Automatic Detection of GGO Candidate Regions by Using Artificial Neural Networks from Thoracic MDCT Images

Abstract

Detection of abnormal areas such as lung nodule, ground glass opacity on multi detector computed tomography images is a difficult task for radiologists. It is because subtle lesions such as small lung nodules tend to be low in contras, and a large number of computed tomography images require a long visual screening times. To detect the abnormalities by use of computer aided diagnosis (CAD) system, some technical method for detecting the abnormalities have been proposed in medical field. Despite of these efforts, their approach did not succeed because of difficulty of image processing in detecting the ground glass opacity (GGO) areas exactly. Thus they did not reach to the stage of automatic detection employing unknown thoracic MDCT data sets. In this paper, we develop a CAD system for automatic detecting of GGO areas from thoracic MDCT images by use of five statistical features which are obtained four density feature and one of shape feature. The proposed technique applied on 31 MDCT image sets. 79.4 [%] of recognition rates and 1.07 of false positive rates was achieved. Some experimental results are shown along with a discussion.