Leveraging image processing techniques to visualize sub-cellular domains in optical photothermal infrared imaging.

Traditionally, vibrational spectroscopic imaging captures a full spectrum at every pixel, a technique commonly referred to as hyperspectral imaging. With the emergence of high-resolution vibrational imaging methods such as optical photothermal infrared (O-PTIR) spectroscopy, it has become a widespread practice to perform imaging at specific wavenumber settings instead of acquiring full hyperspectral data cubes to speed up the acquisition process. Yet, regardless of the dimensionality of the data, spectroscopic imaging modalities require elaborate data pre-processing to reduce background noise, scattering effects, and imaging artifacts that may otherwise distort chemical information and prevent the reliable implementation of feature extraction and classification algorithms. Although the spectral analysis of chemical images has been extensively studied, there is limited research on spatial pre-processing techniques such as masking and spatial filtering, which may become more important when images are acquired at select wavenumbers only. Furthermore, processing pipelines are often vendor-specific, hampering the comparability of research results, or rely on bespoke algorithms and non-standard dependencies, leading to long-term maintenance and compatibility issues. The aim of this paper is hence to present a straightforward, step-by-step protocol for the spatial processing and subsequent segmentation of sparse chemical imaging data using standard open-source libraries. We demonstrate our approach on a set of single-wavenumber images of human cells recorded with a home-built O-PTIR instrument.