Regularized autoregressive models for a spectral estimation scheme dedicated to medical ultrasonic radio-frequency images

Abstract

The local spectral estimation from radio-frequency (RF) signals in medical echographic ultrasound images is not a trivial task due to the noisy nature of the data resulting from a stochastic and nonstationary process, Significant improvements may be obtained by proposing a spatial regularization scheme, smoothing the local spectral estimates while preserving the discontinuities. Based on AR models, the authors propose a 2D regularization scheme in a Bayesian framework. The a-priori knowledge is expressed by means of Markovian Random Fields (MRF) defined on the reflection coefficients. The use of nonquadratic functions allows to preserve discontinuities. First the authors applied their method on simulated data containing spatial discontinuities of spectral characteristics, which showed the efficiency of the regularization technique. Then the technique was used on cardiac RF data. This shows the improvements as well for Integrated Backscatter (IBS) images as for Mean Central Frequency (MCF) Images or whole spectral estimation.