Freezing of Gait Detection in Parkinson’s Disease from Accelerometer Readings

摘要

Almost 10 million individuals are suffering from Parkinson’s globally. It is a neurodegenerative disorder whereby cells responsible for producing dopamine decrease in the substantia nigra segment of the brain. Dopamine is vital for movement as it transmits a signal from the brain to other parts of the body, a decline of which leads to the freezing of gait (FoG). The data we used were of wearable sensors which obtained acceleration in three directions that include axis at x, y, and z. A total of ten subjects suffering from Parkinson’s disease were taken who showed symptoms of the Freezing of Gait (FoG). From ten, only eight subjects showed FoG during research work. Three tasks were performed with wearable accelerometer sensors. The tasks included were straight path random path and daily routine walking. In this paper, we used a random path walking signal. The signal consisted of 6 parts, two sensors reading at the trunk, two at the shank, and two at the ankle. Strong discriminant features were extracted and fed to a classifier to detect normal accelerometer reading and FoG readings through the accelerometer. Bagged Trees showed the highest accuracy among all classifiers used for experimentation that is 90.4%.