Rapid differentiation of patients with lung cancers from benign lung nodule based on dried serum Fourier-transform infrared spectroscopy combined with machine learning algorithms

Abstract

Lung cancer (LC) is associated with poor 5-year survival rates when diagnosed at advanced stages. While low-dose computed tomography (LDCT) screening enables earlier detection, its high false-positive rate, primarily due to benign lung nodules (BLN), necessitates more accurate diagnostic tools. This study developed a rapid and precise LC discrimination method by integrating Fourier transform infrared (FTIR) spectroscopy of dried serum samples with machine learning algorithms. We analyzed dried serum from 58 LC patients, 37 BLN patients, and 36 healthy controls. Five machine learning models, linear discriminant analysis (LDA), support vector machine (SVM), random forest, multilayer perceptron (MLP), and LightGBM, were optimized using FTIR spectral data (1800–900 cm⁻¹ band). All algorithms successfully differentiated the three groups, with LDA achieving the highest accuracy (93.9 %). These results demonstrate that dried serum FTIR spectroscopy coupled with machine learning, particularly LDA, offers a promising approach for distinguishing LC from BLN, potentially augmenting LDCT screening to reduce unnecessary interventions.