Linear prediction modeling for evaluating abnormal intra QRS potentials in the high-resolution electrocardiogram

The purpose of this paper is to obtain abnormal intra-QRS potentials (AIQP) from a signal-averaged electrocardiogram (SAECG) have been proposed to indicate the risk of ventricular arrhythmias. However, the major limitation of current autoregressive moving average modeling is that the model order depends on the database. This study presented a new method based on the linear prediction modeling to improve the limits in AIQP analysis. A total of 154 normal Taiwanese (N), 94 ventricular premature contraction (VPC) patients and 26 sustained ventricular tachycardia (VT) patients were recruited. The AIQP were extracted from the modeling residual of a linear prediction model. From the analyses of all modeling residual curves (modeling residual versus model order), the optimal model order is six. The AIQP was quantified by the root-mean-square value of the modeling residual within QRS interval. The AIQP of VT patients were significantly greater than those of non-VT groups (normal and VPC groups) (p<0.05). No significant differences appeared between normal and VPC groups. A linear combination of AIQP in leads X, Y and Z and three standardized time-domain SAECG parameters provide the best diagnostic performance (specificity 85.9%, sensitivity 88.5% and predictive accuracy 86.2%). It is concluded that the AIQP can be extracted by the linear prediction modeling to evaluate the risk of ventricular arrhythmias, which can enhance the diagnostic performance of time-domain SAECG. And, the linear prediction modeling improves the clinical feasibility of AIQP analysis