅升降梯度等进行特征表现,通过对特征值的准确判断从而确定病因。本文将通过小波变换方法对三十份以上的同类型具有相同病理特征的肺音波形进行多频率的小波分解并根据各个频率带中不同的信号响应,分散率等计算小波特征系数,然后通过BP人工神经网络将所得特征系数作为输入值,优化权重系数,最后进行对为大、中、小湿罗音及哮鸣音等信号的检测验证,实现检测精准率达到90%以上。

Through a large number of studies have indicated that lung sound contains rich lung physiological and pathological information,inhalation of respiratory tract air exhaled through the bronchial tubes, the due to lung and respiratory disease and narrow channel, part of the blockage and accumulation of secretions and in auscultation process will be in lung noise vibration frequency,acoustic amplitude and amplitude fluctuation gradient of features,through the feature value of the accurate judgment to determine the cause. The through wavelet transform method for more than 30 copies of the same type has the same pathological features of lung sound waveform of multi frequency wavelet decomposition and according to the signal of each frequency band in different response,dispersion rate calculation of the wavelet coefficients, and then through the BP artificial neural network will be income characteristic coefficient as input values, optimal weighted coefficient,finally to for large,medium and small wet rale and wheezing sound sound signal testing and certification,to achieve accurate detection rate reached more than 90%.