摘要
目的探讨支持向量机-递归特征消去法（SVM-RFE）分析拉曼光谱在乳腺良恶性疾病鉴别诊断中的价值。方法收集168例手术患者的新鲜乳腺组织标本，其中正常组织51例，良性病变组织66例，恶性病变组织51例，均进行拉曼光谱检测，SVM-RFE方法处理数据，构建模型，马氏距离法判断数据处理方法的优劣。结果共得到1800个拉曼光谱，良性和恶性乳腺组织的特征峰出现在1281、1341、1381、1417、1465、1530和1637cm“处，而正常乳腺组织的特征峰出现在1078、1267、1301、1437、1653和1743cm。处。良性和恶性乳腺组织的主要不同集中在1340和1480cm“处。SVM-RFE判断正常和恶性乳腺组织的正确率分别为100．O％和95．0％，判断良性乳腺组织的正确率为93．0％。结论正常、良性与恶性病变组织的拉曼光谱存在显著差异，SVM-RFE可以用来构建鉴别乳腺病变性质的模型。
Objective To explore the value of application of support vector machine-recursive feature elimination （SVM-RFE） method in Raman spectroscopy for differential diagnosis of benign and malignant breast diseases. Methods Fresh breast tissue samples of 168 patients （all female; ages 22-75） were obtained by routine surgical resection from May 2011 to May 2012 at the Department of Breast Surgery, the First Hospital of Jilin University. Among them, there were 51 normal tissues, 66 benign and 51 malignant breast lesions. All the specimens were assessed by Raman spectroscopy, and the SVM-RFE algorithm was used to process the data and build the mathematical model. Mahalanobis distance and spectral residuals were used as discriminating criteria to evaluate this data-processing method. Results 1 800 Raman spectra were acquired from the fresh samples of human breast tissues. Based on spectral profiles, the presence ofl 078, 1 267, 1 301, 1 437, 1 653, and 1 743 cm-1 peaks were identified in the normal tissues; and 1 281, 1 341, 1 381, 1 417, 1 465, 1 530, and 1 637 cm-l peaks were found in the benign and malignant tissues. The main characteristic peaks differentiating benign and malignant lesions were 1 340 and 1 480 cm-1. The accuracy of SVM-RFE in discriminating normal and malignant lesions was 100.0%, while that in the assessment of benign lesions was 93.0%. Conclusions There are distinct differences among the Raman spectra of normal, benign and malignant breast tissues, and SVM-RFE method can be used to build differentiation model of breast lesions.
出处
《中华肿瘤杂志》
CAS
CSCD
北大核心
2014年第8期582-586,共5页
Chinese Journal of Oncology
基金
国家自然科学基金青年基金（81202078）
吉林省科技发展计划青年基金（20130522030JH）
关键词
乳腺肿瘤
光谱分析
拉曼
支持向量机
递归特征消去法
诊断
鉴别
Breast neoplasms
Spectrum analysis, raman
Support vector machine-recursive feature elimination
Diagnosis, differential
