Feasibility studies of virtual laryngoscopy by CT and MRI-from data acquisition, image segmentation, to interactive visualization

Virtual endoscopy concept has been applied to study the larynx, as well as other hollow organs in recent years, assuming a clean lumen. In this work, we investigated the feasibility of virtual laryngoscopy by (1) studying currently available imaging protocols, (2) developing a suitable image segmentation method, and (3) constructing art efficient visualization system. By utilizing helical computed tomography (CT), the images for laryngeal volume can be obtained during a breath hold with 0.3 mm resolution. A fast pulse sequence using 1.5 T magnetic resonance (MR) imager can achieve 1 mm resolution within few minutes. The gain in tissue contrast on MR images is at the cost of resolution, and motion artifacts must be considered during image segmentation. A first-order Lagrange interpolation was applied to mitigate the reduced resolution, as well as partial volume effect and noise on the MR images. An automatic segmentation algorithm was adapted to extract the wall volume of the larynx. The algorithm considers local voxel property and classifies voxels based on the local property in the KL (Karhunen-Loeve) space. A visualization system was constructed for examining the mucosa and wall geometry with anatomical references in three dimensions. It navigates inside the lumen, as well as outside the larynx interactively with capability of inspecting and zooming into the regions of interest. It can also cut the larynx in any orientation to open the whole volume for viewing the entire inner surface. The procedure was tested on 2 volunteers and 2 patients. The segmentation performed consistently for all the studies and showed to be relatively insensitive to mild respiratory motion artifacts in the MR images. Image processing was accomplished within a few minutes on PC and low-end SGI platforms. These studies demonstrated the feasibility of virtual laryngoscopy for diagnosis of laryngeal abnormalities.