基于神经网络模型的生物组织参数反演算法

Parameters Inversion Algorithm of Biological Tissues Based on a Neural Network Model

针对逆向求解生物组织光学特性参数存在测量精度不高、在体测量困难等问题,提出了一种利用神经网络模型反演生物组织光学参数的方法,该方法以蒙特卡罗算法输出的不同检测距离r处的漫反射率R(r)作为输入,以吸收系数和散射系数作为输出。本文将神经网络算法反演的吸收系数、散射系数值与蒙特卡罗算法获得的吸收系数和散射系数值进行了对比。仿真实验表明:选择r=0.1 cm以及r=0.3 cm距离处的漫反射率作为输入,利用神经网络模型反演的吸收系数和散射系数的平均绝对误差分别为0.003和1.574,一致性决定系数R^(2)分别为0.9997和0.9915。神经网络模型反演的生物组织参数与蒙特卡罗算法获得的吸收系数、散射系数具有较好的一致性,且反演精度高,操作简单,为生物组织光学参数的在体测量提供了新思路。

During the inversion of optical parameters of biological tissues,the measurement accuracy is low and the in vivo measurement is difficult.Therefore,a neural network model to invert the optical parameters of biological tissues was proposed in this paper.In this method,the diffuse reflectance R(r)at different detection distances r from the Monte Carlo algorithm is used as the input,and the absorption coefficient and scattering coefficient are taken as the output.The absorption coefficient and scattering coefficient retrieved by the neural network algorithm are compared with those by the Monte Carlo algorithm.The simulation results show that with the diffuse reflectance at r=0.1 cm and r=0.3 cm as the input,the mean absolute errors are 0.003 and 1.574,respectively for the absorption coefficient and scattering coefficient retrieved by the neural network algorithm,and the consistency coefficient of determination R^(2) can reach 0.9997 and 0.9915,respectively.The biological tissue parameters retrieved by the neural network model agree well with the absorption coefficient and scattering coefficient obtained by the Monte Carlo algorithm.The neural network model has the advantages of high inversion accuracy and simple operation,which provides a new method for the in vivo measurement of optical parameters of biological tissues.