Reproducibility and Sources of Variability in Radiographic Texture Analysis of Densitometric Calcaneal Images

Abstract 

Radiographic texture analysis (RTA) is a computerized analysis of the spatial pattern of radiographic images used as a way of evaluating bone structure. We have shown that RTA performed on high-resolution heel images obtained using a portable densitometer differentiates subjects with and without osteoporotic fractures. In the present study, short-term precision of RTA was examined on densitometric heel images obtained from 33 subjects scanned 8 times each, with 3 observers placing a region of interest (ROI) 3 times on each image. The long-term precision was examined on images obtained from 10 subjects 3 times on each of 3days separated by 1week, with 2 observers placing an ROI on each image. The RTA features examined included the root mean square (RMS) variation, a measure of the contrast between the light and dark areas of the image, the first moment of the power spectrum, a measure of the spatial frequency of the trabecular pattern, and Minkowski fractal (MINK), a measure of roughness/smoothness of the trabecular pattern. The precision of the RTA features expressed as coefficient of variation ranged between the lowest of 0.5–0.7% for MINK and the highest of 14–16% for RMS. The short- and long-term precision was similar, and was not significantly influenced by repositioning and rescanning, or by ROI placement by the same or different observers. Significant sources of variability of RTA were the between-subject differences and differences between regions of the heel, but not differences due to repositioning, rescanning in the same position, or ROI placement by the same or different observers. We conclude that technical aspects of image acquisition and processing are adequate to allow further development of RTA of the densitometric images for clinical application as a method for noninvasive assessment of bone structure.