Enhanced Early Detection of Colorectal Cancer via Blood Biomarker Combinations Identified Through Extracellular Vesicle Isolation and Artificial Intelligence Analysis

Abstract

Colorectal cancer (CRC) remains a major cause of cancer-related deaths worldwide, with early detection being crucial for improving survival rates. Despite the potential of extracellular vesicles (EVs) as blood biomarkers for CRC diagnosis, their clinical utility is limited due to complex and time-consuming isolation methods, unverified biomarkers and low diagnostic performance. Here, we introduce the ZAHV-AI system, which combines the zeolite-amine and homobifunctional hydrazide-based extracellular vesicle isolation (ZAHVIS) platform with AI-driven analysis for enhanced CRC diagnosis. The ZAHVIS platform enables simple, rapid and cost-effective EV isolation and one-step extraction of EV-derived proteins and nucleic acids (NAs), providing a streamlined approach. Using blood plasma samples from 80 CRC patients across all stages and 20 healthy individuals, we identified four EV-derived miRNA blood biomarkers (miR-23a-3p, miR-92a-3p, miR-125a-3p and miR-150-5p) by confirming statistical significance with relative quantification (RQ) values from real-time PCR and integrated these with carcinoembryonic antigen (CEA) levels into an AI-driven diagnostic model. Among 31 combinations used to identify optimal sets, optimal combination (miR-23a-3p, miR-92a-3p, miR-150-5p and CEA) for overall CRC achieved an area under the curve (AUC) of 0.9861, outperforming individual markers and conventional CEA tests. Notably, the system achieved perfect performance in detecting stages 0–1 (AUC: 1.0) and demonstrated high accuracy for stage 2 (AUC: 0.9722) and early-stage CRC (AUC: 0.9861), using stage-specific optimal combinations. Therefore, the ZAHV-AI system offers a reliable and clinically relevant tool for CRC diagnostics, significantly enhancing early detection and monitoring capabilities.