Lung X-Ray Segmentation using Quadrant-Based Tracing Method

Abstract

Chest X-Ray is one of the most popular imaging modalities. Chest X-ray has been a subject of various imaging-related research for years. Among the various research, Lung segmentation is one of the most prominent ones.  Nowadays the trend of research in segmentation is moving toward deep learning however traditional segmentation has advantage of requiring less calculation resources thus still has potential to be explored. In this paper an alternative non-deep learning segmentation method using graph-based method to trace border of the Chest X-Ray lung region is proposed. Chest X-Ray image was treated as a graph with coordinate of the pixels as vertex and value of the pixels as edges. First the image was divided into 4 quadrants, then the border of lung region on each quadrant was traced by finding the minimum spanning tree of the graphs on each quadrant, then the pixels recorded as the tree was smoothed and optimized using Savitzky-Golay filter. The results were analyzed using the confusion matrix by comparing the proposed method results with manual segmentation by a radiologist. The proposed method is successfully segment lung area on lateral view of chest X-Ray with an average accuracy of 0.936. Two sample T-test also employed in order to show that there is no significant difference between the proposed method results and manual segmentation by radiologist.