A Machine Learning Approach to Automated Gait Analysis for the Noldus Catwalk System.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Gait analysis in animal models is crucial for understanding in vivo compound effects in preclinical drug development.
• Automated analysis of gait data can significantly enhance the efficiency and accuracy of these studies.

Purpose of the Study:

• To establish a computational gait analysis approach for the Noldus Catwalk system.
• To enable automatic capture and storage of footprint data for advanced analysis.

Main Methods:

• Developed a machine learning pipeline including step decomposition, feature extraction, and multivariate sequence alignment.
• Employed gradient boosting, random forest, and elastic net classifiers for gait pattern discrimination.
• Utilized animal-wise leave-one-out cross-validation for robust performance assessment.

Main Results:

• Successfully differentiated movement patterns between a Parkinson's disease model and control groups.
• Accurately predicted lesion time points, types, and affected brain regions.
• Provided in-depth feature analysis for classifier interpretation.

Conclusions:

• Established a machine learning method for automated Noldus Catwalk gait data analysis.
• Demonstrated the capability of machine learning to discern pharmacologically relevant animal groups via gait.
• Highlighted the potential for continuous learning and prediction for individual animals in future studies.