基于一维深度卷积神经网络模型的脑肿瘤组织拉曼光谱分类  

Raman spectral classification of brain tumor tissue based on one-dimensional deep convolutional neural network model

作　　者：刘晨曦 魏东 冉义兵 周柳营 马川 LIU Chenxi;WEI Dong;RAN Yibing;ZHOU Liuying;MA Chuan(School of Electrical and Information Engineering,Beijing University of Civil Engineering and Architecture,Beijing 100044,China;Beijing Key Laboratory of Intelligent Processing for Building Big Data,Beijing 100044,China)

机构地区：[1]北京建筑大学电气与信息工程学院,北京100044 [2]建筑大数据智能处理方法研究北京市重点实验室,北京100044

出　　处：《分析试验室》2025年第2期223-232,共10页Chinese Journal of Analysis Laboratory

基　　金：北京市属高校高水平创新团队建设计划项目(IDHT20190506);北京建筑大学高级主讲教师培育计划(GJZJ20220803)资助。

摘　　要：采用深度学习进行正常脑组织与肿瘤组织边缘诊断,对算法效率和准确率均有较高要求,为此提出一种1D-RamNeXt脑肿瘤组织拉曼光谱分类方法。首先,针对样本类别不均衡问题,设计一维数据再平滑数据增强方法,以扩充原光谱数据集;为提升诊断准确率,通过堆叠残差网络(ResNet)块并增加通道数量,构建卷积神经网络(CNN)模型,对光谱数据特征进行提取及整合,并在网络中引入并行拓扑结构残差模块,以解决深度神经网络中层数过多导致的模型退化问题,并减少超参数及算法计算量。针对所提取的小鼠脑部组织实验结果表明,与线性判别分析(LDA)、 K最邻近(KNN)和支持向量机(SVM)类传统算法相比,本文方法准确率提升6%;与AlexNet, VggNet和ResNet一维CNN模型相比,本文方法在准确率提升2%的前提下运行速度提升16%。Utilizing deep learning to distinguish normal brain tissue and tumor tissue at their edges places stringent demands on both algorithmic efficiency and accuracy.Hence,a 1D-RamNeXt Raman spectroscopy classification approach for brain tumor tissues was presented.To address the issue of class imbalance within the dataset,a one-dimensional data smoothing data augmentation method was devised to augment the original spectral dataset.For improving diagnostic accuracy,a convolutional neural network(CNN)model was constructed by stacking Residual Network(ResNet)blocks and increasing the number of channels to extract and integrate spectral data features.Additionally,within the network,parallel topological residual modules were incorporated,mitigating the degradation of the model caused by excessive network depth,while simultaneously reducing hyperparameters and algorithmic computational load.Experimental results obtained from mouse brain tissue samples indicated that the proposed method outperformed traditional algorithms,such as Linear Discriminant Analysis(LDA),K-Nearest Neighbor(KNN),and Support Vector Machine(SVM),yielding a 6%increase in accuracy.Compared to one-dimensional CNN models like AlexNet,VggNet,and ResNet,this proposed method can not only boost accuracy by 2%but also accelerate computation by 16%.

关 键 词：拉曼光谱 卷积神经网络 脑肿瘤 残差结构 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]