Automated segmentation and classification of zebrafish histology images for high-throughput phenotyping

Because of its small size and rapid development, the larval zebrafish is an ideal model organism for studying mutant phenotypes using "high-throughput" histological analysis. Although the preparation and subsequent digitization of zebrafish larval histology specimens can be conducted in parallel, the scoring and annotation of the resulting virtual slides is largely manual and therefore rate limiting, which motivates the development of systems for automated characterization of histology images. We present a prototype for automated segmentation and classification of histology images in animal models, with a pilot study focusing on larval zebrafish eye and gut images. We show that the segmentation of the images into regions of individual cell layers can be conducted with good precision using combinations of widely-used image processing operations, and that the resulting classification system, based on a decision tree algorithm, exhibits promising performance.