OBJECTIVES

One of the major issues of 3D ultrasound echography compounding is image registration. Due to patient's movement, inexact speed of sound estimation and probe movement inaccuracy the different volumes are misregistred and therefore, the compounding is blurred. The complete scan may take some time, while the patient may move and/or breath. Relevant features may be hidden and mislead the diagnostic. The aim is to register the different volumes prior to compute the compounding.

There are many techniques of image registration [7]. The objective of this project is to evaluate a new technique for 3D volume elastic registration presented in [1]. It is based in volume subdivision, block matching using mutual information [2] and thin-plate spline global interpolation. The main goal is to implement these elastic transformations, evaluate them with several cases and propose improvements.

Theoretical work will be needed to understand the underlying principles of these techniques, specially regarding the implementation of the mutual information metric and the probability estimations needed to compute it. Three-dimensional splines need also to be understood and implemented.