Coherence as a measure of noise in the ECG

TY - GEN

T1 - Coherence as a measure of noise in the ECG

AU - Struijk, Johmmes J.

AU - Graff, Claus

AU - Kanters, Jørgen K.

AU - Xue, Joel Q.

AU - Jensen, Ask Schou

AU - Schmidt, Samuel

PY - 2014/1/1

Y1 - 2014/1/1

N2 - The 12-lead ECG can be described as the projection of a time-varying dipole on the lead vectors. This dipole-source assumption implies that the different leads must show a high coherence and that deviations from values close to unity may indicate the presence of noise. We evaluated the average coherence in the band of 5 Hz to 45 Hz in 309 ECGs and compared the coherence measure with an automatic out-of-band noise evaluation, quantified at a scale from 0 (clean) to 10 (very noisy). The generalized magnitude squared coherence (GMSC), which is a global measure of the linear relationships among the leads, was used. A high coherence (GMSC> 0. 95) was always associated with very low noise (level 0 or 1) and 0.9<GMSC<0.95 was associated with levels 0-2. For noise levels > 2 the GMSC was <0.87 in all cases. There were no cases of high noise level and a high coherence, however, in 32% of the cases with noise level 0 or 1 the GMSC was <0.8. For low-noise ECGs the GMSC is close to unity, with a good correlation between noise level and GMSC for normal ECGs. However, a low coherence also seems to be associated with abnormalities in the ECG, which will be subject for further study.

AB - The 12-lead ECG can be described as the projection of a time-varying dipole on the lead vectors. This dipole-source assumption implies that the different leads must show a high coherence and that deviations from values close to unity may indicate the presence of noise. We evaluated the average coherence in the band of 5 Hz to 45 Hz in 309 ECGs and compared the coherence measure with an automatic out-of-band noise evaluation, quantified at a scale from 0 (clean) to 10 (very noisy). The generalized magnitude squared coherence (GMSC), which is a global measure of the linear relationships among the leads, was used. A high coherence (GMSC> 0. 95) was always associated with very low noise (level 0 or 1) and 0.9<GMSC<0.95 was associated with levels 0-2. For noise levels > 2 the GMSC was <0.87 in all cases. There were no cases of high noise level and a high coherence, however, in 32% of the cases with noise level 0 or 1 the GMSC was <0.8. For low-noise ECGs the GMSC is close to unity, with a good correlation between noise level and GMSC for normal ECGs. However, a low coherence also seems to be associated with abnormalities in the ECG, which will be subject for further study.

UR - http://www.scopus.com/inward/record.url?scp=84931299783&partnerID=8YFLogxK

M3 - Conference article

AN - SCOPUS:84931299783

SN - 0276-6574

VL - 41

SP - 37

EP - 40

JO - Computing in Cardiology

JF - Computing in Cardiology

T2 - 41st Computing in Cardiology Conference, CinC 2014

Y2 - 7 September 2014 through 10 September 2014

ER -