Enhancing Non-Invasive Colorectal Cancer Screening with Stool DNA Methylation Markers and LightGBM Machine Learning

Abstract

Abstract Objective: This study evaluates the effectiveness of stool DNA methylation markers CNRIP1, SFRP2, and VIM, along with Fecal Occult Blood Testing (FOBT), in the non-invasive screening of colorectal cancer (CRC), further integrating these markers with the Light Gradient Boosting Machine (LightGBM) machine learning (ML) algorithm. Methods: The study analyzed 100 stool samples, comprising 50 CRC patients and 50 healthy controls, from the First Affiliated Hospital of Gannan Medical University. Methylation Specific PCR (MSP) was used for assessing the methylation status of CNRIP1, SFRP2, and VIM gene promoters. FOBT was performed in parallel. Diagnostic performance was assessed using Receiver Operating Characteristic (ROC) curve analysis, and a LightGBM-based ML model was developed, incorporating these methylation markers and FOBT results. Results: ROC analysis demonstrated that SFRP2 had the highest diagnostic accuracy with an AUC of 0.87 (95% CI: 0.794-0.946) and a sensitivity of 0.88. CNRIP1 and VIM also showed substantial screening effectiveness, with AUCs of 0.83 and 0.80, respectively. FOBT, in comparison, had a lower predictive value with an AUC of 0.67. The LightGBM-based ML model significantly outperformed individual markers, achieving a high AUC of 0.95 (95% CI: 0.916-0.991). However, the sensitivity of the ML model was 0.78, suggesting a need for improvement in correctly identifying all positive CRC cases. Conclusion: Stool DNA methylation markers CNRIP1, SFRP2, and VIM exhibit high sensitivity in non-invasive CRC screening. The integration of these biomarkers with the LightGBM ML algorithm enhances the diagnostic accuracy, offering a promising approach for early CRC detection.