Objective: To select proper optimal methods of regularization parameter for brain electrical impedance tomography. Methods: Four types of regularization parameter choosing methods:L-curve, generalized cross-validation, fixed noise figure method, discrepancy principle, were used in 2-D brain model with high impedance skull, and the performance of the 4 methods were compared by blur radius and position error with statistics of applying 50 different Gauss white noise to measurement voltage. Results:Discrepancy principle proved to be most objective, most robust and most accurate among the four methods. Conclusion:Discrepancy principle is recommended to be used in the reconstruction process to optimize regularization parameter for brain electrical impedance tomography.%目的:寻找一种适用于脑部电阻抗成像(EIT)的正则化参数选取方法。方法:将L型曲线法、广义交叉校验法、固定噪声系数法和偏差原理法4种正则化参数选取算法应用于包含高阻抗颅骨的真实颅脑模型中,然后通过模糊半径和重建位置误差2个指标比较施加50次不同高斯白噪声的重建效果。结果:在4种算法中偏差原理法最客观、最稳定以及最接近真实解。结论:可采用偏差原理法作为脑部EIT的正则参数选取方法。
展开▼