A combination approach to frequency domain analysis of short-time clinical electrocardiogram recordings

Summary form only received as follows: Electrocardiogram interpretation seems to be well understood in time domain compared to frequency domain. Minor rhythm disturbances in P wave and T wave characteristics are not easily visible in single beat spectral estimation methods. The dominance of QRS spectral power over the other segments in an ECG beat adds to the problem. In the present: work, each beat is modeled as a linear combination of the normal sinus rhythm and an arrhythmia component. Normal sinus rhythm is simulated based on the amplitude and interval calculations of each subject's observed EGG. The arrhythmia component is the difference between the observed beat from normal sinus rhythm. The resulting arrhythmia component is modeled using AR spectral estimation methods using Burg's algorithm. In this model based spectral estimation, four different spectral estimations are attempted after subtracting individual components like P wave, QRS complex, T wave apart from subtracting the entire simulated normal sinus rhythm. The present methodology is tested on a group of 80 standard MIT-BIH data segments and 34 clinical data of 6 seconds duration. Preliminary results of the spectral estimations seems to distinguish the common classes of ECG beats like NSR, PVC, APC, ST segment/T wave abnormalities, LBBB and RBBB. Features extracted from these spectral estimations show high sensitivity to even small rhythm disturbances in time.