Sudden Cardiac Death (SCD) is considered one of the main causes of mortality worldwide. Understanding the origin of this heart disease continues to be a challenge for the scientific community. T-wave alternans (TWA) is the term used to describe changes in the T wave's amplitude or shape that are seen. According to the literature review, T wave alternans has been considered an important, non-invasive indicator to detect and stratify the risk of sudden cardiac death. On the other hand, dictionary learning is a digital signal processing technique that allows identify the main characteristics of a signal using a sparse representation. In this context, a new non-invasive method is proposed by mixing TWA spectral methods and dictionary learning. The method identifies the main characteristics of ECG signal by obtaining a sparse representation that adapts a matrix (dictionary) in order to use it for highlighting the TWA characteristics and then use these characteristics for detecting SCD. Experimental results show an improvement of 32% compared to the Physionet TWAnalyser program by using synthetic data set and an improvement of 20% over public databases.

No abstract