Semantic segmentation of lungs using a modified U-Net architecture through limited Computed Tomography images

Abstract

Latest advancements in deep learning have led to an enthusiasm among biomedical researchers to explore the field of semantic segmentation further. Lungs segmentation plays a crucial role in the computer-aided diagnosis of several lung diseases. However, various anatomical varieties make lungs segmentation a challenging task. The main objective of our study is to propose a modified U-Net model that automatically segments the lungs from the computed tomography images. The proposed algorithm is trained on 240 training images. The advantage of this architecture is that it consumes less data and GPU memory. Experimental results show that the proposed architecture obtained 98.3% accuracy, 96.29% dice coefficient, and 93.63% Jaccard index. The segmentation model outperformed the original U-Net and the state-of-the-art methods. Thus, the modified U-Net model is apt for accurate lung segmentation.