乳腺癌细胞拉曼光谱校正的乳腺癌组织精确拉曼光谱诊断模型构建及诊断系统软件研发

Breast conserving surgery (BCS) is widely accepted as a treatment modality for women with early stage breast cancer. The aim of BCS is to achieve tumor-free margins to avoid re-excision and accomplish good esthetic results. Histopathology, the current gold standard of diagnosis, is time-consuming and the specimen needs to be prepared by additional process, which pushes the re-excision rate up to 17%. Hence, we need a rapid, objective, and accurate diagnostic method to reduce the re-excision rate clinically. Raman spectroscopy is a simple, rapid and highly sensitive analytical technology，making it possible to determine the margin status intraoperatively without any pretreatment. However, due to the complexity of breast cancer tissues, there is a relatively high deviation of Raman spectra testing results in the clinical setting. Therefore the accuracy and specificity of the diagnosis model is not enough to meet the needs of the clinical application. This study intends to use immune magnetic beads to purify the breast tumor cells. The breast tumor cells and fresh breast cancer, benign and normal breast tissues are screened by Raman spectroscopy to collect their spectral characteristics respectively. By utilizing the advantage of stable and precise spectral characteristics from breast tumor cell over breast cancer tissues, we calibrate and optimize the real-time spectral breast diagnosis models using Relief and FS with SVM and KNN machine learning algorithms. Thus improving the accuracy of the models and reducing testing time, and further providing a rationale for determining the margin status and nature of lesions by non-pathological methods as well as manufacturing related equipment.

乳腺癌保乳术的目的是最大可能保证乳房美观（保留乳腺组织）的同时确保完整切除病灶，即切缘无病灶残留。但石蜡病理诊断耗时长，导致保乳术后二次手术率高达17%。临床上急需一种快速、客观、准确的诊断方法来降低二次手术率。拉曼光谱是一种简便、快速、灵敏的光谱技术,可直接对组织进行检测，而无需任何前处理，为术中快速判断切缘情况提供了可能。但是乳腺癌组织成分复杂，导致拉曼光谱检测结果差异较大，构建诊断模型准确性难以满足临床需要。本研究采用免疫磁珠法分离新鲜乳腺癌组织的癌细胞，应用拉曼光谱检测癌细胞和新鲜癌组织、良性病变及正常乳腺组织，收集它们的拉曼光谱特征。利用乳腺癌细胞优于癌组织的稳定而明确的拉曼光谱结果，运用ReliefF和FS方法结合SVM和KNN机器学习算法，校正和优化乳腺癌组织即时诊断模型，提高模型的准确性，缩短检测时间，为进一步非病理方法确定病灶切缘阳性与否以及病变性质，相关仪器设备制造奠定理论研究基础。

乳腺癌保乳术的目的是最大可能保证乳房美观（保留乳腺组织）的同时确保完整切除病灶，即切缘无病灶残留。由于临床现有诊断方法有限等原因，乳腺癌保乳术术后二次手术率高是临床上亟待解决的问题。本研究通过细胞表面标志物分选出原代乳腺癌细胞和正常上皮细胞，分析两者之间特征拉曼光谱的差异与联系，最终通过机器学习构建诊断乳腺癌的模型。我们首次检测了流式分选技术获得的原代乳腺癌细胞和正常上皮细胞的增强拉曼光谱，对比两者的特征拉曼光谱可以发现原代乳腺癌细胞较正常上皮细胞的核酸、蛋白质含量高而脂类含量少，并且一些核酸特征峰、蛋白特征峰发生了位移。此外我们首次用随机森林算法对实验获得的增强光谱进行分类，构建模型的正确率78.2％，准确率75.5％，召回率66.7%。机器学习分析原代乳腺癌增强拉曼光谱构建乳腺癌诊断模型，为即时快速无创的判断保乳术手术切缘、早期诊断乳腺癌奠定了基础。