The UC Davis multi-disciplinary Symposium on Biomedical Image Segmentation is intended to foster communications and collaborations between the various UC Davis groups involved in this important field.
CEO, Wiley gave a presentation on two projects he participated in while at the UC Davis Institute for Data Analysis and Visualization (IDAV), involving retinal imaging and facial reconstructive surgery.
Abstract: The acquisition speed of Fourier Domain Optical Coherence Tomography (FD-OCT) has increased so that three-dimensional (3D) volumetric data of human retinas can be captured in a clinical setting. Physicians take advantage of this technology by using software specially designed to process, display, and access quantitative 3D OCT data producing segmented volumes and thickness maps. We described our clinical FD-OCT system used to acquire 3D data from the human retina over the macula and optic nerve head. Individual slices (B-scans) are registered to remove motion artifacts and are then processed with custom 3D visualization and analysis software. Our analysis software includes 3D visualization methods along with a machine-learning support vector machine (SVM) segmentation algorithm that allows users to semi-automatically segment retinal structures and layers. Our software facilitates retinal layer thickness and volume measurements despite the presence of noise and structural deformation associated with retinal pathology. Our software has been tested successfully in clinical settings and has been successful in assessing 3D retinal structures in healthy as well as diseased eyes. Our tool facilitates diagnosis and treatment monitoring of retinal diseases.
As a second application we show preliminary work for computing eye socket volume for the purpose of facial reconstructive surgical planning. We capture an X-ray CT of the patient and place a seed point in the eye globe on a 2D slice through the eye and grow an elliptical snake out from this point until it encounters density values outside a certain tolerance from the seed point. Our preliminary work does this in 2D and we will extend this to 3D by growing a 3D surface using constraints maintaining elliptical curvature as well as only traversing through density similar to the seed point. This causes the expanding surface to stop at bone and air interfaces. When the growth process is finished, we compute the volume contained within the manifold surface using standard methods, for example, decomposing the space with tetrahedra and then summing the volumes of the tetrahedra.