Classification of breast-tissue microarray spots using texton histograms

Breast-tissue microarrays facilitate the survey of very large numbers of tumours but their scoring by pathologists is time consuming, typically highly quantised and not without error. Automated segmentation of cells and intra-cellular compartments in such data can be problematic for reasons that include cell overlapping, complex tissue structure, debris, and variable appearance. This paper proposes a computationally efficient approach that approximates the density of colour and local invariant features by clusters in the feature space, and characterises each spot by a frequency histogram of nearest cluster centres. Spots are classified into four main types based on their histograms. This approach does not rely on accurate segmentation of individual cells. Classification performance was assessed using 344 spots from the Adjuvant Breast Cancer (ABC) Chemotherapy Trial. A two-layer neural network yielded better classification results than a nearest-neighbour classifier or a single-layer network. Some reasons for classification results in disagreement with pathologist-provided labels are discussed and include the existence of spots containing large proportions of different tissue types