Differential diagnosis of Parkinson and essential tremor with convolutional LSTM networks

Oktay A. B., Kocer A.

BIOMEDICAL SIGNAL PROCESSING AND CONTROL, vol.56, 2020 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 56
  • Publication Date: 2020
  • Doi Number: 10.1016/j.bspc.2019.101683
  • Journal Name: BIOMEDICAL SIGNAL PROCESSING AND CONTROL
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, EMBASE, INSPEC
  • Keywords: LSTM, Leap motion controller, Deep learning, Essential tremor, Parkinson's tremor, CNN, NEURAL-NETWORKS, RESTING TREMOR, DISEASE, PREVALENCE
  • Yıldız Technical University Affiliated: No

Abstract

This study aims to present a novel method for differentiation of Parkinsonian tremor (PT) and essential tremor (ET) using both postural and resting tremor. A convolutional long short-term memory (LSTM) that learns hand tremor recorded at both postural and resting positions is proposed for differentiation of PT and ET. 3D landmark points of hands at resting and postural positions are gathered using a Leap Motion Controller. A convolutional LSTM is trained for differentiation of ET and PT after preprocessing tremor data. The method is evaluated and tested on a dataset containing 40 subjects where 23 of them had PT and 17 of them had ET. The experiments showed that the accuracy of combined positions using both resting and postural tremor was higher than using single kind of positional tremor. The accuracy of tremor classification is 90% for the combined positions. (C) 2019 Elsevier Ltd. All rights reserved.