Dynamic analysis of multi lead ECG recordings for detection and categorization of respiratory events during sleep

Our study compares three methods for extraction of respiratory information from single- and multi-lead ECG (QRS-area, Lead-pair loop-area/angle and vector-loop alignment). Polysomnograms and 8-channel Holter-ECGs were simultaneously registered in 90 patients. In 469 segments of 1-5 min duration (total duration: 1553 min), we assessed the similarity of ECG-derived signals with respiratory movements and air-flow by correlation analysis. Moreover, the accuracy for second-by-second detection of central apnea was determined by means of ROC-analysis. Our results show that multi-lead information can substantially improve the quality of ECG-derived respiration. Angle and area of the 2D-QRS-loop spanned by pairs of two leads provided the most consistent and robust estimates, especially in the presence of arrhythmia. Detection of central apnea from the median envelope of the lead-pair loop-angles was possible with 85% sensitivity and 89% specificity.