Resilient embedded system for classification respiratory diseases in a real time

Abstract

Listening to lung sounds using a stethoscope is still one of the most important methods to diagnose respiratory diseases. These sounds are complex and challenging to diagnose, as even trained people may misclassify them. Accurate interpretation of these sounds requires excellent experience from the treating physician. For diagnosing respiratory diseases, sounds were analyzed, and various features were extracted for the proposed hierarchical design consisting of four layers. A random forest classifier was utilized for three layers and deep learning for the last layer. An FPGA implementation of the proposed respiratory processor is validated experimentally on soft and hard resources of the Virtex-5 ML506 FPGA board. Designing the system by field programmable gate array in a hierarchical manner that allows classification without completing all four stages. The resilient four-layer system achieved the highest average accuracy of 100 %, 99.83, 99.62, 99.88, and 99.87 for COPD, Healthy, URTI, Bronchiectasis, and Pneumonia diseases while saving both power consumption 63.8 % and 54.7 % of testing time.