Tumor margin identification and prediction of the primary tumor from brain metastases using FTIR imaging and support vector machines
Tumor margin identification and prediction of the primary tumor from brain metastases using FTIR imaging and support vector machines
复制标题
DOI：
10.1039/c3an00326d
复制
发表时间：
2013-01-01
期刊：
影响因子：
4.2
通讯作者：
Popp, J. uergen
中科院分区：
文献类型：
--
作者：
Bergner, Norbert;Romeike, Bernd F. M.;Popp, J. uergen
红外光谱能够基于其固有的振动指纹识别组织类型，而无需以非破坏性的方式染色。在这里，傅立叶变换红外显微图像收集自22个脑转移组织切片的膀胱癌，肺癌，乳腺癌，结肠癌，前列腺癌和肾细胞癌。本研究的范围是区分癌与正常组织和坏死组织的红外光谱，并利用癌的红外光谱来确定脑转移瘤的原发肿瘤。数据处理遵循先前已经开发的用于分析这些样品的拉曼图像的程序，并且包括解混算法N-FINDR、通过k均值聚类的分割和通过支持向量机（SVM）的分类。与随后的苏木精-伊红染色的训练样本组织切片比较，第一级支持向量机的正确分类率为98.8%的脑组织，98.4%的坏死和94.4%的癌。通过二级支持向量机，对训练数据集的FTIR图像的原发性肿瘤进行了正确预测，总正确率为98.7%。最后，将两水平判别模型应用于四个独立样本进行验证。尽管与训练样本相比分类率略有降低，但独立样本的大部分红外光谱被分配给正确的原发性肿瘤。结果表明，FTIR成像的能力，以补充脑组织诊断的组织病理学工具。
Infrared spectroscopy enables the identification of tissue types based on their inherent vibrational fingerprint without staining in a nondestructive way. Here, Fourier transform infrared microscopic images were collected from 22 brain metastasis tissue sections of bladder carcinoma, lung carcinoma, mamma carcinoma, colon carcinoma, prostate carcinoma and renal cell carcinoma. The scope of this study was to distinguish the infrared spectra of carcinoma from normal tissue and necrosis and to use the infrared spectra of carcinoma to determine the primary tumor of brain metastasis. Data processing follows procedures that have previously been developed for the analysis of Raman images of these samples and includes the unmixing algorithm N-FINDR, segmentation by k-means clustering, and classification by support vector machines (SVMs). Upon comparison with the subsequent hematoxylin and eosin stained tissue sections of training specimens, correct classification rates of the first level SVM were 98.8% for brain tissue, 98.4% for necrosis and 94.4% for carcinoma. The primary tumors were correctly predicted with an overall rate of 98.7% for FTIR images of the training dataset by a second level SVM. Finally, the two level discrimination models were applied to four independent specimens for validation. Although the classification rates are slightly reduced compared to the training specimens, the majority of the infrared spectra of the independent specimens were assigned to the correct primary tumor. The results demonstrate the capability of FTIR imaging to complement histopathological tools for brain tissue diagnosis.