Wi-Fi Channel-State-Information-Based Clustering of Freezing-of-Gait Episodes Caused by Parkinson's Disease Using Short-Time Autocorrelation Time-Series Features.

Because Parkinson's disease (PD) symptoms progress and worsen slowly over time, it is difficult to assess the states and the effects of medications based on short-term clinical diagnoses. Therefore, monitoring of patients at home plays an important role in improving the medical care of PD patients. This paper proposes and constructs a non-contact, unsupervised normal-gait/freezing-of-gait periods clustering system using Wi-Fi channel state information (CSI) based on short-time autocorrelation function (ST-ACF) as a method to continuously monitor symptoms of gait freezing, one of the main symptoms of PD. We use ST-ACF to extract robust periodic signals, followed by short-time complex-frequency principal component analysis (ST-CFPCA) to extract human motion features from the time-series ST-ACF, which is high-dimensional due to the CSI nature. Experiments demonstrate that, with unsupervised clustering, the system identifies the freezing-of-gait episodes with an accuracy of 95.3 %.Clinical relevance This proposed system using short-time autocorrelation function (ST-ACF) and short-time complex-frequency principal component analysis (ST-CFPCA) will realize continuous monitoring of Parkinson's disease at home by using widely available Wi-Fi radio waves, leading to daily checking of the progression of symptoms and the effectiveness of medication.