Refinement of an Algorithm to Detect and Predict Freezing of Gait in Parkinson Disease Using Wearable Sensors

Freezing of gait (FOG) is a debilitating symptom of Parkinson disease (PD). It is episodic and variable in nature, making assessment difficult. Wearable sensors used in conjunction with specialized algorithms, such as our group's pFOG algorithm, provide objective data to better understand this phenomenon. While these methods are effective at detecting FOG retrospectively, more work is needed. The purpose of this paper is to explore how the existing pFOG algorithm can be refined to improve the detection and prediction of FOG. To accomplish this goal, previously collected data were utilized to assess the prediction ability of the current algorithm, the potency of each FOG assessment task(s) for eliciting FOG, and the maintenance of detection accuracy when modifying the sampling rate. Results illustrate that the algorithm was able to predict upcoming FOG episodes, but false positive rates were high. The Go Out and Turn-Dual Task was most potent for eliciting FOG, and the 360-Dual Task elicited the longest duration of FOG. The detection accuracy of the pFOG algorithm was maintained at a sampling rate of 60 Hz but significantly worse at 30 Hz. This work is an important step in refining the pFOG algorithm for improved clinical utility.