Combination of label-free SERS-based nanosensor and machine learning for diagnosis of cholangiocarcinoma

Early detection of cholangiocarcinoma (CCA) is vital for informing therapeutic strategies and predicting survival outcomes in patients. This study aims to demonstrate an accurate, label-free method for diagnosing CCA by analyzing a single drop of serum using surface-enhanced Raman spectroscopy (SERS) with silver nanorod substrate and applying machine learning (ML) to the data. Serum samples (n = 194) included those from CCA cases (n = 58), hepatocellular carcinoma (HCC, n = 48), liver metastases (LM, n = 44), and healthy individuals (HA, n = 44). A 2-µL drop of diluted serum (1:320) with with deionized water) was applied on a label-free silver nanorod SERS chip, and 49 points were randomly examined for each sample. The gathered Raman spectra were analyzed using principal component analysis to reduce their dimensionality, and then partial least-squares discriminant analysis was employed to identify diagnostic clusters. Among different ML models tested, Tthe integration of SERS and machine learningwith linear discriminant analysis (LDA) yielded a diagnostic accuracy of 81% for differentiation among cancers (CCA, HCC, LM) and HA as evaluated by train-test split method. Additionally, the area under the receiver operating characteristic curve achieved 0.95 for separating CCA, HCC, LM, and HA groups. Overall, the results indicate that the use of SERS-based diagnostic techniques in conjunction with machine learning has great potential for accurately distinguishing CCA from other liver cancers, such as HCC and LM. These surface-enhanced Raman spectra are ideally suited for developing cost-effective, clinically relevant, point-of-care diagnostic methods for community-based CCA screening.早期检测胆管癌（cholangiocarcinoma, CCA）对于制定患者治疗策略、预测生存结局至关重要。本研究旨在展示一种精准、无需标记的胆管癌诊断方法：利用搭载银纳米棒基底的表面增强拉曼光谱（surface-enhanced Raman spectroscopy, SERS）分析单滴血清，并将机器学习（machine learning, ML）应用于检测数据。本研究共纳入194份血清样本，其中包括胆管癌患者样本58份、肝细胞癌（hepatocellular carcinoma, HCC）患者样本48份、肝转移瘤（liver metastases, LM）患者样本44份，以及健康个体（healthy individuals, HA）样本44份。将2μL经去离子水按1:320比例稀释的血清滴加至无需标记的银纳米棒SERS芯片上，每份样本随机选取49个检测点采集数据。所采集的拉曼光谱通过主成分分析（principal component analysis）进行降维处理，随后采用偏最小二乘判别分析（partial least-squares discriminant analysis）识别诊断聚类。在测试的多种机器学习模型中，将SERS与线性判别分析（linear discriminant analysis, LDA）相结合的方案，经训练测试划分法评估后，在区分癌症组（CCA、HCC、LM）与健康组的诊断准确率达81%。此外，该方案用于区分CCA、HCC、LM及HA四组的受试者工作特征曲线（receiver operating characteristic curve）下面积可达0.95。整体结果表明，基于SERS的诊断技术与机器学习相结合，在精准区分胆管癌与其他肝癌（如肝细胞癌、肝转移瘤）方面具有巨大应用潜力。此类表面增强拉曼光谱技术非常适合开发成本效益高、符合临床需求的社区人群胆管癌筛查即时诊断方法。