Machine learning-based gait anomaly detection using a sensorized tip: an individualized approach

Abstract

Lower limb motor impairment affects greatly the autonomy and quality of life of those people suffering from it. Recent studies have shown that an appropriate rehabilitation can significantly improve their condition, but, for this purpose, it is essential to know the patient’s functional state and to be able to detect any changes that occur in it as soon as possible. Traditionally, standardized clinical scales have been used to make that assessment, however, as the number of patients to be assessed is high, assessment frequency is usually low. In response to this problem, the aim of the present work is to design a new personalized methodology for developing a Machine Learning-based gait anomaly detector that is able to detect significant changes in the functional state of patients based on data provided by a sensorized tip; a system that will serve as support for the therapist who is treating the monitored patient’s case. Taking into account the variability that exists among patients, the proposed design focuses on an individualized approach, so that the system characterizes the state change of each patient case only on his/her own data. Once developed, the proposed methodology has been validated in ten healthy people of different complexions, achieving an average accuracy of 87.5%. Finally, five case studies have been analyzed, in which data from five multiple sclerosis patients have been captured and studied, obtaining an average accuracy of 82.5%.