Detection of Freezing of Gait Using Unsupervised Convolutional Denoising Autoencoder

At the advanced stage of Parkinson’s disease, patients may suffer from ‘freezing gait’ episodes: a debilitating condition wherein patient’s “feet feel as though they are glued to floor.” The objective, continuous monitoring gait disease with wearable devices has led development many freezing detection models involving automatic cueing rhythmic auditory stimulus shorten or prevent episodes. use thresholding and manually extracted features feature engineering returned promising results. However, these approaches subjective, time-consuming, prone error. Furthermore, their performance varied when faced different walking styles patients. Inspired by state-of-art deep learning techniques, this research aims improve model proposing denoising autoencoder learn salient characteristics Parkinsonian data that is applicable for elimination handcrafted features. Even features, reduction in half window sizes 2s, significant dimensionality learned still managed achieve 90.94% sensitivity 67.04% specificity, which comparable original Daphnet dataset research.