Advancing frontline early pancreatic cancer detection using within-class feature extraction in FTIR spectroscopy

Abstract

This study introduces a novel approach for the early detection of pancreatic cancer through biofluid spectroscopy, leveraging a unique machine learning pipeline comprising class-specific principal component analysis (PCA), linear discriminant analysis (LDA), and support vector machine (SVM) in both real patient and synthetic data. By conducting separate PCA on cancerous and non-cancerous samples and integrating the projections prior to LDA and SVM classification, we demonstrate significantly improved diagnostic accuracy compared to traditional methods. This methodology not only enhances predictive performance but also offers deeper insights into the influence of molecular spectra on model efficacy. Our findings, validated on real patient data, suggest a promising avenue for developing non-invasive, accurate diagnostic tools for early-stage pancreatic cancer detection.