Session PB5.1

Support Vector Regression Model for Assessing Respiratory Effort during Central Apnea Events Using ECG Signals

AH Khandoker*, M Palaniswami

The University of Melbourne
Melbourne, Australia

Central sleep apnea (CSA) is recognized when respiratory effort falls below 15% of pre-event peak to peak amplitude of the respiratory effort. The aim of the present study is to investigate whether wavelet based features of ECG signals during CSA can act as surrogate of respiratory effort measured by respiratory inductance plethysmography (RIP). Therefore, RIP and ECG signals during 125 pre-scored CSA events and 10 seconds preceding the events were collected from 12 patients. Wavelet decompositions of ECG signals up to 20 levels were used as input to the support vector regression (SVR) model to recognize the RIP signals. Using leave-one-event out cross validation, an optimal SVR (linear kernel; C=2^10; e=2^-13 where C is the coefficient for trade-off between empirical and structural risk and e is the width of e-insensitive region)showed that it correctly recognized 115 CSA events (92% detection accuracy) using a subset of selected combination of wavelet decomposition levels (level 8, 9 and 10; 0.13-0.52 Hz) of ECG. Results suggest superior performance of SVR using ECG as the surrogate in recognizing the fall of respiratory effort during CSA. Results also indicate that ECG features could act as a potential surrogate signal of respiratory effort during normal as well as sleep disordered breathing.

(Abstract Control Number: 70)