Automated Lung Segmentation Using Deep Leaning in Lung CT Images

Lung segmentation algorithms are crucial for accurately delineating the areas of infection within the lungs. The main aim of this research work is to generate a highly efficient and reliable deep learning model for lung segmentation. This will be achieved by utilizing chest computed tomography (CT) images and DeepLabV3+ networks. The model will be designed to classify the images into two classes (background and lung field). This work examines the efficacy of the DeepLab V3+network using three pre-trained networks: Xception, ResNet-18, Inception-ResNet-50, MobileNet-v2. The evaluation of the segmentation performance has been conducted using five performance metrics: Intersection of Union (IoU), Weighted IoU, Balance Fl score, pixel accuracy, and global accuracy. The empirical findings of this study validate that the DeepLabv3+ network, when coupled with ResNet-18 and a batch size of 8, exhibits superior performance in two-class segmentation. In the proposed work the mean accuracy and Mean IoU is 99.59% and 98.89% respectively. The automated segmentation model we have built can be enhanced to serve as a clinical diagnosis method for lung diseases.