The liquid and lyophilized blood plasma of patients with benign or malignant thyroid nodules and healthy individuals were studied by terahertz (THz) time-domain spectroscopy and machine learning. The blood plasma samples from malignant nodule patients were shown to have higher absorption. The glucose concentration and miRNA-146b level were correlated with the sample’s absorption at 1 THz. A two-stage ensemble algorithm was proposed for the THz spectra analysis. The first stage was based on the Support Vector Machine with a linear kernel to separate healthy and thyroid nodule participants. The second stage included additional data preprocessing by Ornstein-Uhlenbeck kernel Principal Component Analysis to separate benign and malignant thyroid nodule participants. Thus, the distinction of malignant and benign thyroid nodule patients through their lyophilized blood plasma analysis by terahertz time-domain spectroscopy and machine learning was demonstrated.

中文翻译：

---

通过太赫兹光​​谱分析血浆来区分恶性和良性甲状腺结节


通过太赫兹（THz）时域光谱和机器学习对良性或恶性甲状腺结节患者和健康个体的液体和冻干血浆进行研究。恶性结节患者的血浆样本显示有较高的吸收。葡萄糖浓度和 miRNA-146b 水平与样品在 1 THz 处的吸收相关。提出了一种用于太赫兹光谱分析的两阶段集成算法。第一阶段基于具有线性内核的支持向量机来区分健康参与者和甲状腺结节参与者。第二阶段包括通过 Ornstein-Uhlenbeck 核主成分分析进行额外的数据预处理，以区分良性和恶性甲状腺结节参与者。因此，通过太赫兹时域光谱和机器学习对冻干血浆进行分析，证明了恶性和良性甲状腺结节患者的区别。