Development of a novel transformation algorithm of electromyography and actigraphy signals based on electrical and kinetic energy: An application for sleep healthcare

Article type: Research Article

Authors: Kim, Jeong-Woo^{a; 1} | Kim, Hyo Jae^{a; 1} | Koo, Yong Seo^{a; b; *}

Affiliations: [a] Department of Neurology, Asan Medical Center, Seoul, Korea | [b] Department of Neurology, Korea University Anam Hospital, Seoul, Korea

Correspondence: [*] Corresponding author: Yong Seo Koo, Department of Neurology, Asan Medical Center, 88, Olympic-ro 43-gil, Songpa-gu, Seoul, 05505, Korea. Tel.: +82 2 3010 5920; Fax: +82 2 474 4691; E-mail: yo904@naver.com.

Note: [1] These authors contributed equally to this work.

Abstract: BACKGROUND: Although actigraphy is widely used to measure sleep quality, few studies directly compared actigraphy data with polysomnography data, especially electromyography data. OBJECTIVE: We developed an algorithm which transforms actigraphy and electromyography signals to verify the interchangeability between them and tested the utility of this algorithm in sleep healthcare. METHODS: Thirty-eight subjects underwent polysomnography and actigraphy. We transformed electromyography signals extracted from polysomnography as integrated electromyography (IEMG) and actigraphy signals as integrated acceleration (IACC) using their physical properties. We compared receiver operating characteristic (ROC) curves obtained from transformed datasets with those of raw datasets in distinguishing REM and non-REM sleep. RESULTS: There was no significant correlation between raw electromyography and raw actigraphy data (r= 0.001, p= 0.124). After applying our transformation algorithm, significant correlation between IEMG and IACC was shown (r= 0.392, p< 0.001). In order to overcome small adjusted R2 from simple regression model (adjusted R=2 0.153, p< 0.001), we used panel data regression model to correct individual variances (adjusted R=2 0.542, p< 0.001). In ROC curve for distinguishing REM and non-REM sleep, AUCs were 0.536, 0.735 and 0.729 in raw data, IEMG and IACC respectively. CONCLUSIONS: The transformation algorithm revealed the relationship between electromyography and actigraphy data, and also yielded improved sleep staging ability.

Keywords: Actigraphy, electromyography, panel data, polysomnography, sleep stage

DOI: 10.3233/THC-181527

Journal: Technology and Health Care, vol. 27, no. 3, pp. 243-256, 2019