Lung Diseases Segmentation Using Deep Learning Algorithms

Abstract

Lung diseases pose signifi cant challenges in diagnosing and treating patients, necessitating ac-curate and effi cient analysis of lung images to enable early detection and effective care. This research aims to address the labor-intensive and time-consuming process of detecting lung ab-normalities by employing deep learning-based segmentation algorithms. The objective is to evaluate and compare the performance of advanced models, including U-Net, U-Net++, U-Net 3+, ResU-Net, and Attention U-Net, in automating the identifi cation and delineation of lung abnormalities. The models were trained from scratch, and data preprocessing techniques such as contrast limited adaptive histogram equalization and data augmentation were implementedto enhance the results. The fi ndings consistently demonstrate that the U-Net 3+ architecturesurpasses the other models, exhibiting superior accuracy in segmentation. It achieved state-of-the-art performance with a dice score of 87.61% in COVID-19 segmentation and 89.45in lung tumor segmentation. These results showcase the potential of U-Net 3+ as a promisingsolution for automating the detection of lung abnormalities.Keywords: Image segmentation, Deep Learning, COVID-19, Lung Tumor, CT scans.