TY - GEN
T1 - Probabilistic Data-Driven Method for Limb Movement Detection during Sleep
AU - Cesari, Matteo
AU - Christensen, Julie A.E.
AU - Jennum, Poul
AU - Sorensen, Helge B.D.
PY - 2018
Y1 - 2018
N2 - Periodic limb movement disorder (PLMD) is a sleep disorder characterized by repetitive limb movements (LM) during night. The gold standard for LM detection consists of visual analysis of tibialis left (TIBL) and right (TIBR) electromyographic (EMG) signals. Such analysis is subjective and time-consuming. We here propose a semi-supervised and data-driven approach for LM detection during sleep that was trained and tested on 27 healthy controls (C) and 36 PLMD patients. After preprocessing of the EMG signals, discrete wavelet transform (Daubechies 4 mother wavelet and down to 4th decomposition level) was applied. EMG was reconstructed for each set of detail coefficients, thus obtaining four signals (DI-D4). The pre-processed EMG and DI-D4 signals were divided in 3-s mini-epochs of which traditional EMG features were calculated. Based on the assumption of lack of movements in healthy controls during rapid eye movement (REM) sleep, we used the features during REM of a subgroup of C to build a non-parametric probabilistic model defining the resting EMG distribution. This model was then used to classify the remaining mini-epochs as either resting EMG or LM. The percentages of 3-s mini-epochs with LMs were calculated for each subject and used to distinguish the remaining C and PLMD with a support vector machine and 5-fold cross validation scheme. Results showed that C can be distinguished by PLMD with accuracy higher than 82% in the preprocessed EMG and DI-D3 signals.
AB - Periodic limb movement disorder (PLMD) is a sleep disorder characterized by repetitive limb movements (LM) during night. The gold standard for LM detection consists of visual analysis of tibialis left (TIBL) and right (TIBR) electromyographic (EMG) signals. Such analysis is subjective and time-consuming. We here propose a semi-supervised and data-driven approach for LM detection during sleep that was trained and tested on 27 healthy controls (C) and 36 PLMD patients. After preprocessing of the EMG signals, discrete wavelet transform (Daubechies 4 mother wavelet and down to 4th decomposition level) was applied. EMG was reconstructed for each set of detail coefficients, thus obtaining four signals (DI-D4). The pre-processed EMG and DI-D4 signals were divided in 3-s mini-epochs of which traditional EMG features were calculated. Based on the assumption of lack of movements in healthy controls during rapid eye movement (REM) sleep, we used the features during REM of a subgroup of C to build a non-parametric probabilistic model defining the resting EMG distribution. This model was then used to classify the remaining mini-epochs as either resting EMG or LM. The percentages of 3-s mini-epochs with LMs were calculated for each subject and used to distinguish the remaining C and PLMD with a support vector machine and 5-fold cross validation scheme. Results showed that C can be distinguished by PLMD with accuracy higher than 82% in the preprocessed EMG and DI-D3 signals.
U2 - 10.1109/EMBC.2018.8512254
DO - 10.1109/EMBC.2018.8512254
M3 - Article in proceedings
C2 - 30440364
AN - SCOPUS:85055402125
T3 - Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBS
SP - 163
EP - 166
BT - 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2018
PB - IEEE
T2 - 40th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2018
Y2 - 18 July 2018 through 21 July 2018
ER -