Classification of locomotor activity by acceleration measurement: validation in Parkinson disease

The number of steps per time period is an important ambulatory measure describing an individual's locomotor function with implications for psychological and physical health. Key applications in neurology, psychiatry, psychopharmacology, and sports, behavior or rehabilitation medicine make it desirable to improve step detecting devices. Several pedometer or wrist actigraphy monitors exist today, but are insensitive or confounded by movement style, which may vary for different diagnoses and applications. Presented is an algorithm that detects, classifies and counts steps related to walking, running and shuffling motion. Data is recorded using a novel ambulatory monitoring system (LifeShirt, VivoMetrics, Inc., Ventura, CA, USA) which captures breathing information from respiratory inductive plethysmography (RIP) sensors embedded in a light garment, and acceleration signals from a dual axis accelerometer attached close to the center of body mass. The vertical accelerometer axis measures upward acceleration generated by walking and running, while the other axis measures movement common with shuffling gait. Since these signals often contain noise and artifact due to soft tissue movement or external vibrations they are filtered and autocorrelated using unbiased estimates. The autocorrelation coefficients allow for clearer detection and classification of the cyclic motion during walking, running and shuffling movements. The algorithm is tested during various levels of exercise in healthy individuals and patients suffering from Parkinson disease, which is often characterized by shuffling gait. The results demonstrate an effective locomotor-monitoring algorithm that can produce accurate estimates of frequency and intensity of steps and shuffles and help classify daily locomotor activities.