Presentation Details
| Comparing Video-Based Methods for Spout Lick Detection Georgia Davis, Mia Fox, John Boughter, Max Fletcher. University of Tennessee Health Science Center, Memphis, TN, USA |
Abstract
Lick-based measures such as lick count, inter-lick interval (ILI), and lick burst structure are widely used to assess stimulus consumption. Traditional lickometers can accurately count licks, but fail to capture the animals’ orofacial movements, which can be evaluated as a measure of palatability. To address this limitation, we developed and evaluated video-based models for lick detection, assessing licking and other consummatory behaviors. Ventral-face videos were collected from head-fixed C57BL/6J mice during sucrose and quinine delivery. Videos were analyzed using a custom DeepLabCut model labeling 32 facial features. Spout licks were detected using three approaches: Identifying tongue-tip entries into a defined region of interest (ROI), a random forest classifier (RFC) trained on the mouth and tongue features, and an RFC trained on a reduced, decorrelated feature set of just the top lip and tongue. RFC detection outcomes were sensitive to facial feature selection, valuation of behavioral frames, and selection of lick probability thresholds. RFCs trained on fewer, less-correlated features showed closer agreement with lickometer measures than higher-dimensional models. Identifying licks using an ROI was comparable with other methods, but it works by using spatial recognition rather than feature determination. Results indicated that selecting a global threshold, applied across videos, was less accurate than determining a lick probability threshold per video. Temporal measures, including ILI distributions, were comparable across methods, although lower frame rates caused temporal binning. These results demonstrate that video-based lick detection is a model-dependent measurement and that feature selection, frame rate, and thresholding strategy critically shape lick detection outcomes.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, including photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author.
Content Locked. Log into a registered attendee account to access this presentation.