Semi-automatic segmentation of paranasal sinuses from CT images using active contour with group similarity constraints

Abstract

Computerized tomographic (CT) scanning has dramatically improved the imaging of paranasal sinus anatomy as compared to sinus radiographs. Increasingly, subtle bony anatomic variations and mucosal abnormalities of this region are being detected. The morphological knowledge of nasal cavity and paranasal sinuses has an important clinical value. It is used for the detection of sinus pathologies, for determination of therapy, planning of endoscopic surgeries and for surgical simulations. Current research and industry assisting systems need a workspace definition of the paranasal sinuses, which is realized by segmentation. This paper presents a semi-automatic segmentation method for the paranasal sinuses which allows us to locate structures. In general, the traditional active contour methods like Snake, Levelset can resolve the CT images of paranasal sinuses normal without any anatomic variations caused by sinusitis. However, in the clinical practice, the diseased radiological image has more significances so that these classical methods can not work satisfied very well as the boundaries of sinuses has been covered by impurity inflammation produced. At this point, we proposed a novel method group similarity based on Low Rank to repair the lost part of the boundary. The experiment results proved that our proposed method outperformed conventional algorithms especially in abnormal images.