AN AUTOMATIC CHANGE DETECTION METHOD IN DIFFUSION TENSOR IMAGES: APPLICATION TO MULTIPLE SCLEROSIS. BOISGONTIER Hervéa,b, NOBLET Vincenta, HEITZ Fabricea, RUMBACH Lucienb,c, CATTIN Françoised and ARMSPACH JeanPaulb
Laboratoire des Sciences de l'Image, de l'Informatique et de la Télédétection, LSIIT, UMR CNRSULP 7005, Bd Sébastien Brant, BP 10413, F67412 Illkirch Cedex, France. b Laboratoire d'Imagerie de Neurosciences Cognitives, LINCIPB, Faculté de médecine, UMR CNRSULP 7191, 4 Rue Kirschleger, F67085 Strasbourg Cedex, France c Service de Neurologie, d Service de Radiologie, CHU Minjoz, F25030 Besançon Cedex, France
[email protected]strasbg.fr a
Medical imaging brings important information to physicians for diagnostic, medical decision or patient followup. Data derived for Magnetic Resonance Imaging (MRI) is part of criteria of Multiple Sclerosis (MS) diagnosis. This technique also allows the follow of the evolution of the disease over time [1]. The advent of new quantitative MRI techniques and the constant improvement in spatial resolution yield an increasing amount of 3D image data, whose interpretation is a tedious and errorprone task. Medical image processing assist neurologists for images interpretation and analysis, particularly for patient longitudinal followup. A challenging issue of medical image processing is to design reliable methods for the automatic detection of changes between two or more images. Methods have already been proposed to detect changes in serial conventional MRI [2]. However, few works address automatic change detection in diffusion tensor imaging (DTI), which is a promising technique for monitoring neurodegenerative diseases. We propose an automatic method for change detection between DTI images, with application to the followup of multiple sclerosis lesion evolution. The method relies on the generalized likelihood ratio test [3], while accounting for the non stationarity of noise that corrupts these images. The detection can be done either on scalar images characterizing some diffusion properties such as Fractional Anisotropy (FA) or Mean Diffusivity (MD) or directly on tensor images. Validation has been conducted by simulating lesion appearances in real data. Validation shows that the detection directly on the tensor is more sensitive to registration and interpolation errors. Further work will be done for validating the method on real data from MS patients. Such automatic methods for change detection in diffusionweighted imaging may provide an objective way for characterizing the evolution of MS lesions, in particular for those located in the normalappearing white matter, and a useful tool for studying MS consequences on anatomical and functional brain connectivity. We would like to thanks ARSEP and Région Alsace for their support. [1] Y. Ge, “Multiple sclerosis: The role of MR imaging”, American Journal of Neuroradiology, vol. 27, no. 6, pp. 1165–1176, JuneJuly 2006. [2] M. Bosc, F. Heitz, J.P. Armspach, I. Namer, D. Gounot, and L. Rumbach. “Automatic change detection in multimodal serial MRI: application to multiple sclerosis lesion evolution.” NeuroImage, 20(2):643656, July 2003. [3] S. M. Kay, Fundamentals of statistical signal processing: detection theory, PrenticeHall, Inc., Upper Saddle River, NJ, USA, 1998.