肿瘤预防与治疗2026，Vol.39Issue(1)：12-20,9.DOI:10.3969/j.issn.1674-0904.2026.01.003
应用血清拉曼光谱区分乳腺癌不同的HR状态
Differentiation of HR Status in Breast Cancer Using Serum Raman Spec-troscopy
摘要
Abstract
Objective:To develop a serum-based Raman spectroscopy technique to assess its concordance with core nee-dle biopsy(CNB)for determining hormone receptor(HR)status in breast cancer patients.Methods:A total of 1 710 pa-tients with invasive breast cancer were enrolled from the Department of Breast Surgery at Sichuan Cancer Hospital between Ju-ly 2021 and May 2023.Tumor tissue samples obtained via CNB and peripheral venous blood samples were collected from these patients.Patients who met the inclusion criteria[first diagnosis,complete clinical data,and human epidermal growth factor receptor 2(HER-2)-negative status]underwent CNB.Immunohistochemical staining and in situ hybridization were then performed on the biopsy specimens to determine the molecular subtype and HR status.After preprocessing the peripheral venous blood samples,serum was obtained for Raman spectroscopy acquisition.The resulting spectral data were then prepro-cessed and split into a training set(90%)and an independent test set(10%).Feature extraction was performed on the training set using the t-test,Kruskal-Wallis U test,Pearson correlation,and mutual information.The extracted features were used to build models with logistic regression(LR),random forest(RF)and support vector machine(SVM).These models were then validated on the independent test set.The results from the different modeling methods were compared to evaluate the concordance between the serum Raman spectra and the pathological HR status labels.Results:A total of 231 patients were enrolled in the study,with 134 patients in the HR+group(mean age:51.1 years)and 97 patients in the HR-group(mean age:49.2 years).Feature selection was performed using the t-test,Kruskal-Wallis U test,Pearson correlation,and mutual information.These selected features were then used to train and evaluate three classifiers:LR,RF,and SVM.The overall AUC values across all method-classifier combinations ranged from 0.71 to 0.90.The detailed performance is as fol-lows:t-test:LR=0.85,RF=0.83,SVM=0.86;Kruskal-Wallis U test:LR=0.89,RF=0.84,SVM=0.90;Pearson correlation:LR=0.81,RF=0.80,SVM=0.83;Mutual information:LR=0.71,RF=0.78,SVM=0.71.Among these,the combination of the Kruskal-Wallis U test and SVM achieved the highest AUC(0.90).However,when evaluated for classification consistency using the Kappa statistic,LR demonstrated superior performance(Kappa=0.72)when paired with the Kruskal-Wallis U test.Conclusion:Raman spectroscopy data from serum samples showed strong agreement with CNB results after modeling.Although the Kruskal-Wallis U test combined with SVM achieved the high-est AUC(0.90),the combination with LR demonstrated superior overall performance when both AUC and Kappa were con-sidered,highlighting its greater potential for clinical development.关键词
拉曼光谱/人工智能/乳腺癌/激素受体/液体活检Key words
Raman spectroscopy/Artificial intelligence/Breast cancer/Hormone receptor/Liquid biopsy分类
医药卫生引用本文复制引用
李彦君,陈鳕姨,刘悦,杜雨杭,郭豪,李俊杰,张倩,王硕,李林涛..应用血清拉曼光谱区分乳腺癌不同的HR状态[J].肿瘤预防与治疗,2026,39(1):12-20,9.基金项目
四川省医学会医学科研项目(编号:S20250023) （编号:S20250023）
成都市科技局技术创新研发项目(编号:2024-YF05-01955-SN) This study was supported by grants from Sichuan Medical Association(No.S20250023)and Chengdu Sci-ence and Technology Bureau(No.2024-YF05-01955-SN). （编号:2024-YF05-01955-SN）
