甲状腺结节
结核(地质)
甲状腺
线性判别分析
放射科
医学
甲状腺癌
诊断准确性
人工智能
计算机科学
内科学
古生物学
生物
作者
Xiao Liu,Meihuan Wang,Kaining Zhang,Huawei Zhang,Yongchao Lai
标识
DOI:10.1016/j.cej.2023.144794
摘要
The diagnostic rate of thyroid nodules has increased considerably in recent years. Malignant nodules can reduce the quality of and even endanger the lives of patients. Fine needle aspiration (FNA) with cytological diagnosis is the most common modality for diagnosing malignant and benign nodules. However, a certain percentage of nodules still cannot be diagnosed definitively, especially when they are less than 10 mm in diameter. Surface-enhanced Raman spectroscopy (SERS) is a molecule-specific analysis method suitable for tumor diagnosis. In this study, the SERS technique was used to analyze FNA samples of thyroid nodules smaller than 10 mm and combined with multiple machine learning algorithms (principal component analysis-linear discriminant analysis, partial least squares discriminant analysis (PLS-DA), and support vector machines) to develop a novel and precise diagnosis model for malignant and benign thyroid nodules. The results showed that the hyphenated method of SERS and the PLS-DA algorithm efficiently diagnosed malignant and benign thyroid nodules, with an area under the curve (AUC) of 97.479% and accuracies of 95.59% and 85.58% in the training and cross-validation sets, respectively. Moreover, the classification model outperformed cytology in diagnosing FNA samples from microscopic thyroid cancers. This study demonstrates that a novel diagnostic strategy based on SERS and machine learning can constitute a complementary diagnosis method for nodules with unknown FNA cytological findings. Furthermore, this strategy can improve the diagnostic accuracy of thyroid nodules smaller than 10 mm, achieve an early and exact diagnosis of microscopic thyroid cancer, and prevent misdiagnoses, underdiagnoses, and overtreatment of thyroid nodules.
科研通智能强力驱动
Strongly Powered by AbleSci AI