Mobile Diagnosis of COVID-19 by Biogeography-based Optimization-guided CNN

Abstract

Since 2019, COVID-19 has profoundly impacted human health around the world. COVID-19 is extremely contagious, so fast automated diagnosis is necessary. In the field of COVID-19 detection, there are many studies based on convolutional neural networks (CNN). This article introduces the Biogeography-based Optimization (BBO) algorithm to tune three hyperparameters of CNN: \({\beta }_{1}\) for calculating the exponential decay rate of the past gradient, \({\beta }_{2}\) for calculating the exponential decay rate of the square of the past gradient and the learning rate \(\mathrm{\alpha }\). A mobile COVID-19 diagnosis application based on BBO-CNN is developed. The sensitivity of BBO-CNN is 94.46% ± 1.45%, the specificity is 93.72% ± 1.86%, the precision is 93.80% ± 1.64%, the accuracy is 94.09% ± 0.92%, the F1-score is 94.11% ± 0.88%, the Matthews Correlation Coefficient (MCC) is 88.21% ± 1.81%, and the Fowlkes-Mallows Index (FMI) is 94.12% ± 0.88%. Compared with six other deep learning-based state-of-the-art methods, BBO-CNN performs superior. BBO-CNN automates COVID-19 detection. The developed mobile diagnosis application helps to diagnose COVID-19 quickly in remote areas where radiologists are scarce.