ABSTRACT
The recent pandemic COVID-19 had invaded all the countries without any exceptions. The main reason of the virus to become pandemic is its intensity is infection contagiousness. Hence, rapid diagnosis is to be carried out to prevent the infection from spreading. As a rapid diagnosis framework, this research proposes an Artificial Intelligence (AI) based framework for identifying the presence and the severity of COVID-19 with the help of chest CT slices of the subject. This paper proposes a new AI severity classifier called Atom Search Optimized Gated Recurrent Units based Neural Network (ASO-GRU-NN), which is a GRU model with recurrent weights optimized by ASO algorithm. This model is trained for severity classification using different feature extraction techniques on the infection segmented images. The feature extraction phase aims to estimate the statistical features, gray-level co-occurrence matrix (GLCM) features, and region features of the identified infected areas in the lungs. The feature vectors are given as one feature group and as combinations of three to assess the performance of the ASO-GRU-NN model for the identification and the severity classification of COVID-19. The performance of the classification model trained with different features are analyzed and compared with existing models with the help of various performance measures such as accuracy, error rate, sensitivity, specificity, and execution time.
COVID-19, Feature extraction, Gated recurrent unit, Gray-level co-occurrence matrix, Severity classification
