Pre-processing techniques for colour digital pathology image analysis

Abstract

Digital pathology (DP) can provide extensive information from captured tissue samples and support accurate and efficient diagnosis, while image analysis techniques can offer standardisation, automation, and improved productivity of DP. Since DP images are intrinsically captured as colour images, such analysis should appropriately exploit the colour or spectral information residing in the obtained data. However, before analysing colour DP images, some pre-processing is typically necessary to both support and enable effective analysis. Colour calibration aims to ensure accurate colour information is recorded, while colour enhancement is useful to be able to obtain robust performance of image analysis algorithms which are otherwise sensitive to imaging conditions and scanner variations. Employed methods range from calibrating the cameras and scanners over correcting the displayed colours to transferring the image to another colour representations that in turn can improve e.g. segmentation or other subsequent tasks. Colour deconvolution allows to effectively separate the contributions of stains localised in the same area and thus allows analysis of stain specific images, while variations in colour appearance of histopathology images due to e.g. scanner characteristics, chemical colouring concentrations, or different protocols, can be reduced through application of colour normalisation algorithms. In this paper, we give an overview of commonly employed colour image pre-processing techniques for digital pathology, summarising important work in colour calibration, colour enhancement, colour deconvolution and colour normalisation.