A Comparative Approach for Investigating an Efficient Method for Reduction of Search Space in Lung Image Analysis

Majority of the lung disorders are diagnosed and treated using chest radiographs and volumetric CT scans. The experts are examining mainly the lung parenchyma for disease predictions and treatment. Though the radiology experts could decipher the lung boundary manually, it is very lengthy and susceptible to observer variations. The effective and automated segmentation of the region of interest (ROI) guarantees the speed and accuracy of clinical examinations. This study mainly focuses on investigating an automated and efficient lung segmentation (LS) technique that will effectively reduce the search space and improve the accuracy of clinical investigations. This study reports the comparative analysis of the traditional LS methods viz. segmentation using image processing and Watershed technique with the supervised learning strategy (SLS) using modified U-net. SLS outperformed the existing frameworks and achieves the highest segmentation accuracy of 99.7%.