Akademik Veri Yönetim Sistemi
International Journal on Technical and Physical Problems of Engineering, cilt.17, sa.3, ss.1-10, 2025 (Scopus)
Accurately segmenting the lungs is a critical task in medical imaging as it plays an essential role in diagnosing and treating various pulmonary diseases. The complexity of anatomical structures and the presence of noise in medical images make lung segmentation a challenging task. Nevertheless, deep learning approaches have shown promise in previous decade to address the issue of lung segmentation. Among these methods, U-Net architecture has emerged as a popular choice due to its capacity to capture spatial information effectively. The principal aim of the paper is to create a dependable model for segmenting lung image scans, which could aid in the diagnosis of lung-related ailments, including tumors. The original contribution of this paper is its comprehensive analysis of U-Net++ architecture for lung segmentation based on Computed Tomography (CT) scans. In this article, a lung segmentation model based on the U-Net++ is presented which was performed on 267 lung CT scans. This study also offers an evaluation of U-Net, U-Net++ and ResUNet, outlining their advantages and disadvantages in clinical applications. U-Net++ gave a higher accuracy of 95% for test data compared to 74% using U-Net and 75% using ResUNet architectures. The U-Net++ approach accomplished a loss of 0.0510 on the validation set, much less than the other models. By presenting detailed engineering case studies and insights, this study demonstrates the practical value of U-Net++ in precise lung segmentation and underscores its potential for advancing pulmonary disease diagnosis and treatment.