Three classical machine learning classifiers are employed for classification of VCG signals into healthy control and BBB category. The maximum classification accuracy of 99.57% is achieved over Physikalisch-Technische Bundesanstalt diagnostic (PTBD) dataset by K-nearest neighbor classifier employing equal and squared-inverse Chebyshev distance with 10 and 20 neighbors. Moreover, with the proposed approach, we have achieved the sensitivity and specificity of 99.68% and 99.18%, respectively. The proposed approach thus has the potential to be used in real time clinical setting for accurate detection of BBB.

Bundle branch block (BBB) refers to cardiac condition that causes a delay in the path of electrical impulses, which makes it difficult for the heart to pump blood efficiently throughout the body. Early diagnosing BBB is important in cases where prior heart anomalies exist. Generally, the 12-lead electrocardiogram (ECG) is used to detect the BBB. To ease the ECG recording procedure, vectorcardiography (VCG) has been proposed with three leads ECG system. Manual diagnosis of BBB using ECG is subjective to the expertise of the doctor. To facilitate the doctors, in the present study, we have proposed a novel framework to automatically detect BBB from VCG signals using multivariate Fourier–Bessel series expansion-based empirical wavelet transform (MVFBSE-EWT). The MVFBSE-EWT is applied over the three channels of VCG signal, which results in the varying number of multivariate Fourier–Bessel intrinsic mode functions (MVFBIMFs). To process further, first six number of MVFBIMFs are selected due to their presence in the entire dataset. Each MVFBIMF is represented in higher dimensional phase space. From each phase space trajectory, fractal dimension (FD) is computed with three scales. The feature space is reduced with metaheuristic feature selection algorithm.