Image Processing Reference
In-Depth Information
Trajectory evaluation and
behavioral scoring using
JAABA in a noisy system
H. Chen 1 ; P. Marjoram 1 ; B. Foley 2 ; R. Ardekani 2
1 Department of Preventive Medicine, Keck School of Medicine, USC, Los Angeles,
2 Molecular and Computational Biology, Department of Biological Sciences, USC, Los Angeles, CA, USA
Typical methods for generating and analysing tracking data for animals such as Drosophila require ideal-
ized experimental conditions that are frequently difficult, expensive, or undesirable to replicate in the
lab. In this chapter, we describe and implement methods for improving robustness in nonideal condi-
tions. Our method involves an initial processing step in which tracks are constructed from noisy video
data, followed by a subsequent application of machine learning algorithms to further improve perform-
ance. We demonstrate that our methods are capable of generating a substantial improvement in perform-
ance in fly identification, and allow for effective construction of tracks for individual flies. We then use
this data to train sex and behavior classifiers, which we employ to detect and describe behavioral difer-
ences among test experiments. As such, our algorithm provides a path for groups who wish to track fly
or characterize their behavior, in less than ideal conditions.
Machine learning
The authors gratefully acknowledge funding from NSF and NIMH through awards DMS
1101060 and MH100879. The material contained in this chapter reflects the views of the authors,
and not necessarily those of NSF or NMH.
Search WWH ::

Custom Search