Chest CT images analysis with deep learning algorithms for COVID-19 diagnostic for Iraqi center

Abstract

Since the spread of coronavirus disease 2019 (COVID-19), the efforts evolved to find fast and accurate diagnosis methods. The Computerized Tomography (CT) scan has proven its high efficiency in COVID-19 diagnosis. Meanwhile, the deep learning algorithm in computer-aided diagnosis (CAD) system developed largely. For that reason, deep learning models for automatic COVID-19 detection from CT images were proposed by many studies. In this study, a feature extracted with deep learning models included constructed Convolutional neural network (CNN) and pre-trained CNN (VGG-16) has been introduced, and it was compared with other handcrafted features extraction models in terms of the gray level co-occurrence matrix (GLCM) texture features. For classification, a multi-layer dense classifier was implemented for axial lung CT scans into two classes COVID-19 and Non-COVID-19. The dataset used in this work was locally collected from Ibn Al-Nafis Teaching Hospital-Baghdad-Iraq. On the relatively limited dataset the deep learning-based models including the crafted CNN and the VGG16 network performed with high classification accuracy where their accuracy results of (99.5 %), (86.1 %) respectively. These accuracy results as compared to the GLCM classification accuracy results of (64.1 %) approved the efficiency of deep learning algorithms in COVID-19 diagnosis. Finally, the proposed method might be a useful tool for early diagnosis to help the radiologist in the identification of COVID-19, especially in our Iraqi centers. These outcomes can be improved when larger datasets are available.