2026-09-18T19:04:32Zhttps://api.repository.cam.ac.uk/server/oai/request

oai:www.repository.cam.ac.uk:1810/3992412026-02-28T01:43:14Zcom_1810_213729com_1810_256065col_1810_219485

Methods for Comparative Gait Analysis in Heterogeneous Dog Populations https://doi.org/10.17863/CAM.127848 Mitchell, Lyndie Sutcliffe, Michael Allen, Matthew Vertebrate locomotion is a complex process utilising both the nervous and musculoskeletal systems. The study of gait can therefore provide significant insights concerning the health and function of the brain, spinal cord, nerves, muscles, bones, and joints. Clinical gait analysis can be used to understand, diagnose, monitor, rehabilitate, and design treatment plans for a variety of musculoskeletal and neurological abnormalities. Utilisation of gait analysis can be limited in comparisons of heterogeneous dog populations due to the significant differences in size and shape that can exist between different breeds and mixed-breeds of dogs. These differences create difficulties in determining what normal parameter ranges should be for a given dog. As a result, definitively detecting the presence and location of abnormalities in a dog’s gait is challenging, particularly in cases of subclinical lameness or multi-limb or multi-joint involvement. This thesis explores the use of machine learning and mechanical modelling methods to predict expected temporospatial gait parameter ranges in a heterogeneous dog population and identify the presence of gait abnormalities. A preliminary study validated the use of an instrumented treadmill system to enable rapid collection of kinetic and temporospatial gait data and improve repeatability of gait measurements through increased gait cycle sample size and control of velocity. The effects of various measurement qualities on the resultant parameter values were explored, and criteria for data inclusion and processing were determined. A case study of objective kinetic and temporospatial gait changes in a dog before and after a spinal cord injury was also found to suggest that methods for comparison between different breeds, sizes, and conformations of dogs in existing control group studies are not adequate. A scalable biomechanical model consisting of non-invasive morphometric measurements was used to inform machine learning models in the prediction of temporospatial gait parameter ranges and identification of gait abnormalities. This approach resulted in excellent models for prediction of expected ranges of step length, stride length, and paw contact surface area parameters in a heterogeneous dog population. Moderately successful predictive models for the hind reach, velocity, and cadence parameters were also developed. A preliminary classification model for identifying the presence of a gait abnormality was also successful in identifying all clinical lameness cases in the test data. These findings are a significant step in support of the development and implementation of machine learning methods to enable better objectivity and abnormality detection in canine gait studies. Further research is needed to expand and validate these models in a larger heterogeneous dog population for clinical implementation. University of Cambridge 2025-08-06 Thesis Doctoral Doctor of Philosophy (PhD) eng Gates Cambridge Trust https://www.repository.cam.ac.uk/handle/1810/399241 https://www.repository.cam.ac.uk/bitstreams/a129a6db-17f2-4091-8993-ae8050ad1df6/download e3e63c94e82c79400a70347bd1866757 https://www.repository.cam.ac.uk/bitstreams/7b6ae06d-a317-43f4-b55c-301dc113677b/download 51eb72a87015d564e24f66fbe966bb95 https://www.repository.cam.ac.uk/bitstreams/a94db49f-b838-402d-90d0-94ec22fb1f88/download 78c40d9a8a22fd4ed63c9e841188e08e https://www.repository.cam.ac.uk/bitstreams/02d3a704-9b23-40e2-acee-bce98ca8bd59/download d19c11422b23fff03dfbd76025c3eb1a https://www.repository.cam.ac.uk/bitstreams/4693bafd-8f82-4f77-a3d2-c10b6fcf8e72/download 87eda9de84448d1f82354d60eee3eb5f http://purl.org/NET/rdflicense/allrightsreserved Dog Gait Analysis Machine Learning