Research On Automatic Segmentation Of 3D Inner Ear MRI Images Based On Statistical Shape Models

The inner ear segmentation can provide reliable anatomical information for clinical diagnosis and treatments,and can also provide the basis for the three-dimensional visualization and spatial orientation measurement of the inner ear,especially the standard spatial coordinate-based inner ear model is of guiding value for the theoretical research on the diagnosis and treatment of otolith disease.Because the gray intensity distribution of the MRI image of the inner ear is similar to the adjacent cerebrospinal fluid and the inner ear has complicated topological structures,the boundaries between which are difficult to distinguish.The traditional methods of image segmentation usually uses the information about the grayscale of the image and the prior information such as the anatomical position and the shape of the tissue and organ have not been fully explored,resulting in unsatisfactory results of automatic segmentation of the inner ear.To this end,an innovative inner ear automatically segmentation method has been proposed which is based on the combination of the 3D level set algorithm and the statistical shape models(SSMs).First of all,the SSMs is created by a set of training inner ear shapes which are segmented by otologists manually;Second,the region of interest(ROI)volume of the inner ear is determined by a rigid registration;Third,to realize the automatic locating of the initial contour,the mean shape model of the inner ear SSMs is used as an initial contour of level set which is projected to the ROI of the inner ear by a model to volume registration method;Finally,the 3D threshold level set segmentation algorithm is used to refine the inner ear segmentation.The segmented inner ears are visualized by surface rendering and compared with the manual segmentations by otologists.The experimental results demonstrated that the proposed method is capable of segmenting the inner ear automatically and the segmentation accuracy can achieve the clinical requirement.In addition,the problem of GPU-based data processing in Fourier domain optical coherence tomography system for real time imaging of human skin is also studied.