DeepLab V3+ Based Semantic Segmentation of COVID -19 Lesions in Computed Tomography Images

Abstract

Abstract- Coronavirus 2019 spreads rapidly worldwide causing a global epidemic. Early detection and diagnosis of COVID-19 is critical for treatment as it causes respiratory syndrome appears in the chest medical images, such as computed tomography (CT) images, and X-ray images. The CT images are more sensitive and have more details compared to the X-ray images. Thus, automated segmentation plays an imperative role in detecting, diagnosing, and determining the spreading of COVID-19. In this paper, the DeepLabV3+ combined with MobileNet-V2 model was implemented. To validate this combination, we conducted a comparative study between the DeepLabV3+ variants by its combination with MobileNet-V2 against DeepLabV3+ combined with different CNN, namely ResNet-18, and ResNet50. Also, a comparative study with the basic traditional U-Net and modified Alex for segmentation was carried out. The experimental results showed the superiority of the using DeepLabV3+ combined with MobileNet-V2 for COVID-19 segmentation by achieving 97.5% mean accuracy, 95.2% sensitivity, 99.7% specificity, 99.7% precision, 99.3 % weighted Jaccard coefficient, and 97.5% weighted dice coefficient.

Keywords

• keywords- covid-19
• semantic segmentation
• deeplabv3+
• mobilenet-v2
• depth-wise separable convolution layer