Tracking C. elegans swimming for high-throughput phenotyping

The model organism C. elegans is widely used for phenotype studies, where the behaviors of large numbers of individuals have to be measured accurately. Swimming motions exhibit unique aspects of their behavior, but are significantly more difficult to track than crawling motions. Filming multiple worms is necessary for high-throughput phenotype experiments, but adds to the complexity of tracking. Here we present a novel method to segment and track multiple swimming worms. Segmentation is done by low-level-vision processing followed by articulated model fitting. Tracking is performed by iteratively adjusting the articulated models to the successive frames. We tested our method on 404 videos and 2020 worms, and report results of 94.1% of worms tracked, which are valid 92.1% of the time. The models we use enable a cheap computation of a variety of measures of behavior. We report a quantitative comparison of three mutants using existing and novel phenotype measures, showing statistically significant differences of behavior.