Abstract
<title>Abstract</title> <p>Non-invasive screening is critically important for the early diagnosis of colorectal cancer (CRC). Surface-enhanced Raman scattering (SERS) has shown great potential in liquid biopsy, yet it still faces the challenge of poor signal reproducibility in complex serum matrices. In this study, a flexible three-dimensional SERS platform based on a 4-mercaptophenol (4-MP)-modified glass fiber filter membrane (GFF@4MP) was developed and combined with machine learning for label-free discrimination of CRC serum. Through sodium chloride modification, silver nanoparticles were uniformly immobilized onto the three-dimensional porous scaffold. Crucially, 4-MP, serving as a surface functionalization modifier, was able to modulate the interfacial physicochemical microenvironment and provide specific binding sites. This optimized the ordered adsorption of serum targets within SERS “hot spots”, significantly reduced non-specific interference, and improved spatial uniformity (relative standard deviation of 7.63%). In the analysis of clinical serum samples (30 CRC cases and 30 healthy controls), the substrate modified with 4-MP functional groups yielded SERS spectra with superior intergroup discrimination. Combined with principal component analysis (PCA), the support vector machine (SVM) model achieved the best classification performance on an independent test set, with a prediction accuracy of 93.33% and an area under the ROC curve (AUC) as high as 0.988. This study effectively enhanced the stability of SERS detection in complex biological samples through interfacial chemical modification, thereby providing a reliable methodological strategy for non-invasive clinical screening of colorectal cancer.</p>