Application of ATR-FTIR spectroscopy and multivariate statistical analysis in cancer diagnosis

Abstract

Lung cancer is one of the most prevalent and lethal malignant tumors worldwide. Currently, clinical diagnosis primarily relies on chest X-ray examinations, histopathological analysis, and the detection of tumor markers in blood. However, each of these methods has inherent limitations. The current study aims to explore novel diagnostic approaches for lung cancer by employing attenuated total reflection-Fourier transform infrared (ATR-FTIR) spectroscopy in conjunction with multiple machine learning models. Fourier transform infrared spectroscopy can detect subtle differences in the material structures that reflect the carcinogenic process between lung cancer tissues and normal tissues. By applying principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) to analyze infrared spectral data, these subtle differences can be amplified. The study revealed that the combination of spectral bands within the 3500–3000 cm^-1 and 1600–1500 cm^-1 ranges is particularly significant for differentiating between the two groups. Three classification models—Support Vector Machine (SVM), k-Nearest Neighbor (kNN), and Linear Discriminant Analysis (LDA)—were constructed for spectral analysis of various band combinations. The results indicated that in detecting lung cancer samples, the combination of the 3500–3000 cm^-1 and 1600–1500 cm^-1 bands offers significant advantages. The analysis of the receiver operating characteristic (ROC) curve demonstrated that the area under the curve (AUC) exceeded 0.95 for all models, with the LDA model achieving an accuracy rate of 99.4% in identifying lung cancer patients compared to healthy individuals. The findings suggest that the integration of ATR-FTIR spectroscopy with multiple machine learning models represents a promising auxiliary diagnostic method for clinical lung cancer diagnosis, enabling detection at the molecular level.