K-Means Clustering Analysis of the Membrane Structure of Giant Unilamellar Vesicles from the Hyperspectral Dark-Field Microscopy

Giant Unilamellar Vesicles (GUVs) serve as simplified models mimicking the biological membranes. To investigate the membrane integrity of GUVs, we introduced cholesterol (CHOL) in concentrations ranging from 0 to 40 mol%. The reflectance spectra of GUVs, resulting from the absorption or scattering of photons upon their vesicular interactions, were captured spatially via the hyperspectral dark-field microscopy. To analyze the spectral variation inside the lipid membranes, we utilized the K-Means clustering technique, a machine-learning algorithm that categorizes data into distinct "clusters" based on similarities in their characteristics. We used K-Means from the Spectral Python (SPy) Module to section the membrane structure of GUVs based on the similarities in their spectral profiles. The K-Means algorithm was found to be very effective in segmenting the membrane structure with submolecular resolution. Our results are promising in revealing the patterns of GUVs' microstructure in relation to the CHOL concentrations based on the spectral attributes. Integrating machine learning to the analysis of hyperspectral microscopy data leads to the exceptional submolecular resolution otherwise unavailable in the live sample imaging.