Fundamental study on creation of stimulus patterns with artificial neural network for functional electrical stimulation

Hajime Murakami, Tamotsu Machino, Takashi Watanabe, Ryoko Futami, Nozomu Hoshimiya, Yasunobu Handa

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

Functional Electrical Stimulation (FES) is very useful to restore motor functions of paralyzed extremities. It has been already reported that coordinated movements of the paralyzed upper limb of the quadriplegic patients could be restored by FES with electromyogram (EMG) -based stimulus patterns. The stimulus patterns based on EMG analysis reflect the cooperative activation of muscles. However it is necessary to measure EMG signals from many normal subjects every time when we try to restore a new movement. Hence we have studied a creating method of stimulus patterns for various movements. Stimulus patterns were created with the inverse model of the controlled object, i.e., the musculoskeletal system of the subject. Here we made Artificial Neural Network (ANN) learn the inverse model by the direct inverse modeling. Especially in this paper, the controlled object which has redundancy in an angle-stimulus voltage characteristic was considered. The direct inverse modeling method is not applicable to such redundant object. We introduced a constraint to obtain the angle-stimulus voltage characteristic which is not redundant. We controlled normal subject's wrist angles (angle of radial/ulnar flexion and palmar/dorsi flexion) with stimulus patterns created by this method. First, we measured stimulus voltage-angle characteristic of the controlled object, and obtained a forward model by making ANN learn the measured characteristic. Since the stimulus voltage-angle characteristic was many-to-one relationship, we decided one-to-one stimulus voltage-angle characteristic by using constraint. As the constraint, we minimized sum of normalized stimulus voltages to the muscles. It was expected that the muscle fatigue was suppressed by this constraint. Next, we made ANN learn the inverse model of one-to-one stimulus voltage-angle characteristic by the direct inverse modeling. Then we created stimulus patterns for four trajectories with the obtained inverse model (angle-stimulus voltage characteristic). The created stimulus patterns were applied to the muscles for wrist movements, and were confirmed to restore the desired movements. The results indicate the fundamental validity of the method.

Original languageEnglish
Pages (from-to)407-413
Number of pages7
Journaljapanese journal of medical electronics and biological engineering
Volume35
Issue number4
Publication statusPublished - 1997 Dec 1
Externally publishedYes

ASJC Scopus subject areas

  • Biomedical Engineering

Fingerprint Dive into the research topics of 'Fundamental study on creation of stimulus patterns with artificial neural network for functional electrical stimulation'. Together they form a unique fingerprint.

  • Cite this