广义极大似然估计在OCT无创血糖监测中的应用

Application of Maximum Likelihood Type Estimates in Noninvasive Blood Glucose Monitoring in vivo Using Optical Coherence Tomography

在光学相干层析术(OCT)无创血糖监测过程中,预测模型的建立容易受异常点的干扰。采用广义极大似然估计(M估计)建立的预测模型能够有效地通过权函数降低异常点在模型中的权重。通过人体血糖钳夹临床实验和口服葡萄糖耐量测试实验,利用M估计和最小二乘估计法(OLS估计)两种方法建立了血糖预测模型,采用交互验证法对两种模型的均方根误差(RMSE)进行了比较。对比结果表明,M估计能有效地降低血糖预测模型的RMSE值。此外,利用克拉克误差表格分析法对两个模型的预测结果进行评估,评估结果表明采用M估计建立的血糖预测模型的准确性和稳定性高于OLS估计,因此M估计更适合临床上的OCT无创血糖监测应用。

During the model building process, the blood glucose monitoring model can be easily damaged by abnormal points. Maximum likelihood type estimates （M estimates） introduced in this paper can decrease the weight of abnormal points in the blood glucose model. In glucose clamp experiment and human oral glucose tolerance test, M estimates and ordinary least sum of squares estimates （OLS estimates） are applied to build the blood glucose monitoring models, respectively. Root mean square error （RMSE） of the model built by M estimates is calculated by using interactive verification method. It shows that M estimates can effectively reduce the RMSE value of blood glucose prediction results. In addition, predicted values of blood glucose by the two models are evaluated by Clarke error grid analysis. The results show that the veracity and stability of the blood prediction model built by M estimates are higher than that built by OLS estimates. Thus, the method of M estimates is more suitable for clinical application of noninvasive blood glucose sensing using optical coherence tomography.