Coherent Unwrapping and Segmentation of the Spiral Cochlea
Abstract
Volumetric Optical Coherence Tomography and Vibrometry (VOCTV) were recently developed to record very sensitive spatially resolved vibratory measurements in the unopened mouse cochlea. It is a challenge to do quantitative analysis and segmentation of the volumetric data with the spiral structure of the cochlea. Here I present a newly developed technique to label the soft tissues, including the Reissner’s membrane, tectorial membrane, and basilar membrane by unwrapping the spiral cochlea coherently thus “linearizing” the volume image. The acquired volume is rotated by hand so that the cochlea has an axis parallel to the z-axis of the Cartesian coordinate system. The 3D spiral is used to estimate the location of the spiral limbus by calculating the spiral’s radius and height. The parameters of the model are determined by the given points from the actual vertices of the spiral limbus. Then, the frame is resampled around the estimated point and stacked to generate a new linearized cochlea. The linearized cochlea is segmented and points are selected from the segmented mask by algorithm. With the selected points, polygons are generated for labeling. Selecting points, however, can be challenging if the SNR is low or the segmentation result is not suitable to automatically select points. In this case, linearization is beneficial since the regions to be labeled can be found by interpolating points. After labeling, the labeled regions are rewrapped to observe the region at the original volume image. Since the frames are not continuously extracted, the rewrapped volume is made to be continuous by closing operation. This algorithm can be further developed by segmenting and labeling the organ of Corti with detail, and automatizing the rotation and the model production.
Citation
Lee, Da Saem (2016). Coherent Unwrapping and Segmentation of the Spiral Cochlea. Master's thesis, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /187416.