Visibility-Graph Analysis of Raman Spectra in Chalazion and Lipid Dysregulation: An Exploratory Study

A chalazion is an inflammatory lesion of the eyelid caused by a blockage in the sebaceous glands. To identify subtle biochemical differences in chalazion secretions and patient serum, we apply a graph-theory-based visibility-graph approach to Raman spectra, focusing on its exploratory potential for detecting band-specific spectral differences. As a classical baseline, we also employ Principal Component Analysis (PCA); however, the visibility-graph approach provides complementary, band-localized information that may support interpretation of subtle spectral differences. The study included samples from chalazion patients and control subjects, comprising healthy individuals and patients with lipid disorders. Raman spectra were analyzed by transforming the data into visibility graphs, enabling identification of hidden correlations and signal dynamics. Within the VG representation, group differences were concentrated mainly in the 1100–1300 cm−1 range, with localized extrema near 1177, 1184, and 1260 cm−1 in chalazion secretions and near 1172, 1174, 1219, and 1257 cm−1 in serum. The identified positions overlap regions with possible lipid- and protein-related contributions, including C–C, =CH, and amide III vibrations, and are treated as candidate spectral positions rather than markers of individual molecular compounds. From the perspective of complex-network-based signal analysis, the visibility-graph transformation provides an alternative way of representing Raman spectra and extracting structural information that is not readily apparent in the original spectral domain. This information-oriented approach is consistent with entropy-related studies of complex signals, where nonlinear structure, local organization, and hidden dependencies in data are of central interest. These findings suggest that visibility-graph analysis may be useful as an exploratory tool for identifying candidate lipid-related spectral features, although further validation using larger sample sets is required. Full-range PCA was also examined over 200–2000 cm−1. It revealed global covariance patterns distributed across several spectral regions, whereas VG localized the pairwise differences within the 1100–1300 cm−1 range at specific Raman-shift positions.