A Comparative Analysis of Deep Learning Architectures for Segmentation in Lung

Abstract

This study explores the application of deep learning techniques to segment lung computed tomography (CT) scans, with a focus on cases involving COVID-19 and lung tumors. Utilizing a diverse dataset encompassing a wide range of CT scans, we conduct an extensive evaluation of various state-of-the-art deep neural network architectures. Our experimental results demonstrate the high efficiency and accuracy of deep learning models in performing image segmentation tasks, achieving impressive dice scores of 95.12% and 82.89% on COVID-19 and lung tumor data, respectively. These findings highlight the signif-icant potential of deep learning in medical imaging applications. Furthermore, we conduct thorough ablation studies, meticulously analyzing the performance of each network architecture. These studies provide valuable insights into the specific strengths and limitations of different deep learning approaches, facilitating the identification of the most effective methods for lung CT scan segmentation. This research not only underscores the promising capabilities of deep learning in medical image analysis but also offers a detailed understanding of how various models can be optimized to enhance performance in clinical applications.