Session S74.5

Principal Component Analysis for the Detection and Assessment of T-Wave Alternans

G Bortolan*, I Christov

ISIB-CNR
Padova, Italy

T wave alternans (TWA) is an electrophysiologic phenomenon associated with a risk factor of sudden cardiac death. In the framework of PhysioNet/Computers in Cardiology 2008 Challenge, a set of 100 ECG recordings were collected in order to test different algorithms for the detection and the quantification of TWA. The ECGs were preprocessed for power line interference, electromyographic noise and baseline wander suppression. QRS detection and T waves' delineations were automatically performed. Two kind of algorithm were tested, considering the temporal scale and frequency domain. 1) The T amplitude was tested with statistical analysis. 2) Principal component analysis (PCA) has been used for quantifying the complexity of the T waves and for the detection of possible alternans. In particular the PCA has been applied to the intervals of T waves, and the ratio of 2nd/1st eigenvalue is the parameter used for this analysis. Different methods for capturing beat-to-beat alternation of the T-wave parameters were tested. Global indices and local detection, considering a variable length window of RR intervals, were used. Non parametric statistical tests provided the significance of the detected differences.
Results: The Computers in Cardiology Challenge made available a score system in order to rank the various methods. The score takes values in the range -1, +1. Our study proved a high correlation between alternans of T amplitude and the alternans of T wave complexity index. For example the two indices were in agreement in 88% and 74% in the global and local detection respectively. The results of a preliminary algorithm combining the outcome of the two methods for the detection of TWA showed a score of 0.658.

(Abstract Control Number: 368)