A new region growing algorithm is proposed for the automated segmentation o
f three-dimensional images. No initial parameters such as the homogeneity t
hreshold or the seeds location have to be adjusted. The principle of our me
thod is to build a region growing sequence by increasing the maximal homoge
neity threshold from a very small value to a large one. On each segmented r
egion, a 3D parameter that has been validated on a test image assesses the
segmentation quality. This set of values called assessment function is used
to determine of the optimal homogeneity criterion. Our algorithm was teste
d on 3D MR images for the segmentation of trabecular bone samples in order
to quantify osteoporosis. A comparison to automated and manual thresholding
showed that our algorithm performs better. Its main advantages are to elim
inate isolated points due to the noise and to preserve connectivity of the
bone structure. (C) 2002 Elsevier Science B.V. All rights reserved.