Meeting Abstract
Animal behavior research often requires the capture and subsequent analysis of behavior recorded from live organisms. Quantifying the expression of behavior (especially in insects) can be challenging when using human observers alone. The purpose of this project was to develop a computer-automated analysis method for discrete behaviors (measured qualitatively) in ants which would yield a high level of coherence with the evaluations of a separate human observer. We were specifically interested in examining the expression of forward antennal movements in a sample of animal data (n = 4; 25 observations/animal) collected from a previous, unpublished study. Western harvester ants harnessed in a modified 100 μL pipette tip were exposed to a total of twenty-five presentations of an odorant stimulus and/or untreated cotton-tipped applicator. Forward movements of the antenna during stimulus presentations were scored as a binary outcome (yes/no) by a blind researcher. Videos were analyzed in MATLAB by defining a region of interest and measuring the spatially-averaged intensity and spatially-averaged intensity gradient for each frame of the video file. Threshold values were set to a multiple (1, 2, 3, 4, and 5x) of the noise level, which was computed using the median of the absolute deviation, and the expression of forward movement was determined by whether a measurement during the period of stimulus presentation exceeded the threshold value. Overall, this automated method yielded high levels agreement (μ = 85%) and inter-rater reliability with a human observer at low thresholds, but lower levels of agreement and inter-rater reliability at threshold values of 3 or more. This automated method will help accelerate rapid and accurate analysis of ant behavior in future studies.
