A portable Raspberry Pi-based system for diagnosis of heart valve diseases using automatic segmentation and artificial neural networks

• Published in: Cogent engineering Vol. 7; no. 1
• Main Authors: Joukhadar, Abdulkader, Chachati, Louay, Al-Mohammed, Mohammed, Albasha, Obada
• Format: Journal Article
• Language: English
• Published: Abingdon Cogent 01.01.2020; Taylor & Francis Ltd; Taylor & Francis Group
• Subjects: Algorithms; Artificial neural networks; artificial neural networks (ANNs); automatic segmentation; Back propagation; Back propagation networks; Classification; Diagnostic software; Diagnostic systems; Disease; Domains; Feature extraction; Hardware; Heart; Heart valves; Neural networks; phonocardiogram (PCG) signal; Physicians; Raspberry Pi; Segmentation; Signal processing; Touch screens
• ISSN: 2331-1916; 2331-1916
• DOI: 10.1080/23311916.2020.1856757

This study proposes a Raspberry Pi-based system for the diagnosis of heart valve diseases as a primary tool to improve the diagnostic accuracy of physicians. The proposed system is able to detect and classify nine common valvular heart cases encompassing eight types of heart valve diseases as well as the normal case of valves. The design and development of the proposed system are mainly divided into two phases, namely development of a disease classification approach, and design and implementation of the diagnostic hardware system. The developed disease classification approach is comprised of five stages, namely obtaining phonocardiogram (PCG) signals, preprocessing, segmentation using a proposed automatic algorithm, feature extraction in three domains (time, frequency, and wavelet decomposition domains) and classification using a backpropagation neural network. The hardware of the diagnostic system consists of a PCG signal acquisition module connected to a processing and displaying unit, which is represented by a Raspberry Pi connected to a touch screen. Where the developed disease classification approach is implemented in the software of the Raspberry Pi to enable it to detect the diseases in real time and fully automatically. The proposed system was clinically tested on 50 real subjects encompassing the nine cases. The performance of the diagnostic system is obtained with an accuracy of 96%, sensitivity of 95.23%, and specificity of 100%.