Multivariate classification of fourier transform infrared hyperspectral images of skin cancer cells
Multivariate classification of fourier transform infrared hyperspectral images of skin cancer cells
复制标题
皮肤癌细胞傅里叶变换红外高光谱图像的多元分类
DOI：
10.1109/eusipco.2016.7760464
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
J. Schnekenburger
中科院分区：
文献类型：
--
作者：
Francisco Peñaranda;V. Naranjo;L. Kastl;B. Kemper;G. Lloyd;J. Nallala;N. Stone;J. Schnekenburger
描述了从傅里叶变换红外图像中提取的光谱的多级分类的多级框架。这种学习结构用于区分从两批四种不同皮肤培养细胞（两种正常细胞和两种肿瘤细胞）的高光谱图像中提取的光谱，其中一批细胞已用荧光活细胞染料染色。在框架的每个阶段都探索了不同的选项，特别是在光谱预处理和所采用的分类算法中。特别注意优化学习模型并通过交叉验证客观地估计泛化性能。对于所有未染色的皮肤细胞类型都获得了非常高的辨别性能。然而，正如多个分类实验所证明的那样，污渍的存在会引入光谱伪影，从而恶化类别分离。
A multilevel framework for the multiclass classification of spectra extracted from Fourier transform infrared images is described. This learning structure was employed to discriminate the spectra extracted from hyperspectral images of two batches of four different skin cultured cells (two normal and two tumor), where the cells of one batch had been stained with fluorescence live cell dyes. Different options were explored in each stage of the framework, specifically in the spectral pre-processing and the employed classification algorithm. Special care was taken to optimize the learning models and to objectively estimate the generalization performance by means of cross-validation. A very high discriminative performance was obtained for all the unstained skin cell types. However, the presence of the stains introduces spectral artifacts that worsen the class separation, as has been demonstrated in several classification experiments.