Abstract: T wave alternans detection is the basis of the quantification of T wave for an electrocardiogram (ECG) signal. According to the dynamical modeling and estimation theory, a nonlinear Bayesian filtering framework was proposed to approximate the T wave under impulsive noise with a novel dynamical model for the asymmetric T wave. Results of tests on realistic ECG data demonstrate that the proposed approach can effectively quantify the T wave on a beat-to-beat basis from a single ECG channel, enabling the T wave alternans quantification on a beat to beat basis with limited channels.