基于ECA-1D-CNN的TDLAS的静脉用药浓度定量分析方法研究  

Study on Quantitative Analysis Method of TDLAS Intravenous Drug Concentration Based on ECA-1D-CNN

作　　者：朱永炳 蔡玉琴 蒋力耀 雷春 滕龙 王德旺[3] 陶治 ZHU Yong-bing;CAI Yu-qin;JIANG Li-yao;LEI Chun;TENG Long;WANG De-wang;TAO Zhi(School of Electronics&Information Engineering,Nanjing University of Information Science&Technology,Nanjing 210044,China;School of Integrated Circuits,Nanjing University of Information Science&Technology,Nanjing 210044,China;The First Affiliated Hospital of Nanjing Medical University,Nanjing 210029,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044 [2]南京信息工程大学集成电路学院,江苏南京210044 [3]南京医科大学第一附属医院,江苏南京210029

出　　处：《光谱学与光谱分析》2025年第5期1341-1347,共7页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(62205153);国家重点研发计划课题(2022YFB4401301);国家自然科学基金重大研究计划项目(92264103)资助。

摘　　要：静脉用药溶质浓度的定量分析一直是静配中心药物检测的研究热点,但常规的调配和复核手段都是借助人工操作,存在药液浓度把控受限、人工复核压力繁重且低效等问题,因此提出一种对静脉药物溶质浓度准确、便捷、无损的检测方法显得至关重要。由于传统的近红外光谱对低浓度液体检测有一定局限性,基于可调谐激光吸收光谱技术(TDLAS),研究了了一种基于高效注意力机制一维卷积神经网络(ECA-1D-CNN)的葡萄糖混合溶液浓度定量检测模型。为检测低浓度葡萄糖混合溶液,以TDLAS技术为基础,选择光强吸收率最高的980 nm波段作为激光器光源,通过光电传感器,获取药物浓度的透射光强信号,由锁相放大模块解调为二次谐波信号得到一共600个不同浓度的自建数据集,将样本按8∶2的比例划分为训练集和测试集。以600个药物浓度透射光强的二次谐波信号作为研究对象,采用ECA-1D-CNN进行葡萄糖混合溶液浓度的定量检测。该模型共有4个卷积层,均采用Relu激活函数激活,每个卷积层后添加1个BN层,每两个卷积层添加1个池化层,在第2个池化层后添加1个ECA,可以帮助网络模型更好地学习特征之间的关系,减少参数数量和改善模型的鲁棒性。首先,为了凸显1D-CNN模型的优势,使用相同的原始数据集在PCR、SVR、PLSR上进行建模并对比4种不同模型的预测效果。其次,在6种不同数据预处理的基础上,将ECA-1D-CNN模型与1D-CNN模型进行对比,以决定系数R2、绝对误差MAE、均方根误差RMSE作为评价指标来分析预测模型的泛化能力。结果表明,SG+Normalization预处理下的ECA-1D-CNN模型效果最优,该方法能够对6~30 mg·100 mL^(-1)的葡萄糖混合溶液浓度进行有效预测,其模型训练集R2可达到0.998,MAE为0.295,RMSE为0.343,测试集的R^(2)可达到0.993,MAE为0.498,RMSE为0.691。采用所提出的方法可以精准的预测静脉用药溶质的浓The quantitative analysis of intravenous drug solute concentration has always been the research hotspot of drug detection in static dispensing centers.Still,the conventional means of mixing and reviewing are operated manually.There are problems such as limited control of the concentration of the drug solution,laborious pressure of manual review,and inefficiency,so it is crucial to propose an accurate,convenient,and non-destructive detection method for intravenous drug solute concentration.Due to the limitations of traditional near-infrared spectroscopy for the detection of low-concentration liquids,based on tunable laser absorption spectroscopy(TDLAS)technology,a quantitative detection model of glucose mixed solution concentration based on efficient attention mechanism one-dimensional convolutional neural network(ECA-1D-CNN)was investigated.In order to detect the low concentration of glucose mixed solution,based on the TDLAS technology,the 980 nm band with the highest light intensity absorption rate was selected as the laser light source,and through the photoelectric sensor,the transmitted light intensity signal of the drug concentration was acquired,which was demodulated into the second harmonic signal by the phase-locked amplification module to obtain a total of 600 self-constructed datasets of different concentrations,and the samples were divided into training and testing sets in the ratio of 8∶2.Aiming at the second harmonic signal of the transmitted light intensity of 600 drug concentrations as the research object,a glucose mixed solution concentration detection model based on the one-dimensional convolutional neural network model with efficient attention mechanism(ECA-1D-CNN)is proposed,with a total of four convolutional layers,all of which are activated by the Relu activation function,and a BN layer is added after each convolutional layer,a pooling layer is added after every two convolutional layers,and a pooling layer is added after the 2nd pooling layer,and a pooling layer is added after the 2nd poolin

关 键 词：葡萄糖混合溶液 TDLAS 一维卷积神经网络 高效通道注意力机制 

分 类 号：O657.319[理学—分析化学]