Metric learning for Parkinsonian identification from IMU gait measurements

• Published in: Gait & posture Vol. 54; no. NA; pp. 127 - 132
• Main Authors: Cuzzolin, Fabio, Sapienza, Michael, Esser, Patrick, Saha, Suman, Franssen, Miss Marloes, Collett, Johnny, Dawes, Helen
• Format: Journal Article
• Language: English
• Published: England Elsevier B.V 01.05.2017
• Subjects: Adult; Aged; Aged, 80 and over; Algorithms; Female; Gait; Gait - physiology; Gait Disorders, Neurologic - diagnosis; Gait Disorders, Neurologic - physiopathology; Hidden Markov models; Humans; Inertial Measurement Unit; Machine Learning; Machine Learning algorithms; Male; Metric learning; Middle Aged; Orthopedics; Parkinson Disease - physiopathology; Parkinson's; Walking - physiology; Parkinson's; Inertial Measurement Unit; Gait; Hidden Markov models; Machine Learning algorithms; Metric learning; Machine Learning Algorithms; Hidden Markov Models; Metric Learning
• ISSN: 0966-6362; 1879-2219
• DOI: 10.1016/j.gaitpost.2017.02.012

•A metric learning approach to identify Parkinsonian gait from IMU data is proposed.•The approach learns the best classification strategy for the given training data.•Consequently, it can cope with larger cohorts with better generalisation power.•We achieve 85.51% accuracy over 580 subjects, the best yet over such large cohort. Diagnosis of people with mild Parkinson's symptoms is difficult. Nevertheless, variations in gait pattern can be utilised to this purpose, when measured via Inertial Measurement Units (IMUs). Human gait, however, possesses a high degree of variability across individuals, and is subject to numerous nuisance factors. Therefore, off-the-shelf Machine Learning techniques may fail to classify it with the accuracy required in clinical trials. In this paper we propose a novel framework in which IMU gait measurement sequences sampled during a 10m walk are first encoded as hidden Markov models (HMMs) to extract their dynamics and provide a fixed-length representation. Given sufficient training samples, the distance between HMMs which optimises classification performance is learned and employed in a classical Nearest Neighbour classifier. Our tests demonstrate how this technique achieves accuracy of 85.51% over a 156 people with Parkinson's with a representative range of severity and 424 typically developed adults, which is the top performance achieved so far over a cohort of such size, based on single measurement outcomes. The method displays the potential for further improvement and a wider application to distinguish other conditions.