Digital stain separation for histological images.

It is often desirable to perform digital image analyses on sections prepared for human interpretation, e.g. nuclear chromatin texture analysis or three-dimensional reconstructions using sections requiring human delineation of structures of interest. Unfortunately such analyses are often more effective using stains with less complex contrast. Here an automated selective 'de-staining' method for digital images is presented. The method separates an image into its red, green and blue and hue, saturation and intensity components. A mask of stained tissue is prepared by automatic percentile thresholding. A single weighted inverted colour channel is then added to each of the three primary colour channels separately by an iterative algorithm that adjusts the weights to give minimum variance within the mask. The modified red, green and blue channels are then recombined. This method is automatic requiring no pre-definition of stain colours or special hardware. The method is demonstrated to 'de-stain' nuclei in haematoxylin and eosin (H&E) sections (and a separate haematoxylin image can be derived from this). An image of isolated brown reaction product is produced with immunoperoxidase preparations counterstained with haematoxylin. Furthermore trichrome (haematoxylin van Gieson, picrosirius red) and other common stains may be separated into their components with modifications of the same algorithm. Although other methods for colour separation do exist (e.g. spectral pathology and colour deconvolution) these require special apparatus or precise calibration and foreknowledge of pure dye colour spectra. The present method of digital stain separation is fully automatic with no such prerequisites.