Detection of Freezing of Gait using Convolutional Neural Networks and Data from Lower Limb Motion Sensors

Abstract

<div>Freezing of Gait is the most disabling gait disturbance in Parkinson’s disease. For the past decade, there has been a growing interest in applying machine learning and deep learning models to wearable sensor data to detect Freezing of Gait episodes. In our study, we recruited sixty-seven Parkinson’s disease patients who have been suffering from Freezing of Gait, and conducted two clinical assessments while the patients wore two wireless Inertial Measurement Units on their ankles. We converted the recorded time-series sensor data into continuous wavelet transform scalograms and trained a Convolutional Neural Network to detect the freezing episodes. The proposed model achieved a generalisation accuracy of 89.2% and a geometric mean of 88.8%.</div>

Comparison of gait phase detection using traditional machine learning and deep learning techniques

“…Multi-layer structures of DL networks help extract higher-level features from the input signal. Examples of these networks are Multi-Layer Perceptron (MLP) [33], Convolutional Neural Networks (CNN) [34], Long Short Term Memory (LSTM) [35] and combined ones like CNN-LSTM [36]. However, the number of trainable parameters in these models is huge and therefore requires lots of training data to train all these parameters and develop an accurate model accurately.…”

“…Since these models are based on supervised learning methods, they require labelled data for training. This can limit the application as data must be collected in different controlled scenarios [34]- [36]. Alternatively, this study uses mathematical models to extract the desired labels from other relevant information existing in the data [37], allowing to utilise data that has been collected for other purposes.…”