IEEE/SP 13th workshop on Statistical Signal Processing July, 17-20, 2005 - Bordeaux - France

All papers by session

Tissue Classification in Dynamic Contrast Enhanced Magnetic Resonance Imaging by a Fast Regularization Method
Plenary Lecture 4

Abstract

A dynamic contrast enhanced MR clinical exam produces a set of images which monitor in time the concentration of a contrast agent in the vessels of an organ. The portion of tissue represented by each pixel can be classified as a benign or malignant tumoral one, according to the qualitative behavior of the time series associated to it. The time series can be considered as the noisy output of a pharmacokinetic distributed model whose parameters have an intrinsic diagnostic importance. For technical reasons, the SNR and the number of observations can not be both large. This makes difficult parameters identification and classification, especially when short computation time is required. A family of functionals parameterized by a few hyperparameters is considered. The problem is then formulated as a constrained optimization one in order to estimate jointly the pharmacokinetic parameters, the classification labels and the hyperparameters, therefore providing an automatic method. An heuristic algorithm is proposed which produces results similar to those obtained by a MCMC based approach. However the MCMC algorithm is about ten times slower. The algorithm is an extension of an iterative algorithm, whose convergence can be proved, to solve a similar, but simpler, constrained optimization problem. The performances of the method are illustrated on a real data set.