AbstractTime-to-fall off an accelerating rotating rod (rotarod) is widely utilized to evaluate rodent motor performance. We reasoned that this simple outcome could be refined with additional measures explicit in the task (however inconspicuously) to examine what we call movement sub-structure. Our goal was to characterize normal variation or motor impairment more robustly than by using time-to-fall. We also hypothesized that measures (or features) early in the sub-structure could anticipate the learning expected of a mouse undergoing serial trials. Using normal untreated and baclofen-treated movement-impaired mice, we defined these features and automated their analysis using paw video-tracking in three consecutive trials, including paw location, speed, acceleration, variance and approximate entropy. Spectral arc length yielded speed and acceleration uniformity. We found that, in normal mice, paw movement smoothness inversely correlated with rotarod time-to-fall for the three trials. Greater approximate entropy in vertical movements, and opposite changes in horizontal movements, correlated with greater first-trial time-to-fall. First-trial horizontal approximate entropy in the first few seconds predicted subsequent time-to-fall. This allowed for the separation, after only one rotarod trial, of different-weight, untreated mouse groups, and for the detection of mice otherwise unimpaired after baclofen, which displayed a time-to-fall similar to control. A machine-learning support vector machine classifier corroborated these findings. In conclusion, time-to-fall off a rotarod correlated well with several measures, including some obtained during the first few seconds of a trial, and some responsive to learning over the first two trials, allowing for predictions or preemptive experimental manipulations before learning completion.
Revisiting the volatility of bitcoin with approximate entropy
