Predicting Three Types of Freezing of Gait Events Using Deep Learning Models

Abstract--Freezing of gait is a Parkinson's Disease symptom Naghavi et al. discovered that using The best performing model achieves a score of 0.427 One machine learning model uses time-series plantar pressure I. Each Freezing of gait (FOG) is a common Parkinson's disease PD patient is required to complete a 25-meter walking task, (PD) mobility disturbance that episodically inflicts PD patients during which a set of 16 features related to the center of with the inability to step or turn while walking. In advancing pressure coordinates, center of pressure velocities, center of stages of PD, 60% of PD patients could experience FOG pressure accelerations, and ground reaction forces is collected events [1]; each FOG event could last up to a few minutes. A 2-layer LSTM neural network FOG episodes often occur at the initialization of walking (start architecture and a 3-layer LSTM neural network architecture hesitation), turning, or during walking periods, during which show similar performance, achieving 82.1% mean sensitivity PD patients would experience dystonic gait during the "on" and 89.5% mean specificity and 83.4% mean sensitivity and state and hypokinetic gait during the "off" state of FOG [2]. However, plantar pressure insole sensors in to FOG, such as narrow passages, being time pressure, the research are for single use, which means that this detection distractions, dual-tasking, and male sex [3, 4] and actions system cannot generalize to larger scale experiments or reallife that could alleviate FOG, such as emotion, excitement, and detection systems [1].