Title: Hierarchical fusion in feature and decision space for detection of valvular heart disease using PCG signal

Authors: M.K.M. Rahman; Ainul Anam Shahjamal Khan; Tasmeea Rahman

Addresses: Department of Electrical and Electronic Engineering, United International University, Madani Avenue, Badda, Dhaka-1212, Bangladesh ' Department of Electrical and Electronic Engineering, Chittagong University of Engineering and Technology, Bangladesh ' Department of Electrical and Electronic Engineering, United International University, Madani Avenue, Badda, Dhaka-1212, Bangladesh

Abstract: Detection of valvular heart disease from phonocardiogram (PCG) signal is an important non-invasive and low-cost tool that can have a big impact on the healthcare market. We have developed two techniques namely 'weighted fusion of features in decision space (WFFDS)' and 'hierarchical fusion in feature and decision space (HFFDS)' that combined information from multiple feature domains to improve the disease-detection accuracy. We have shown that fusion of multiple features improve the detection-accuracy compared with individual features. The accuracy is further improved by WFFDS technique, where the fusion is performed in decision space instead of feature space. In WFFDS, classifiers of same type are trained on different feature sets and some weights are calculated from confusion-matrix, which are then used to combine information in decision space for classifying new data. In HFFDS, fusion is performed both in feature and decision space. Our experimental results corroborate that both WFFDS and HFFDS performs better than traditional representations of features and their straight-forward fusion.

Keywords: phonocardiogram; PCG; valvular diseases; neural network; feature fusion; decision fusion.

DOI: 10.1504/IJBET.2022.125576

International Journal of Biomedical Engineering and Technology, 2022 Vol.40 No.2, pp.184 - 204

Received: 17 Apr 2019
Accepted: 31 Mar 2020
Published online: 16 Sep 2022 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article