The Classification of Multiple Interacting Gait Abnormalities Using Insole Sensors and Machine Learning

In this work we investigate the effectiveness of a wireless in-shoe pressure sensing system used in combination with a type of machine learning referred to as long term short term memory networks (LSTMs) to classify multiple interacting gait perturbations. Artificially induced gait perturbations consisted of restricted knee extension and altered under foot centre of pressure (COP). The primary aim was to assess the capacity to diagnose gait abnormalities without the need to attend a gait laboratory or visit a clinical healthcare professional, through the use of technology. Ultimately, such a system could be used to autonomously generate therapeutic guidance and provide healthcare professionals with accurate up to date information about a patients gait. The results show that LSTMs are capable of classifying complex interacting gait perturbations using in-shoe pressure data. When testing, 11 of 12 perturbation conditions were correctly classified overall and 58.8% of all data instances were correctly classified (8.3% is random classification). This work illustrates that an automated low cost, non-invasive gait diagnosis system with minimal sensors can be used to identify interacting gait abnormalities in individuals and has further potential to be used in a healthcare setting.