Machine Learning-Enabled Non-Invasive Screening of Tumor-Associated Circulating Transcripts for Early Detection of Colorectal Cancer

Abstract

Colorectal cancer (CRC) is a major cause of cancer-related mortality, highlighting the need for accurate and non-invasive diagnostics. This study assessed the utility of tumor-associated circulating transcripts (TACTs) as biomarkers for CRC detection and integrated these markers into machine learning models to enhance diagnostic performance. We evaluated five models-Generalized Linear Model, Random Forest, Gradient Boosting Machine, Deep Neural Network (DNN), and AutoML-and identified the DNN model as optimal owing to its high sensitivity (85.7%) and specificity (90.9%) for CRC detection, particularly in early-stage cases. Our findings suggest that combining TACT markers with AI-based analysis provides a scalable and precise approach for CRC screening, offering significant advancements in non-invasive cancer diagnostics to improve early detection and patient outcomes.