Bundle Branch Block Detection Models Based on Heart Rate Variability Parameters

Objective: Bundle branch block (BBB) is an arrhythmia disease. The left and right BBB (LBBB and RBBB) are the two most common types of BBB. The ECG features of BBB are very complicated. The computer-aided diagnosis (CAD) are urgently needed to improve the ECG-based diagnosis accuracy of BBB. Methods: ECG data is pre-processed to eliminate the baseline drift; R-peaks are located for extracting RR interval sequence; The characteristic HRV parameters are screened from thirty-three HRV parameters using the independent sample t-test; The characteristic HRV parameters are combined with support vector machine (SVM) to construct the models detecting BBB. Through the performance tests, the optimal detection model is obtained. Results: The optimal detection model of LBBB was composed of HR mean, HR SD, peakLF, αl and α2 + SVM, which accuracy rate and area under the curve (AUC) reached 88.89% and 0.977, respectively. The optimal detection model of RBBB is composed of SD1, Sampen and DET + SVM, which accuracy rate and AUC are 83.33% and 0.926, respectively. Conclusions: The optimal detection models of the LBBB and RBBB have better comprehensive advantages compared with other related methods. Thus, the proposed models has a good clinical application prospect.