Behaviors detected with Behavior recognition
Definitions
See also the reference in Notes.
Rooting with the muzzle or digging with the front paws in the bedding material (for Mouse Behavior Recognition only).
The subject licks at the spout of the water bottle.
The subject eats at the feeder or from the floor, or is eating while holding food in front paws.
The subject grooms snout, head, fur or genitals. Includes scratching and licking of paws during a grooming session.
The subjects moves forward with both hind limbs at the same time. Head goes first, followed by the rear. Unlike Jumping, Hopping is scored as a point event with no duration, and does not interrupt behavior states (for Mouse Behavior Recognition only).
The subject moves quickly forward with both hind limbs at the same time. Jumping interrupts other states like Walking (for Rat Behavior Recognition only).
The subject stands in an upright posture, with front paws not in contact with any object. Includes the rise and descend.
The subject stands in an upright posture, leaning with front paws against the cage-wall. Includes the rise and descend.
The subject rests with hardly any moving, either sits or is lying down. Includes sleeping. Apparently no interest in the environment.
The subject makes slight movements of the head, possibly with slight, discontinuous body displacement. Includes sniffing the air, the wall, the floor and other objects.
The subject makes sudden and short movements of the body or head. Includes body shake and head shake. Twitching is scored as a point event with no duration, and does not interrupt behavior states (for Rat Behavior Recognition only).
Twitching is defined as a point event, that is, an event marking a point in the time line, but with no duration.
The subject moves to another place, and hind legs move as well.
Note the difference between Walking and Movement: Both dependent variables Walking and Movement are based on displacement of the subject’s body point in the 2D space. However, Walking is based on many more video frames than Movement, and therefore Walking is a more accurate measure of walking behavior.
How to specify a behavior for Behavior recognition
1.Click the Add button next to the behavior you want to analyze.
2.Under Behavior decision method, select:
▪Default: Statistics will be calculated by using all samples scored as that behavior (no filtering based on the behavior’s probability).
▪Probability greater than: Statistics will be calculated by using only the samples for which the probability of that behavior is higher than a specific value. Select this value from the list (0-99%).
3.Under Behavior duration threshold, next to Exclude instances shorter than, enter how long the current state must last before it changes from the previous state In seconds). If the bout length passes this threshold, the previous samples in the bout are included in the new state. If the bout length does not pass the threshold, the previous state ends, but no new state is defined.
4.Under Calculate statistics for, select which state you would like to calculate statistics for.
5.Complete the procedure to add the variable. See Calculate statistics: procedure.
Accuracy
According to our tests, the average recall rate, that is, the proportion of ground truth manually-scored behaviors that are correctly recognized, is around 70%. For more information, see the paper mentioned below.
▪Drinking can be confused with Sniffing especially when the Drinking Spout Points are not defined.
▪Make sure to draw the Feeder Zones in the Arena Settings to maximize the detection accuracy of Eating.
▪To maximize accuracy of detection of Rearing supported and reduce false positives, carefully draw the Wall zone in the Arena Settings.
Notes
▪For Twitching and Hopping, which is an event with no duration, the Behavior duration threshold and the Calculate statistics for options are not available.
▪For more information on Behavior recognition in EthoVision XT, see the following paper:
van Dam, E., J.E. van der Harst, C.J.F. ter Braak, R.A.J. Tegelenbosch, B.M. Spruijt, L.P.J.J. Noldus (2013). An automated system for the recognition of various specific rat behaviours. Journal of Neuroscience Methods 218 (2), 214–224.
http://dx.doi.org/10.1016/j.jneumeth.2013.05.012.
▪The Behavior decision method acts as a filter to ignore samples for which the behavior is associated with a low probability, making it easier to extract more reliable data. By default, EthoVision XT uses the original scores (thus no filtering based on probabilities). See the figure below for an example.
Top: The probability of Grooming, and another behavior plotted against time (for simplicity, other behaviors are ignored here). Bottom: When defining Grooming in the Analysis profile, consider the following options:
▪With Default selected, Grooming occurs every time its probability exceeds the probability of any other behavior (default behavior recognition).
▪With Probability greater than 50% selected, Grooming is limited to the samples for which its probability is higher than 50%. The higher the value, the more conservative the definition.
▪The Behavior duration threshold acts as a filter to ignore short transitions between states. For example, if the behavior Grooming is scored for 0.8 s and you select a Behavior duration threshold of 1.0 s, the Grooming state does not exceed that threshold, and is therefore excluded from analysis.
▪For all behaviors but Twitching and Hopping, the percentage statistics are calculated including Unknown. For example, when Grooming lasts 30 seconds, Not Grooming 1 min 30 s, and Unknown 30 s, then the Cumulative Duration within Track (%) for Grooming is 30/150 = 20%. If you want to exclude Unknown from the percentage, in the Data profile make two nesting intervals, one with [behavior name] and the other with Not [behavior name], then combine the two Nest boxes in parallel. Next, in the Analysis profile specify the behavior and its statistic, Cumulative Duration within Nesting (%). In the example above, the result would be 30/120 = 25%.
▪The duration of the behaviors may not add up to 100% of the track duration, also when including Unknown. This could result, for example, when you filter a behavior using a probability threshold. When visualizing behaviors, samples filtered out appear as gaps between behaviors.
▪When you specify a Behavior probability threshold for one behavior, the other behaviors (which are mutually-exclusive by definition) are not re-calculated according to that setting. Depending on the value of the probability threshold, the behavior may become overlapping with other behaviors. For example, if you select a probability threshold of 15% for Grooming, some parts of the track may be scored as Grooming where also Eating was scored based on the default settings (see below). To prevent this from happening, either use Default or choose the same probability threshold for all behaviors.
Below: Event plots of behaviors Grooming and Eating. Top: Behavior probability. Middle: Grooming is defined with a Behavior probability greater than 15%. Bottom: Eating is defined with Default probability settings, and is scored for most of the time due to the high probability (see the probability plot). As a result, Grooming is partially overlapping with Eating, although the two behaviors are supposed to exclude each other. With default settings, Grooming would not be scored.
See also