Maryellen Giger PhD, PRESENTER: Nothing to Disclose

Abstract: HTML Purpose: We have been developing radiographic texture analysis (RTA) methods for the assessment of bone structure from radiographic images of the spine, hip, and heel. Presently, we have reported that the combined analysis of BMD and RTA on images obtained using a clinical densitometer for the calcaneus shows promise in differentiating patients with and without vertebral fractures. In this study, we present our latest methods for optimizing the preprocessing and texture features involved in RTA. Methods and Materials: Our database includes images of the calcaneus from 160 subjects (44 with and 116 without vertebral fractures) acquired with a peripheral densitometer system. In addition, BMD of the lumbar spine, femoral neck, and calcaneus are obtained. Radiographic texture analysis consists of (1) extracting a region of interest (ROI) from the digital radiographic image of the heel, (2) preprocessing the ROI of the trabecular structure, (3) characterizing the fractal nature of the trabecular pattern using a Baysian neural network, and (4) estimating the probability that a fracture exists. ROC analysis was employed to evaluate the performance of BMD and RTA in the task of distinguishing between patients with and without vertebral fractures. A resubstitution (self consistency) method was used to determine an upper bound in the performance and a leave-on-out jackknife method was used to determine the robustness of the method. Results: BMD alone yielded Az values (area under the ROC curve) of 0.59, 0.65, 0.54 for the spine, hip, and heel, respectively in the task of differentiating cases without and with a vertebral fracture. Use of preprocessing prior to texture analysis improved the fractal features in terms of Az from 0.55 to 0.78 (resubstitution analysis). Compared to BMD values alone, the new texture features (Az = 0.78, resubstitution; Az = 0.70, leave-one-out jackknife) were better than spine, hip, and heel BMDs with p-values all less than 0.05 in the resubstitution and with p-values of 0.16, 0.53, 0.08, for spine, hip, and heel, respectively, in the jackknife evaluation. Conclusion: In conclusion, RTA has the potential to perform as well as hip BMD (and better than heel BMD) in the task of predicting those at risk for vertebral fracture. Continued improvement is expected as larger databases are obtained. (M.G. is a shareholder in and received a grant from R2 Technology, Sunnyvale, CA.) Questions about this event email: m-giger@uchicago.edu

Giger PhD, M, Performance of Radiographic Texture Analysis (RTA) in Evaluating Patients with and without Vertebral Fractures.  Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL. http://archive.rsna.org/2003/3107356.html