ID 39804
Title Transcription
BP ニューラル ネットワーク オ モチイタ サーカディアン リズムゲン ノ システム ドウテイ
Title Alternative
BP neural netrworks approach for identifying biological rhythm source in circadian data fluctuations
Author
Nagashino, Hirofumi Department of Radiologic Science and Engineering, School of Health Sciences, The University of Tokushima Tokushima University Educator and Researcher Directory KAKEN Search Researchers
Kinouchi, Yohsuke Department of Electrical and Electronic Engineering, Faculty of Engineering, The University of Tokushima Tokushima University Educator and Researcher Directory KAKEN Search Researchers
Akutagawa, Masatake Department of Radiologic Science and Engineering, School of Health Sciences, The University of Tokushima Tokushima University Educator and Researcher Directory KAKEN Search Researchers
Cisse, Youssouf Department of Electrical and Electronic Engineering, Faculty of Engineering, The University of Tokushima|Department of Medicine, Faculty of Medicine, Laval University
Keywords
circadian rhythms
sleep-wake rhythm
system identification
neural network
moving average process
Content Type
Journal Article
Description
Almost all land animals coordinate their behavior with circadian rhythms, matching their functions to the daily cycles of lightness and darkness that result from the rotation of the earth corresponding to 24 hours. Through external stimuli, such as dairy life activities or other sources from our environment may influence the internal rhythmicity of sleep and waking properties. However, the rhythms are regulated to keep their activity constant by homeostasis while fluctuating by incessant influences of external forces. A modeling study has been developed to identify homeostatic dynamics properties underlying a circadian rhythm activity of sleep and wake data measured from normal subjects, using an MA (Moving Average) model associated with backpropagation (BP) algorithm. As a result, we found out that the neural network can capture the regularity and irregularity components included in the data. The order of MA neural network model depends on subject’s behavior. The last two data are usually dominant in the case without strong external forces. The adaptive changes of the dynamics are evaluated by the change of weight vectors, a kind of internal representation of the trained network. The dynamics is kept in a steady state for more than 20 days. Identified properties reflect the subject’s behavior, and hence may be useful for medical diagnoses of disorders related to circadian rhythms.
Journal Title
四国医学雑誌
ISSN
00373699
NCID
AN00102041
Publisher
徳島医学会
Volume
59
Issue
6
Start Page
304
End Page
314
Sort Key
304
Published Date
2003-12-25
Remark
EDB ID
FullText File
language
jpn
departments
Science and Technology
Medical Sciences