A novel algorithm to distinguish sudden cardiac death subjects from other cardiac patients and healthy individuals

Abstract

Abstract For the timely diagnosis of sudden cardiac death (SCD), selecting accurate features and increasing the specificity of the diagnosis algorithms are essential. Therefore, the HRV signal of subjects who suffered from SCD was examined in the present study. The signal has been studied in one-hour duration before the incident to obtain significant signal changes in subjects' cardiac signals. In the proposed methodology, the patient's HRV signals are divided into 5 minutes segments. Each of these segments is decomposed into four sub-signals. Afterward, the corresponding energy and instantaneous amplitude of each sub-signal are determined. Subsequently, the transfer entropy between each pair of instantaneous amplitude signals and the sample entropy of energy sub-signals are determined. The segment representing a radical change in comparison to its previous segment is detected. A support vector machine (SVM) classifier is used to identify subjects exposed to SCD, based on the hypothesis that these radical changes can be recognized as indicators of the SCD process. This methodology has the advantage of not being limited to any particular subclass of cardiac diseases. The results represent 100% and 89.47% specificity respectively for healthy subjects and cardiac patients 15 minutes before the incident.