支持向量回归在人体血红蛋白无创检测中的应用

发布时间：2018-06-10 15:11

  本文选题：无创检测 + 人血红蛋白　； 参考：《分析化学》2017年09期

【摘要】：采用线性渐变滤光片(Linear variable filter,LVF),优化设计高性能、便携式的人体血液成分近红外检测设备,研究了支持向量回归(Support vector regression,SVR)模型对人体血红蛋白(Hemoglobin,Hb)的预测能力及稳定性,以实现贫血疾病的无创诊断。无创采集100位志愿者食指前端光谱信息并划分定标集、验证集1和2。应用网格搜索方法优选惩罚参数与核函数参数c=5.28,g=0.33,用以建立稳健的SVR模型。随后,分别对验证集1和2中Hb水平进行定量分析。实验结果表明:预测标准偏差(RMSEP)分别为10.20 g/L和10.85 g/L,相对预测标准偏差(R-RMSEP)为6.85%和7.48%,测量精度较高且SVR模型对不同样品的适应性较强,基本满足临床检测要求。基于SVR算法自行设计的LVF型近红外光谱检测设备在贫血症的无创诊断中有着良好的应用前景。
[Abstract]:Linear variable filter was used to optimize the design of high performance and portable near infrared detection equipment for human blood composition. The predictive ability and stability of support vector regression (SVR) model for human hemoglobin (HB) were studied. In order to achieve non-invasive diagnosis of anemia. Noninvasive spectral information was collected from the front end of index finger of 100 volunteers and the calibration set was divided into two sets: 1 and 2. The penalty parameter and kernel function parameter cn5.28g / 0.33 are selected by grid search method to establish a robust SVR model. Then, the HB levels in validation set 1 and 2 were quantitatively analyzed respectively. The experimental results showed that the prediction standard deviation (RMSEPs) was 10.20 g / L and 10.85 g / L respectively, and the relative standard deviation (R-RMSEPV) was 6.85% and 7.48%, respectively. The accuracy of the prediction was higher and the SVR model was more adaptable to different samples, which basically met the requirements of clinical detection. LVF NIR detector based on SVR algorithm has a good application prospect in noninvasive diagnosis of anemia.
【作者单位】： 中国科学院长春光学精密机械与物理研究所;中国科学院大学;吉林大学第一医院肿瘤中心;
【基金】：国家高技术研究发展计划(863计划)(No.2012AA022602) 国家自然科学基金(Nos.61308067,61475155) 吉林省科技发展计划项目(No.20140204078GX) 应用光学国家重点实验室基金资助项目~~
【分类号】：O657.33;R446.11
，

本文编号：2003651