TY - GEN
T1 - Unsupervised discrimination of motor unit action potentials using spectrograms
AU - Pham, Thuy T.
AU - Fuglevand, Andrew J.
AU - McEwan, Alistair L.
AU - Leong, Philip H.W.
N1 - Publisher Copyright: © 2014 IEEE.
PY - 2014/11/2
Y1 - 2014/11/2
N2 - Single motor unit activity study is a major research interest because changes of MUAP morphology, MU activation, and MU recruitment provide the most informative part in diagnosis and treatment of neuromuscular disorders. Intramuscular recordings often provide a more than one motor unit activities, thus MUAP discrimination is a crucial task to study single unit activities. Most neurology laboratories worldwide still need specialists who spend hours to classify MUAPs. In this study, we present a new real-time unsupervised method for MUAP discrimination. After automatically detect MUAPs, we extract features of spectrogram images from the wavelet coefficients of MUAPs. Unlike benchmark methods, we do not calculate Euclidean distances which assumes a spherical distribution of data. Instead, we measure correlation between spectrogram images. Then MUAPs are automatically discriminated without any prior knowledge of the number of clusters as in previous works. MUAP were detected on a real data set with a precision PPV of 94% (tolerance of 2 ms). We obtained a similar result in MUAP classification to the reference. The difference in percentages of MU proportions between our method and the reference were 3% for MU1, 0.4% for MU2, and 12% for MU3. In contrast, F1-score for MU3 reached the highest level at 91% (PPV at the highest of 96.64% as well).
AB - Single motor unit activity study is a major research interest because changes of MUAP morphology, MU activation, and MU recruitment provide the most informative part in diagnosis and treatment of neuromuscular disorders. Intramuscular recordings often provide a more than one motor unit activities, thus MUAP discrimination is a crucial task to study single unit activities. Most neurology laboratories worldwide still need specialists who spend hours to classify MUAPs. In this study, we present a new real-time unsupervised method for MUAP discrimination. After automatically detect MUAPs, we extract features of spectrogram images from the wavelet coefficients of MUAPs. Unlike benchmark methods, we do not calculate Euclidean distances which assumes a spherical distribution of data. Instead, we measure correlation between spectrogram images. Then MUAPs are automatically discriminated without any prior knowledge of the number of clusters as in previous works. MUAP were detected on a real data set with a precision PPV of 94% (tolerance of 2 ms). We obtained a similar result in MUAP classification to the reference. The difference in percentages of MU proportions between our method and the reference were 3% for MU1, 0.4% for MU2, and 12% for MU3. In contrast, F1-score for MU3 reached the highest level at 91% (PPV at the highest of 96.64% as well).
UR - http://www.scopus.com/inward/record.url?scp=84929455577&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84929455577&partnerID=8YFLogxK
U2 - 10.1109/EMBC.2014.6943514
DO - 10.1109/EMBC.2014.6943514
M3 - Conference contribution
C2 - 25569882
T3 - 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2014
SP - 1
EP - 4
BT - 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2014
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2014 36th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2014
Y2 - 26 August 2014 through 30 August 2014
ER -