TY - GEN
T1 - Power spectral analysis for identifying the onset and termination of obstructive sleep apnoea events in ECG recordings
AU - Khandoker, Ahsan H.
AU - Karmakar, Chandan K.
AU - Palaniswami, Marimuthu
PY - 2008
Y1 - 2008
N2 - The high prevalence of obstructive sleep apnoea (OSA) requires a simplified, unattended screening device that would be useful for diagnosis at the early stage. This study presents a method for screening individual OSA event based on sleep ECG signal. The overnight ECG recordings were divided into 5-second epochs containing normal (N) breathing and onset (O), maximum (M) & termination (T) of OSA events. Power spectral analysis of ECG epochs was employed to extract features. The area under receiver operating characteristics curve was estimated to determine the discrimination capability of each feature (or power in each frequency bin). The maximum ROC areas for N/O, N/M and N/T were found to be 0.78, 0.81, 0.71 in the ranges of powers of 57-65 Hz, 52-72 Hz, 52-66 Hz bands respectively. An heuristic rule was applied to recognize the individual OSA events from spectral features of N,O,M,T epochs. Results show good agreement with the original annotations in an overnight sleep study. These results, therefore, could have considerable potential in ECG based screening and can aid sleep specialist in the assessment of patients with suspected sleep apnoea syndrome.
AB - The high prevalence of obstructive sleep apnoea (OSA) requires a simplified, unattended screening device that would be useful for diagnosis at the early stage. This study presents a method for screening individual OSA event based on sleep ECG signal. The overnight ECG recordings were divided into 5-second epochs containing normal (N) breathing and onset (O), maximum (M) & termination (T) of OSA events. Power spectral analysis of ECG epochs was employed to extract features. The area under receiver operating characteristics curve was estimated to determine the discrimination capability of each feature (or power in each frequency bin). The maximum ROC areas for N/O, N/M and N/T were found to be 0.78, 0.81, 0.71 in the ranges of powers of 57-65 Hz, 52-72 Hz, 52-66 Hz bands respectively. An heuristic rule was applied to recognize the individual OSA events from spectral features of N,O,M,T epochs. Results show good agreement with the original annotations in an overnight sleep study. These results, therefore, could have considerable potential in ECG based screening and can aid sleep specialist in the assessment of patients with suspected sleep apnoea syndrome.
UR - http://www.scopus.com/inward/record.url?scp=63449127549&partnerID=8YFLogxK
U2 - 10.1109/ICECE.2008.4769180
DO - 10.1109/ICECE.2008.4769180
M3 - Conference contribution
AN - SCOPUS:63449127549
SN - 9781424420155
T3 - Proceedings of ICECE 2008 - 5th International Conference on Electrical and Computer Engineering
SP - 96
EP - 100
BT - Proceedings of ICECE 2008 - 5th International Conference on Electrical and Computer Engineering
T2 - 5th International Conference on Electrical and Computer Engineering, ICECE 2008
Y2 - 20 December 2008 through 22 December 2008
ER -