Multiclass COVID-19 Detection using Attention-Based 3D CNN with Grey Wolf Optimization

Abstract

The emergence and spread of COVID-19 have raised a need for accurate and quick diagnosis because of the high infection rate. Towards this goal, a brand-new solution for multiclass COVID-19 detection is presented in this study using deep learning. The methodology is a synergy of feature extraction of an advanced MobileNetV2 model trained on the COVID-19 data set and the classification strength of attention-based 3D Convolutional Neural Networks. MobileNetV2 also is acting as a feature extractor obtaining detailed information from various COVID-19 imaging datasets. Thereafter, an attention-based 3D CNN architecture is used for the classification process to distinguish COVID-19 classes effectively. To fine-tune the parameters of the presented model and increase modeling performance, the Grey Wolf Optimization (GWO) algorithm is added to improve the convergence to the best solution. The second concept; spatial attention is applied in order to enhance the feature selectivity in addition to enhancing the classification outcomes. The experimental data also prove the efficiency of the developed approach, which shows high accuracy as well as sensitivity and specificity in the multiclass COVID-19 detection problems. The proposed hybrid framework, has a clear potential to enhance diagnostic power in healthcare thus using the disease management and control systems.