Innovative molecular diagnostic strategy for salivary gland lesions based on ATR-FTIR spectroscopy coupled with machine learning.

OBJECTIVE

This study aimed to identify the biochemical signatures that distinguish pleomorphic adenoma (PA) and mucoepidermoid carcinoma (MEC) from normal parotid tissue (NP), while also exploring molecular differences relevant to benign-malignant tumor differentiation using Attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy combined with machine-learning (ML) approaches.

STUDY

DESIGN

A total of 48 formalin-fixed, paraffin-embedded (FFPE) salivary gland tissue sections (5 µm thickness) were analyzed, including 8 NP, 24 PA, and 16 MEC specimens. ATR-FTIR spectra were acquired directly from routine histopathological slides over the 4000-400 per cm range without chemical deparaffinization. The spectral data were then processed using multivariate analysis (PCA and OPLS-DA) to reveal biochemical variations among the tissue groups. Supervised classification was then performed using Support Vector Machine (SVM) and k-Nearest Neighbours (kNN) algorithms with 6-fold cross-validation to evaluate classification performance.

RESULTS

Distinct spectral differences were detected among NP, PA, and MEC tissues, particularly in protein (amide I-II), lipid (CH-stretching), and nucleic acid (phosphate) regions. OPLS-DA demonstrated clear class discrimination (3D: J₁ = 4.1250, J₂ = 2.2121). The SVM classifier achieved the highest diagnostic accuracy (F1-scores: NP = 1.00, PA = 0.96, MEC = 0.94; overall accuracy = 95.83%), outperforming the kNN model (accuracy = 93.75%).

CONCLUSIONS

Combining ATR-FTIR spectroscopy with ML-based multivariate analysis provides a highly accurate method of differentiating NP, PA and MEC salivary gland tissues based on their biochemical fingerprints. This integrative approach establishes the basis for a rapid, non-invasive, label-free diagnostic tool that is compatible with clinical practice and applicable to routine histopathological samples.