Session P74.6

Dynamic Analysis of ECG Inter-Lead Transfer Properties for Detection and Categorization of Respiratory Events during Sleep

C Maier*, V Rödler, P Laguna, H Dickhaus

Heilbronn University
Heilbronn, Germany

This study deals with recognition of respiratory activity from the ECG with the final aim to delineate episodes of sleep apnea and categorize them into central, obstructive and mixed type. The surface ECG is modulated according to the variation of the thoracic impedance between heart and electrodes. Since the various types of respiration exhibit characteristic differences in intra-thoracic pressure and accumulation or displacement of air and blood, our hypothesis is that these cause anisotropic modification of the conductive properties. Although the true cardiac signal source is not available, it might be possible to estimate the temporal variation of anisotropy by dynamically analysing the transfer properties between pairs or sets of leads, and these might exhibit typical patterns which allow for differentiation of the type of respiration. To verify our hypothesis, we analyzed Holter ECGs (Mortara H12+, 8-channels, 1 kHz sampling rate per channel) registered simultaneously to polysomnograms (PSGs). From a total of 122 overnight recordings in 109 patients, we extracted 474 segments (duration 1-5 min) containing a wide variety of respiratory patterns. We implemented an adaptive RLS-algorithm which provides linear predictive models on a beat-to-beat basis, expressing each single ECG lead as weighted linear superposition of arbitrary other leads. The models’ transfer coefficients are estimated from signal windows of 150 ms duration centred at each QRS-complex, and their temporal variation is considered a mirror of the altered transfer properties. As two alternative and simpler methods to dynamically quantify inter-lead dependencies we considered, for each pair of ECG leads, first the relations between mean QRS amplitude modulus, and second, the area of the 2D-loop spanned by paired QRS-samples of two leads. First qualitative results indicate that all three approaches contain relevant information also on the type of respiratory activity. During central apnea, we generally find very low variability in all series. In the transitions between paradoxical and normal breathing, we observed phase shifts (out-of-phase vs. in-phase or vice versa) between QRS amplitude series in several cases, however in not confined to fixed lead pairs for different patients. Similarly, the sign of QRS-loop area change was altered between these breathing modes for some lead pairs. The transfer coefficient series also clearly reflect respiratory activity and changes in ventilation. Models of low order (1-5) with a history of 0-25 ms combining up to two or three leads appear suitable and sufficient. Currently, experts are annotating our data second-by-second for respiratory events. Statistical analysis of the annotated database will reveal whether there exist reproducible transfer-patterns between ECG leads which allow conclusions on the type of respiratory activity, and whether there is additional benefit from the transfer coefficients over loop-area or amplitude estimates.

(Abstract Control Number: 100)