Principal component analysis in application to Brillouin microscopy data

Abstract

Brillouin microscopy has recently emerged as a new bio-imaging modality that provides information on the microscale mechanical properties of biological materials, cells and tissues. The data collected in a typical Brillouin microscopy experiment represents the high-dimensional set of spectral information, i.e. each pixel within a 2D/3D Brillouin image is associated with hundreds of points of spectral data. Its analysis requires non-trivial approaches due to subtlety in spectral variations as well as spatial and spectral overlaps of measured features. This article offers a guide to the application of Principal Component Analysis (PCA) for processing Brillouin imaging data. Being unsupervised multivariate analysis, PCA is well-suited to tackle processing of complex Brillouin spectra from heterogeneous biological samples with minimal a priori information requirements. We point out the importance of data pre-processing steps in order to improve outcomes of PCA. We also present a strategy where PCA combined with k -means clustering method can provide a working solution to data reconstruction and deeper insights into sample composition, structure and mechanics.

Keywords

• Brillouin microscopy
• principle components
• unsupervised learning
• hyperspectral data