%0 PDF %T Semi-Automated Analysis of the Texture of Osteoarthritis in Trabecular Bone. %A Nochlin, Jason Lyons. %8 2005-06-20 %I Tufts Archival Research Center %R http://localhost/files/m326mc385 %X In digital images, texture refers to patterns in the intensity of pixels of the image. By ana- lyzing the texture of an image, it is possible to generate data about the subpixel structure of the image's subject. Specifically, in the problem of diagnosing the severity of osteoarthritis (OA) in the periartricular tibia (i.e., bone just under the knee joint), texture analysis can be used to quantify changes in the trabecular structure of the bone even when the trabeculae are not resolved in the MRI image. Texture-based techniques provide medical researchers a tool for understanding OA and how it changes over time. To explore the utility of texture techniques, a process was developed which predicts OA severity relative to an existing benchmark, the relative perarticular Bone Mineral Density (paBMD) as determined by dual-energy X-ray absorptiometry (DXA). This processed was developed and tested using a dataset consisting of knee MRI images from 50 patients. After exploring a variety of methods, a three step process was implemented. First, image segmentation is performed in order to identify the region of interest (ROI) in the image with minimal human input, a task previously performed entirely manually. Second, spatial filters are applied to each ROI in order to generate texture maps. A bank of 40 filters was used which consisted of two classes of filters: Laws texture energy masks and Gabor filters. Statistics are taken on the texture maps (mean, standard deviation, skewness, kurtosis, entropy). Third, the statistics are then used as features for a regression problem. To solve the regression problem, kernel-Support Vector Regression with a 4th degree polynomial was found to be the best performing algorithm. In the final analysis, the knee images of 39 patients were used in a cross-validation scheme. A prediction root-mean square error of 11% was obtained, suggesting that texture analysis techniques can be used as tools for analyzing trabecular structure in MRIs. %G eng %[ 2022-10-07 %9 text %~ Tufts Digital Library %W Institution