Early screening for colorectal cancer: study on a serum detection method based on SERS and machine learning
10.3760/cma.j.cn114452-20240930-00542
- VernacularTitle:结直肠癌早期筛查：基于SERS与机器学习的血清检测方法研究
- Author:
Limao LI
1
;
Yong HUANG
;
Zhenguang WANG
;
Jiaxiang LIN
;
Zheng WU
;
Xiaowei CAO
;
Wei WEI
Author Information
1. 扬州大学附属江都人民医院胃肠外科，扬州225200
- Publication Type:Journal Article
- Keywords:
Spectrum analysis, raman;
Serum;
Artificial intelligence;
Principal component analysis;
Support vector machines
- From:
Chinese Journal of Laboratory Medicine
2025;48(2):214-222
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To establish a serum detection method of surface-enhanced Raman spectroscopy (SERS) combining with machine learning for early screening of colorectal cancer (CRC).Methods:Serum samples were collected from 150 CRC patients diagnosed at Jiangdu People′s Hospital, Affiliated to Yangzhou University, and also from 37 healthy subjects. Gold nanohexapod (AuNHs) arrays were prepared using an oil-water interface self-assembly method. A 5 μl serum sample was applied onto the AuNHs array. Scatheless and rapid detection for serum were performed using a Renishaw inVia Raman spectrometer at room temperature with a laser wavelength of 785 nm, exposure time of 10 s, and power of 5 mW. The raw SERS spectra were preprocessed using Savitzky-Golay smoothing, AsLS baseline correction, and Min-Max normalization with Origin 2019 software. Furthermore, the principal component analysis (PCA)-support vector machine (SVM) model was constructed using Python′s scikit-learn library. Leave-One-Out Cross-Validation (LOOCV) was used to evaluate the model′s accuracy, sensitivity, specificity, and area under the curve (AUC).Results:The AuNHs arrays exhibited uniform morphology. The relative standard deviation (RSD) of the SERS intensity at 1 080 cm -1 was 5.69%, and the RSD of the SERS intensity at 1 340 cm -1 was 6.20%. The limit of detection (LOD) of the AuNHs array was 9.42×10 -12 mol/L. The PCA-SVM model achieved an accuracy of 90.91% (170/187), sensitivity of 96.79% (181/187), specificity of 99.47% (186/187), and an AUC of 0.98. The most significant characteristic peaks distinguishing different CRC stages were at 747, 940, 1 000, 1 447, and 1 612 cm -1. Conclusion:The serum detection method based on SERS combined with machine learning can accurately screen CRC with higher accuracy, sensitivity, and specificity, demonstrating potential clinical application value.
