Over the past decade, confocal Raman microspectroscopic (CRM) imaging has matured into a useful analytical tool to obtain spatially-resolved chemical information on molecular composition of biological samples and found its way into histopathology, cytology and microbiology. A CRM imaging dataset is a hyperspectral image in which Raman intensities are represented as a function of three coordinates: a spectral coordinate encoding the wavelength and two spatial coordinates x and y. Understanding CRM imaging data is 2 challenging because of its complexity, size, and moderate signal-to-noise-ratio. Spatial segmentation of a CRM imaging data is a way to reveal regions of interest and is traditionally performed using non-supervised clustering which relies on spectral domain-only information.Their main drawback is the high sensitivity to noise. We present a new pipeline for spatial segmentation of CRM imaging data which includes pre-processing in the spectral and spatial domains, and k-means clustering. Its core is the pre-processing routine in the spatial domain, edge-preserving denoising (EPD), which exploits the spatial relationships between Raman intensities acquired at neighboring pixels. Additionally, we propose to use spatial correlation to identify Raman spectral features co-localized with defined spatial regions and confidence maps to assess the quality of spatial segmentation. For CRM data acquired from a mid-saggital Syrian hamster (Mesocricetus auratus) brain cryosections, we show how our pipeline benefits from the complex spatial-spectral relationships inherent in the CRM imaging data. EPD significantly improves the quality of spatial segmentation that allows us to extract the underlying structural and compositional information contained in the Raman microspectra.