Pathological Lung Segmentation in Chest X-Ray Images using Modified U-Net-Based Architecture

Lung segmentation is often considered to be a critical task in radiography and medical-imaging analysis as it enables the measurement of lung volumes and the detection of abnormalities such as fluid buildup, and inflammation in the lung tissues, tumours and nodules. This paper proposes a method for lung segmentation based on a type of convolutional neural network (CNN) called U-Net which is designed for biomedical image segmentation. The proposed method combines the U-Net architecture with Dilated Convolutions and Bi-linear Interpolation to improve the performance and robustness of the model. The proposed methodology has been trained and tested on publicly available datasets, outperforming existing lung segmentation methods with respect to both Dice and Intersection Over Union Scores. The obtained results substantiate the effectiveness of the proposed U-Net-based approach for lung segmentation.