(489a) High-Throughput Characterization of Morphological Phenotypes for Automatic Screening of C. Elegans Using Granulometry and Microfluidics
A major challenge in image processing is the characterization of densely packed overlapping objects. Image analysis on biological samples routinely involves characterization of structures that are overlapping and occluded, yet proper detection of these structures is still very difficult, limiting rapid analysis and large-scale screening. Segmentation algorithms, such as thresholding, edge detection and watershed implementations, have been used for object recognition; however, these techniques are becoming less powerful in the realm of biological imaging due to the shift toward higher content applications where separating individual objects with a high degree of occlusion is challenging and complex. To solve this issue, we present an algorithm for semi-quantitative textural analysis of biological samples using the principles of granulometry.
Granulometry is a morphological processing algorithm where a series of morphological openings allow for the detection of the objects in an image. To demonstrate the capability of this algorithm, we applied the modified granulometry algorithm to the biological application of lipid droplet size characterization in the model organism C. elegans. Lipid droplets are distributed anisotropically in the intestinal cells throughout the length of the worm in a highly occluded manner, which would otherwise be analyzed poorly by segmentation algorithms. Because metabolism and lipid storage pathways are conserved between humans and C. elegans, classification of various lipid droplet phenotypes would allow for the discovery of important genes. Nevertheless, accurate phenotype characterization throughput is low due to the current methods such as hand curating the diameters of individual lipid droplets from images. By coupling the modified granulometry algorithm with microfluidics, a dramatic increase in the throughput can be achieved, giving the ability to conduct an automated high-throughput screen of lipid droplet size distribution phenotypes in C. elegans. This allows for the identification of new lipid storage mutants at approximately 500 times faster than previously capable. Ultimately, this algorithm can be used for identification of lipid droplet phenotypes in other organisms, and determining genes that disrupt lipid storage in C. elegans, which can broaden future investigations that will impact the genetic understanding of metabolism and fat storage.