Mobile Gait Analysis using Personalised Hidden Markov Models for Hereditary Spastic Paraplegia Patients

Gait analysis provides a quantitative method to assess disease progression or intervention effect on gait disorders. While mobile gait analysis enables continuous monitoring in free living conditions, state of the art gait analysis for diseases such as hereditary spastic paraplegia (HSP) is currently limited to motion capture systems which are large and expensive. The challenge with HSP is its heterogeneous nature and rarity, leading to a wide range of ages, severity and gait patterns as well as small patient numbers. We propose a sensor-based mobile solution, based on a personalised hierarchical hidden Markov Model (hHMM) to extract spatio-temporal gait parameters. This personalised hHMM achieves a mean absolute error of 0.04 s ± 0.03 s for stride time estimation with respect to a GAITRite^® reference system. We use the successful extraction of initial ground contact to explore the limits of the double integration method for such heterogeneous diseases. While our personalised model compensates for the heterogeneity of the disease, it would require a new model per patient. We observed that the general model was sufficient for some of the less severely affected patients.