Automatic Detection of Ground-Glass Opacity Shadows by Three Characteristics on MDCT Images

Abstract

Medical imaging systems such as computed tomography, magnetic resonance imaging provided a high resolution image for powerful diagnostic tool in visual inspection fields by physician. Especially MDCT image can be used to obtain detailed images of the pulmonary anatomy, including pulmonary diseases such as the pulmonary nodules, the pulmonary vein, etc. Image segmentation is an important problem in medical image processing and computer vision. In X-ray CT images, several approaches have been reported for segmenting lung cancer area. However, manual segmentation of the lung area may require several hours for analysis. Furthermore, MDCT images contain more than 300 slices. Therefore, manual segmentation method cannot apply for clinical application in the MDCT images. Ground-Glass Opacity (GGO) is defined as increased attenuation of lung parenchyma without obscuration of the pulmonary vascular markings on CT images. It is one of the important features in lung cancer diagnosis of computer aided diagnosis. If the area, occupied by GGO is large on the CT image, medical doctor can extract the GGO comparatively easily. However, the possibility to overlook the light gray shadow becomes higher when GGO exists as a small area. In this paper, we propose a new technique for automatic detection of GGO shadows by three characteristics on the thorax MDCT image. In the first step, we extract the region of interests in order to segment the lung area. To segment the lung area, we start preprocessing the CT slices by employing binarization, labeling, shrinking and expansion processing. In the second step, we calculate characteristics of GGO shadows such as mean value, standard deviation, and semi interquartile range. In the final step, the GGO shadow’s regions are extracted by linear discriminant function. The proposed technique was applied to 5 lung MDCT image sets (each case containing 210 to 250 CT slices) and all of the GGO shadows are detected correctly on the image sets.