Lung Fields Segmentation Algorithm in Chest Radiography - Advances in Image and Graphics Technologies - page 138

imaging modality in lung cancer detection because of its low cost and low dose radia-
tion. In addition, the imaging equipment of chest radiography is simpler than other
imaging modalities. Since accurate segmentation of lung fields is the basis of auto-
matic detection of lung nodules, it has become one of the hotspots in the fields of
medical image processing. The segmentation algorithm of lung fields based on
the analysis of feature images was introduced by Xu et al. [5]. The top of lungs and
the contour of chest cavity were determined by the second derivative of contour, and
the right hemidiaphragm edges were determined by edge gradient analysis. The start-
ing points were determined based on a “standard rule” to search for the left hemi-
diaphragm edges. Ginneken et al. [6] used active shape models, active appearance
model and a multi-resolution pixel classification method for the segmentation of lung
fields. Shi et al. [7] proposed a new deformable model by use of both population-
based and patient-specific shape statistics to segment lung fields from serial chest
radiographs. Soleymanpour et al. [8] used adaptive contrast equalization and non-
linear filtering to enhance the original images. Then, an initial estimation of lung
fields was obtained based on morphological operations, and it was improved by grow-
ing this region to obtain the accurate contour. Zhenghao Shi et al. [9] modified the
conventional fuzzy c-means algorithm. Then the Gaussian kernel-based fuzzy
clustering algorithm with spatial constraints was used for automatic segmentation
of lung fields. Yan Liu et al. [10] proposed the lung segmentation algorithm based
on the flexible morphology and clustering algorithm. In our study, we established
an initial outline model and segmented the lung fields by use of gray and shape
similarity information in feature images. However, the initial position of lung
fields may be far away from the actual boundary of lung fields in some images.
During segmentation with gray and shape similarity information, the lung
outline cannot be covered by search area. Therefore, we modified the lung outline
based on Active Shape Model (ASM) [11] algorithm to obtain a better segmentation
result.

2 Establishment of an Initial Outline Model

The establishment of an initial outline model consists of three main sections: marking
the training set, aligning the training set, and establishing an initial outline model. The
process was described as follows.

2.1 Marking the Training Set

The points along the lung fields were used to mark the training set. The points in-
cluded the following three classes: 1) the points with specific application; 2) the
points without specific application, e.g. extreme points of the target in a specific di-
rection or the highest point of the curvature; 3) the points between 1) and 2).