Automatic classification and clustering of caenorhabditis elegans using a computer vision system

Abstract

In this paper, we introduce a computer vision system for automatic classification and clustering of C. elegans according to their behavioral phenotypes. We extract three kinds of features such as worm movement, body size, and body shape. A total of 117 features are extracted for each worm. Then the features are used to build an optimal classification tree using the CART(Classification and Regression Tree). We also try to find optimal clusters by using the gap statistic and hierarchical clustering method. For the experiment, we use 860 sample worms of 9 types (wild, goa-1, nic-1, egl-19, tph-1, unc-2, unc-29, unc-36, and unc-38). According to our experimental results, average success classification rate for wild, goa-1, nic-1, and egl-19 types is 92.3% while the rate for the other types is 70.3%. And the optimal number of clusters is 8 in our case. © Springer-Verlag 2003.