Towards Efficient Automated Characterization of Irregular Histology Images via Transformation to Frieze-Like Patterns

Abstract

Histology is used in both clinical and research contexts as a highly sensitive method for detecting morphological abnormalities in organ tissues. Although modern scanning equipment has enabled high-throughput digitization of high-resolution histology slides, the manual scoring and annotation of these images is a tedious, subjective, and sometimes error-prone process. A number of methods have been proposed for the automated characterization of histology images, most of which rely on the extraction of texture features used for classifier training. The irregular, nonlinear shapes of certain types of tissues can obscure the implicit symmetries observed within them, making it difficult or cumbersome for automated methods to extract texture features quickly and reliably. Using larval zebrafish eye and gut tissues as a pilot model, we present a prototype method for transforming the appearance of these irregularly-shaped tissues into one-dimensional, "frieze-like" patterns. We show that the reduced dimensionality of the patterns may allow them to be characterized with greater efficiency and accuracy than by previous methods of image analysis, which in turn enables potentially greater accuracy in the retrieval of histology images exhibiting abnormalities of interest to pathologists and researchers.