Lung Segmentation Using Prediction-Based Segmentation Improvement for Chest Tomosynthesis

Abstract

Chest radiography which is the most common imaging method for lung is difficult to distinguish the lung vessels and nodules due to characteristics of representing the chest in a single image and shading occurred by organs. Computed Tomography(CT) scan has excellent lung nodule detection sensitivity because it produces chest images as volume data, but it has a large amount of exposure dose and is expensive. Chest tomosynthesis which generates volume data through continuous shooting comes to the forefront as an early lung cancer screening method with high lung nodule detection sensitivity than chest radiography and lowdose than CT image. However, chest tomosynthesis is difficult to have computer-based automatic segmentation because of blurring occurred while generating the image. Therefore, we propose prediction-based segmentation improvement method based on the central slices with less blurring after performing lung segmentation using region-growing. Using the proposed method, it is to improve the lung segmentation performance by improving the incorrect segmentation results on the outer slices where many blurring occurs. The experiment results showed the improvement of incorrectly segmented lung region.