Comparison of Two Deep Learning Models for the Recognition of Parkinson’s Disease Gait Patterns

In this paper, a method based on gait analysis and deep learning was developed and explored for the recognition of Parkinson’s disease (PD). The study comprised 33 patients with PD and 15 healthy subjects (CTRL). Two deep learning models were designed and compared: convolutional neural network (CNN) and long short-term memory network (LSTM). The networks were trained on data that was recorded using a wireless and wearable system comprised of inertial and force sensors while performing a gait task. The influence of different data lengths on the performance of models was also examined. The best result was obtained using a CNN model fed with 8 seconds long walking sequences, with an accuracy of 92.49% for the classification of all PD patients and CTRL subjects. The obtained results showed that methods based on deep learning and gait analysis could contribute to objective and automatized recognition of Parkinson’s disease with high accuracy.