The development, validation and demonstration of an automated rodent tracker and whisker detector

Quantitatively assessing behaviour to measure animal behaviour and motor control is challenging because there is a lack of unobtrusive behavioural models. Some studies have suggested that measuring whisker movements might be a good, quantitative behavioural model. However, whiskers are very thin, small and move very fast; and there is not yet an automated program that can detect whiskers in a fully-intact, freely-moving animal. Therefore, this thesis develops, validates and demonstrates a novel, fullyautomated rodent tracker and a whisker annotator, that simultaneously measures locomotion and exploration behaviours as well as whisker movements. The �rst step in designing an automated rodent tracker and whisker detector, is to extract a reliable ground truth from which to compare any tracked points to. Therefore, the Manual Whisker Annotator (MWA) was designed as a validator and calibrator for the subsequent trackers and detectors. The second step is to provide a reliable body and head contour. Therefore, the Automated Rodent Tracker (ART) was developed and validated, compared to a semi-automated tracker (Ethovision) and the MWA. Finally, a fully-automated whisker detector (FAWD) was designed and validated, using two existing semi-automatic whisker trackers (BWTT and Whisk) and the MWA. FAWD incorporates a variety of image-processing algorithms, including super sampling, dilation and subtraction and frangi �ltering to reliably detect whiskers. Both ART and FAWD were also successfully demonstrated on videos collected from SOD1 mice, a model of Amyotrophic Lateral Sclerosis, from day 30 to 120. The development of this software enables whisker movements and locomotion to be tracked in a repeatable fashion, and the fully-automated nature of the software means that many videos can be collected and quickly processed with minimal user input. This thesis develops and validates a suite of behavioural software that provides robust and quantitative measures of rodent behaviour for basic research or drug discovery. Future work will be to demonstrate this software on a larger range of rodent models of neurodegeneration, to further showcase the exibility and quantitative nature of this behavioural model.