Relevant aspects of unmixing/resolution analysis for the interpretation of biological vibrational hyperspectral images

Abstract

The use of multivariate unmixing/resolution methods for the analysis of biological vibrational hyperspectral images is crucial to characterize the morphology and spectral signatures of the different biological tissues or cell compartments. This work provides a general data analysis protocol to interpret Raman and FT-IR hyperspectral images of biological samples. To do so, Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) is the core tool proposed. The protocol starts by describing dedicated preprocessing steps suitable to handle the nature and artifacts of the spectroscopic technique used. Later on, the focus is on the use of MCR-ALS to analyze single images or image multiset structures that contain images with related information. Relevant issues related to advanced use of MCR-ALS on biological image analysis, such as the modeling of mixed non-biological signal contributions or the elucidation and active use of information related to the presence/absence of biological contributions in different images (samples) in multiset structures, are described. Additional aspects, such as the only use of the FT-IR fingerprint region vs. using extended spectral ranges in FT-IR to improve the differentiation among contributions, are also considered. The proposed data analysis methodology is demonstrated on real Raman and FT-IR images from zebrafish tissue cryosections.