基于BP神经网络的左心室心肌组织参数反演方法的研究

Study on the inversion method for left ventricular myocardial tissue parameters based on the back propagation neural network

因为蕴含着心肌组织特性变化等病理特征,人体左心室的变形和动力学特性已成为心脏疾病临床诊断的重要依据。本研究基于BP神经网络方法,通过对左心室临床诊断数据的反演,开展左心室心肌组织参数识别研究。首先,使用Matlab语言编写图像识别程序提取人体左心室CT影像中内外膜位置点,在SolidWorks软件中建立左心室的真实几何模型,通过Abaqus软件建立左心室的有限元分析模型。其次,采用Mooney-Rivlin超弹性模型模拟心肌组织特性,运用Abaqus有限元软件,对左心室有限元模型进行动态数值分析,获得3个特征时刻下对应的45组BP神经网络的输入-目标向量。最后,使用Matlab语言编写BP神经网络程序,对输入-目标向量进行BP神经网络训练,建立左心室诊断数据与心肌组织参数之间的非线性映射关系。对实例的分析结果表明,BP神经网络可很好地用于基于临床数据的心肌组织参数反演,可望成为临床诊断因心肌组织特性变化引起左心室病变的一种有效方法。

The mechanical dynamic responses of human left ventricle have become an important basis for clinically diagnosing heart diseases because they contain the pathological characteristics such as variation of myocardial tissue parameters.Based on the back propagation(BP)neutral network,we studied the myocardial tissue parameters via the inversion of clinical response data.First,positions of internal and external membranes were extracted from the CT images of human left ventricle by Matlab software to write image recognition program,the geometric model was established via Solidworks software,and the finite element modeling and analysis(FEM)mesh was eventually generated via Abaqus software.Next,the BP neutral network was developed by means of Matlab tools,the nonlinear mapping was established between the tissue parameters and the responses after the BP network was trained with 45 sets of BP input-target vectors corresponding to three characteristic instants.The inversed results of four examples indicate that the BP neutral network is well applicable to the inversion of tissue parameters based on clinical data,and may be an effective approach to clinically diagnosing the heart diseases caused by change of myocardial tissue parameters in the human left ventricle.