Colorectal Cancer Detection in Sweat Samples Using Data Processing Techniques and the Cyranose 320 Electronic Nose

Abstract

Colorectal cancer (CRC) remains a leading cause of cancer-related mortality worldwide, underscoring the urgent need for non-invasive and cost-effective screening strategies. This study evaluates the feasibility of using the Cyranose 320 electronic nose to discriminate between CRC patients and healthy controls through volatile organic compound (VOC) analysis of sweat samples. A total of 65 sweat samples (31 CRC, 34 controls) were analyzed. The data processing pipeline included Relative Difference (RD) feature extraction, Quantile Transformer scaling, Orthogonal Signal Correction (OSC), and Principal Component Analysis (PCA), followed by supervised machine learning classification. PCA revealed strong class separability, with the first three principal components explaining 95.72% of the total variance (PC1: 91.28%). Supervised classification using nested cross-validation demonstrated robust performance across seven algorithms. Random Forest achieved the best results, with 95.4% accuracy, 93.5% sensitivity, 97.1% specificity, and an AUC of 0.967. Decision Tree showed comparable performance, while all evaluated models exceeded an AUC of 0.90. Confusion matrix analysis confirmed high true positive rates and minimal false positives, particularly for tree-based ensemble methods. These findings demonstrate that sweat-derived VOC profiling using the Cyranose 320, combined with advanced data preprocessing and multivariate analysis, provides strong discriminative capability for CRC detection. The results support the potential of sweat-based electronic nose systems as a non-invasive, scalable, and patient-friendly screening approach, warranting validation in larger independent cohorts.