Cross-Task Translation of Motor Recovery in Mouse Models of Stroke

Abigail Smith

Document Type

Article

Publication Date

2025

Keywords

JMG

Abstract

The mouse model has been used to study stroke recovery through a variety of behavioral assays, but results are not easily translatable across tasks. This project connects data from two such assays examining motor deficits after stroke: the string-pull task, in which a mouse pulls on a string to reach a reward, and the pasta-reach task, in which the mouse reaches through a glass window to retrieve a piece of pasta. DeepLabCut pose estimation and DeepEthogram behavior labeling algorithms are used to track the mouse’s movement from videos, and kinematic features including speed, acceleration, and joint angles were extracted from the data. We used machine learning to predict different phases of stroke (pre-stroke, acute, or chronic) and found that joint and wrist angles were universally important in making these predictions. We then created a framework for translation across the string-pull and pasta-reach tasks using CycleGAN, a generative and adversarial neural network. We found that angles of pull were most translatable across the two tasks. In the future, we hope to use this pipeline to translate shared behavioral features in mice to those in humans, since it is difficult to gather standardized data on human stroke.

Please contact the Joan Staats Library for information regarding this document.

COinS