Recognition of Parkinson disease using Kriging Empirical Mode Decomposition via deep learning techniques

Parkinson's disorder (PD) is a chronic, irreversible neurological disorder that is hard to identify and manage.

In a clinical environment, doctors typically examine the gait irregularity using visual inspections and other indications to determine the gait disruption and significant symptoms of PD. The existing evaluation methods heavily rely on the doctors' knowledge and experiences, which might result in misinterpretation. Many previous studies use spatiotemporal features and monitoring systems to assist doctors in classifying PD.

Recent studies involve the decomposing techniques for the gait signals in order to lighten the dataset and computational time. In this paper, PD categorization from gait data is proposed using Kriging Empirical Mode Decomposition (KEMD) with several machine learning approaches and Deep learning techniques to estimate the accuracy of algorithms respectively. The outcome of the techniques were evaluated using accuracy, sensitivity and specificity.

The LSTM method produced promising results among the ML and DL techniques, with the highest classification accuracy of 99.10 %, and it outperformed compared to other methods.