Application of image segmentation and adaptive interpolation techniques to 3D reconstruction of the human temporal bones

Abstract

Three dimensional models aid otolaryngologists in understanding the complex anatomical features of the human temporal bone. Many of these models are generated by reconstructing histological sections. The goal of this thesis is to provide improvements on these existing 3D reconstruction methods. Presented are a segmentation framework and a contour finding algorithm for histological slices, followed by Gaussian filtering and error analysis. An adaptive interpolation algorithm based on monotonic piecewise cubics is used to automatically generate missing anatomical structure. Part of the algorithm development was completed on CT scans with proposals for extension to histological slices. The contour finding and Gaussian filtering algorithms output valid data points for interpolation. The adaptive interpolation algorithm produces satisfactory results with the interpolation error for the malleus being 1.80% when half of the data is used. The equivalent 3D model volume difference was 0.24%.