Development of a Diagnostic Model for Colorectal Cancer Utilizing Surface-Enhanced Raman Spectroscopy Coupled with Machine Learning

Background:Colorectal cancer is a prevalent malignancy with diagnostic challenges. Colonoscopy, the current standard, is invasive and poorly tolerated by patients, while non-invasive modalities suffer from limited sensitivity. This research delves into serum surface-enhanced Raman spectroscopy (SERS) variances between colorectal cancer patients, healthy controls, and individuals at various disease stages. METHODS:Silver nanoparticles were utilized as an enhancement substrate in conjunction with a HORIBA ARAMIS confocal Raman spectrometer to swiftly analyze serum samples from 45 colorectal cancer patients and 35 healthy controls in a label-free manner. Independent samples t-test and t-distributed random neighborhood embedding techniques were employed to compare distinct serum SERS spectra. Subsequently, a Random Forest algorithm was utilized to construct a diagnostic model for colorectal cancer screening and staging based on serum Raman spectroscopy data. The diagnostic performance of the Random Forest model was comprehensively assessed using error rate curves, subject operating characteristic curves, and 5-fold cross-validation. RESULTS:The serum SERS spectra exhibited distinct spectral peaks differentiating colorectal cancer from healthy controls, notably at 491, 635, 724, 957, 1001, 110, 1205, and 1452 cm⁻¹. In the cancer group, higher average spectral intensities were observed at the 724, 957, 1001, 110, and 1452 cm⁻¹peaks, while lower intensities were noted at the 491, 635, and 1205 cm⁻¹ peaks. Integration of SERS spectroscopy with a random forest model yielded a colorectal cancer detection sensitivity of 0.857, specificity of 0.904, and an AUC of 0.881.At spectral peaks of 724 cm⁻¹, 957 cm⁻¹, 1001 cm⁻¹, 110 cm⁻¹, and 1452 cm⁻¹, the intensity of serum SERS spectral peaks exhibited higher levels in TNM stages III-IV compared to stages I-II. Similarly, at characteristic peaks of 492 cm⁻¹ and 635 cm⁻¹, the intensity was also elevated in TNM stages III-IV. Utilizing the SERS technique in conjunction with the random forest model yielded a sensitivity of 0.723 and a specificity of .803 for discriminating between different stages, achieving an AUC value of 0.763. Conclusion:Substantial disparities in peak intensities characteristic of Surface-Enhanced Raman Spectroscopy (SERS) were evident when comparing serum samples from individuals with colorectal cancer to those from healthy controls, as well as across various stages of colorectal cancer.