An investigation of segmentation methods and texture analysis applied to tomographic images of human vertebral cancellous bone

The goal of this study is to determine architectural and textural parameter s on computed tomographic (CT) images, allowing us to explain the mechanica l compressive properties of bone. Although the resolution (150 mu m) is of the same order of magnitude as the trabecular thickness, this method enable s the possibility of perfecting an in vivo peripheral CT system with an acc eptable radiation dose for the patient. This study was performed on L2 vert ebrae cancellous bone specimens taken after necropsy in 22 subjects aged 47 -95 years (mean: 79 years). The segmentation process is a crucial point in the determination of accurate architectural parameters. In this paper the u se of two different segmentation methods is investigated, based on an edge enhancement and a region growing approach. The images are compared and the architectural parameters extracted from the images segmented by both method s lead to a quantitative evaluation. The parameters are found to be globall y robust towards the segmentation process, although some of them are much m ore sensitive to the approach used. Highly significant correlations (P < 0. 0005) have been obtained between the two segmentation methods for all the p arameters, with rho ranging from 0.70 to 0.93. In order to improve the asse ssment of bone architecture, texture analysis (run length method) was inves tigated. New features are obtained from an image reduced to 16 grey-levels. Textural parameters in addition to architectural parameters in a multivari ate regression model increase significantly (P = 0.01) the prediction of th e maximum compressive strength (variation of r(2) from 0.75 up to 0.89).