基于多波长法的胆红素浓度测量研究

Study on bilirubin concentration measurement based on multi-wavelength method

面向新生儿黄疸无创测量,提出了多波长法的测量方法,以去除肤色差异对测量结果的影响。采用光谱测量法对胆红素浓度进行测量,配制具有不同黑色素浓度的模拟组织液来模拟新生儿的不同肤色;利用微型光谱仪对模拟组织液的透射光谱数据进行采集,并使用S-G(Savitzky-Golay)平滑、暗噪声去除及吸光度转换方法,对光谱数据进行预处理;基于预处理后得到的吸收光谱数据进行量化分析,建立偏最小二乘回归(PLSR)模型,并结合随机蛙跳(RF)波长选择算法进行特征波长提取。实验结果表明:经过RF算法选择出的波长数为25个,仅为原来的8.7%,很好地简化了原有PLSR模型。且简化后模型的R_c~2为0.882,R~2_p为0.871,RMSEC与RMSEP分别为1.6 mg/dL与1.7 mg/dL,具有较好的模型预测性能。

A multi-wavelength method for noninvasive measurement of neonatal jaundice is proposed to remove the effect of skin color difference on the measurement results.The contents include:measuring bilirubin concentration by spectrometric method,preparing simulated tissue fluid with different melanin concentration to simulate different skin color of newborn.The transmission spectral data of simulated tissue fluid are collected by micro spectrometer,and the spectral data are preprocessed by S-G(Savitzky-Golay)smoothing,dark noise removal and absorbance conversion methods.Quantitative analysis is carried out based on the preprocessed absorption spectrum data,the PLS regression(PLSR)model is established,and the characteristic wavelengths are extracted by combining the random frog(RF)wavelength selection algorithm.The experimental results show that the number of wavelengths selected by the RF algorithm is 25,which is only 8.7%of the original number and simplifies the original PLSR model very well.The R2c and R2p of the simplified model are 0.882,0.871,RMSEC and RMSEP are 1.6 mg/dL and 1.7 mg/dL,respectively,that shows good prediction performance of the model.