Chest CT Images Analysis with Deep-Learning and Handcrafted Based Algorithms for COVID-19 Diagnosis
Abstract
After the outbreak of coronavirus disease 2019 (COVID-19), researchers have been working to develop fast and reliable diagnostic methods. The Computerized Tomography (CT) scan has shown to be highly effective in the diagnosis of COVID-19. Meanwhile, Computer-Aided Diagnosis (CAD) systems in terms of deep learning algorithms have progressed significantly. As a result, deep learning models for automatic COVID-19 detection from CT images were proposed. This study proposed Deep-learning, handcrafted, and Hybrid models for classification of axial lung CT-scans into two groups (COVID-19 and NonCOVID-19). Convolutional Neural Network (CNN) and Gray Level Co-occurrence Matrix (GLCM) act to extract Deep learning and Handcrafted features respectively, also a hybrid features (CNN and GLCM ) model was constructed to combine the effect of two models. For classification a dense multi-layer classifier was introduced. Finally to further enhance model performance a feature selection algorithm was employed. For training and testing of the models a Dataset collected for Iraqi patients from Ibn Al-Nafis teaching hospital was collected. The results obtained approved that the combined features extracted significantly improved the classification performance, with maximum classification accuracy achieved by the CNN+GLCM (with ANOVA feature selection) of 99.3% for the collected dataset.