A locally constrained statistical shape model for robust nasal cavity segmentation in computed tomography

Accurate segmentation of the nasal cavity plays a pivotal role for the creation of patient specific nasal models which are essential for the diagnosis and treatment planning of nasal-related disorders and diseases, as well as for research on nasal drug delivery. However, the structure of the nasal cavity is difficult to segment due to the lack of boundary distinction to other connected airway components, such as the paranasal sinuses, which exhibits the same intensity range. Existing algorithms fail at differentiating the cavity from its surrounding structures. This paper presents a new method at segmenting the nasal cavity, designed to differentiate between the nasal and non-nasal airway. Our algorithm makes use of a range of statistical knowledge of the complex nasal anatomy through the introduction of a robust multi-atlas initialization for seeds derivation and the incorporation of a similarity guided statistical shape model (SSM). Our approach combines the shape variation energy of the SSM together with a modified locally constrained random walk algorithm to segment the nasal cavity. Our proposed algorithm was evaluated on 20 CT images and outperformed comparative state-of-the-art and conventional algorithms.