ResNet-EfficientNet powered framework for high-precision cough-based classification of infectious diseases

COVID-19 is a extremely contagious disease triggered by the SARS-CoV-2 virus which mostly affects the human breathing system. Furthermore, the COVID-19 was emerged in late 2019 and escalated rapidly into a global pandemic which impacted health and economic challenges across globe. Similar to other infectious diseases, it transmits through respiratory droplets and the rapid diagnosis is more important in controlling transmission and managing patient health care. In this work, a deep learning framework for COVID-19 classification using cough sounds has been proposed. Furthermore, the various deep learning models such as one-dimensional Convolutional Neural Network (1D-CNN), Depth-wise Convolutional Neural Network (DS-CNN), EfficientNet v2 and ResNet are utilized for the identification of normal and cough sounds produced by COVID-19. Also, the performances of all the deep learning models are analyzed using performance metrics such as accuracy, recall, precision, Matthews Correlation Coefficient (MCC), F1_Score and False Positive Rate (FPR). Results demonstrate that the performance of pre-trained models namely EfficientNet v2 and ResNet is better when compared to existing Deep Learning (DL) models. Additionally, the accuracy, precision, recall, F1_Score, MCC and false positive rate of ResNet is 98.5%, 98.99, 98, 0.9849, 0.9699 and 0.01 respectively shows that the ResNet is superior to the other models. The proposed work focus on the early intervention which helps physicians to isolate or treat patients which reduces transmission.