Abstract—Although bone mineral density measurements constitute one of the main clinical indicators of osteoporosis, we know that bone fragility risk is also related to deteriorations of osseous architecture. Medical imaging constitutes one means to appreciate in vivo bone screen, what is particularly important in the follow up of the osteoporosis. This paper presents a method of bone textural MRI and CT scan classification, based on the use of multifractal analysis by the WTMM-2d method, we propose the choice of three features to realize these images classification: the Hölder exponents average at the peaks of Legendre spectrums, the wavelet transform skeletons density by pixel, and variance of directions of gradients. The preliminary results of 40 images directly resulting from two medical imaging (MRI and CT scan), prove to be interesting since 90% of cases are well estimated, and two classes instantaneous clustering of the results (one healthy patient class and one osteoporotic patient class) quite separate. I. INTRODUCTION LTHOUGH bone mineral density measurements constitute one of the main clinical indicators of osteoporosis, we know that bone fragility risk is also related to osseous architecture [1]. The conventional medical imaging approach (tomography) [2] restore cut by cut the matter and in particular here the osseous trabeculation. Certain sites at the risk (vertebra, heel, and wrist) territories constitute whose study can reveal deteriorations of osseous architecture: in the osteoporosis early detection case. This is why many works are devoted to osteoporosis imaging study, in particular by the fractal and multifractal analysis [3, 4, 5]. The scientific stake that we want to overcome is to be able to propose a discrimination method of osseous trabeculation images, from examination in routine clinical condition. Indeed, in mammography for example (the resolution is very close to 10 µm) seek for micro calcifications of 200 µm average size [6]. The actual clinical imagers resolution is about 100 µm, when the cortical thickness is about 10 times less. The method that we propose is inspired from Arneodo and al [7], the calculation of CWT 2d was done with recursive filter of Deriche for the Gaussian and its first derivate, the images are oversampled what increased the number of skeletons of wavelet transform, consequence of the distance between WTMM contours improvement. The takings into account of the bordering areas then of all image structural variability, and more particularly of those osteoporotic were made possible by the calculation of Hölder exponents average at the peak of multifractal spectrums. This latter were obtained from two different scale grids: the first by decreasing the large-scale analysis of the grid, and the second by taking all the possible scales as in the traditional approach. Studying the variance of directions of gradients revealed a clear difference between normal cases and pathological cases caused by an apparent anisotropy in the osteoporosis. The skeletons of wavelet transform density were also associated in the classification operation who informs us about the number of forms smoothed by the Gaussian on a finest wavelet analysis scale. II. PROCEDURE FOR PAPER SUBMISSION The calculations are related to 40 images resulting from various (MRI and CT scan) imagers of the CHRU of Lille. (http://www.u703.fr/imagesbase/). They result from areas of interest (rectangular and various sizes) positioned by the clinicians on each native image of kind to cover with the territories trabeculation strictly. They are coded on 8bits with format BMP. The statue clinical (normal or osteoporotic) is revealed only at the end of calculations. A. THE WTMM-2d METHOD: The image is regarded as rough surface, described perfectly by the relation which gives the intensities of the gray levels Z=f(x,y). We use an oversampling in lines and columns which does not affect of anything the local degree of irregularity, when we view the image from (σminﾆσmax) to (2σminﾆ2σmax) analysis scales. In fact, the Hölder condition for an α exponent in one dimension case is given by [9]: (1) α y x c y F x F - ≤ - ) ( ) ( With c>0 and α>0 this remains always valid after an oversampling of two ) 2 , 2 ( ) , ( \ \ y x y x = what gives: (2) α 2 2 ) 2 ( ) 2 ( \ \ \ \ y x c y F x F - ≤ - This simple procedure is efficient since it improves the separation distance between WTMM contours, what contributes directly in the increasing skeletons of wavelet transform in the case of weak resolution images (fig.1 and fig.2). Great variability in the structure, which characterizes certain images, more precisely in the cases of osteoporosis Classification of trabecular bone texture from MRI and CT scan images by multi resolution analysis Mohamed Khider*, Abdelmalik Taleb-Ahmed*, Patrick Dubois**, Boualem Haddad*** *LAMIH UMR CNRS 8530, Valenciennes and Hainaut Cambresis University, Valenciennes, France **Laboratoire de biophysique, INSERM, U 703, Faculté de Médecine, CHRU Lille, France *** Université Houari Boumedienne, Alger, Algérie taleb@univ-valenciennes.fr A Proceedings of the 29th Annual International Conference of the IEEE EMBS Cité Internationale, Lyon, France August 23-26, 2007. SaP2B1.22 1-4244-0788-5/07/$20.00 ©2007 IEEE 5589