Laser- und Medizin-Technologie GmbH, Berlin (LMTB), Fabeckstraße 60-62, 14195 BerlinbTechnical University Berlin, Institute for Optics and Atomic Physics, 10587 Berlin.
Laser- und Medizin-Technologie GmbH, Berlin (LMTB), Fabeckstraße 60-62, 14195 Berlin.
J Biomed Opt. 2014 Feb;19(2):027007. doi: 10.1117/1.JBO.19.2.027007.
Raman spectroscopy based discrimination of cervical precancer and normal tissue has been shown previously in vivo with fiber probe based measurements of colposcopically selected sites. With a view to developing in vivo large area imaging, macro raster scans of native cervical cone biopsies with an average of 200 spectra per sample are implemented (n=16). The diagnostic performance is evaluated using histopathological mapping of the cervix surface. Different data reduction and classification methods (principal component analysis, wavelets, k-nearest neighbors, logistic regression, partial least squares discriminant analysis) are compared. Using bootstrapping to estimate confidence intervals for sensitivity and specificity, it is concluded that differences among different spectra classification procedures are not significant. The classification performance is evaluated depending on the tissue pathologies included in the analysis using the average performance of different classification procedures. The highest sensitivity (91%) and specificity (81%) is obtained for the discrimination of normal squamous epithelium and high-grade precancer. When other non-high-grade tissue sites, such as columnar epithelium, metaplasia, and inflammation, are included, the diagnostic performance decreases.
先前已有研究表明，基于光纤探头的拉曼光谱技术可对阴道镜下选择的部位进行体内检测，从而区分宫颈癌前病变和正常组织。为了开发体内大面积成像技术，我们对未经处理的宫颈锥形活检标本进行了平均每个样本 200 次光谱的宏观光栅扫描（n=16）。通过对宫颈表面的组织病理学图谱进行评估，来评价诊断性能。比较了不同的数据降维和分类方法（主成分分析、小波、k-最近邻、逻辑回归、偏最小二乘判别分析）。通过自举法估计敏感性和特异性的置信区间，得出不同光谱分类程序之间的差异不显著的结论。根据分析中包含的组织病理学差异，评估分类性能，并使用不同分类程序的平均性能进行评估。对于正常鳞状上皮和高级别癌前病变的区分，获得了最高的敏感性（91%）和特异性（81%）。当包括柱状上皮、化生和炎症等其他非高级别组织部位时，诊断性能会下降。
