Histopathological image analysis

The invention relates to an apparatus and a computer-implemented method for training a machine-learning algorithm to perform histopathological analysis. The method comprises obtaining 210 a plurality of first microscopic images of first histological specimens that have been stained with a first marker, followed by the scanning 212 of a related group of second microscopic images of second histological samples that have been stained with a different dye. The method further comprises obtaining 220 a respective plurality of mask images (Figs. 12, 13) generated for the second microscopic images, each mask identifying a histological feature of interest tinted in the associated second image by the second marker. The technique also comprises training 240 the machine-learning algorithm to predict, from a first microscopic image, a histological feature of interest that would be highlighted in the same specimen by the second marker. A region of interest (ROI) may be selected in each first image and a third marker may be used. The first marker may comprise haemotoxylin, eosin and a first immunohistochemical marker, while the second marker may comprise a second immunohistochemical marker. All markers may be fluorescent. The machine-learning algorithm may be a neural network.