Classification of spinal cord injury in rats by machine learning of gait parameters

Spinal cord injury (SCI) represents a critical neurological condition with high morbidity, significantly impacting sensory motor functions. This thesis introduces a novel approach to classify the time after SCI and rats which recieved neuromodulation therapy by using machine learning (ML) based analysis of gait parameters and locomotor scores. Utilizing data derived from video imaging of rat locomotion, this study evaluates the accuracy and feasibility of the approach compared to previous literature. Key parameters analyzed include stance duration, swing duration, stride distance, limb duty factor, and paw area, alongside traditional metrics like Basso, Beattie and Bresnahan (BBB) locomotor rating scale and Von Frey (vF) withdrawal thresholds. The integration of markerless pose estimation tool, DeepLabCut (DLC), allowed for detailed extraction of gait parameters, overcoming challenges associated with lateral plane-focused methodologies. By using images from lateral and bottom views, footsteps were associated with animals' body movement and the gait pattern was extracted by deep-learning methods. Gait parameters, BBB scores and vF thresholds were input as features to Ensemble Learning including various ML methods to predict time after injury and neuromodulation treatment. In animals with SCI, gait parameters and BBB scores from both fore-hindlimbs, allowed prediction of post injury class (6 end-points) with \%43 accuracy. The same feature set predicted treatment with \%71 accuracy. Results showed that the methodology can effectively differentiate between treatment groups and time after SCI. However, additional work is needed to improve accuracy. Parameter extraction by deep-learning also provides an accessible and cost-effective solution for SCI research.