A Novel Approach for the Colour Deconvolution of Multiple Histological Stains

Colour Deconvolution (CD) is a commonly used tool in histological medical image analysis that separates histochemical or immunohistochemical stains into their component parts. Traditional CD uses matrix inversion to change the Red, Green and Blue (RGB) channels of an image into a new domain that is representative of reference colours but this limits the total number of stains that can be separated. This is problematic for histological staining protocols that use more than three stains, such as FAST staining. This limitation has restricted the use of multicolour staining in light microscopy. To address this issue, this paper evaluates the use of Non Negative Matrix Factorisation (NNMF) and Non Negative Least Squares (NNLS) to enable the decomposition of multistained histological sections into its source components. It will be shown that NNLS is better suited to imaging modalities such as Whole Slide Image (WSI) scanners and that the multiple staining metrics produced from a single sample are analogous to those generated by applying single reagents to contiguous tissue sections.