Recalage d’images échographiques par apprentissage pour l’échographie 3D main libre sans capteur de position

Abstract

Sensorless freehand 3D ultrasound imaging is an approach which allows the reconstruction of ultrasound volumes by analyzing the contents of the 2D scans in a freely acquired sequence to estimate their relative positions, instead of relying on a position sensor. Indeed, the distance between scans can be inferred using a speckle decorrelation model which describes the relationship between image correlation and elevational distance, while in-plane motion can be estimated by conventional registration methods. However, each type of motion (in-plane and out-of-plane) occuring in freehand motion is a source of decorrelation, and these multiple sources cause motion estimation errors. A study of these errors based on experiments on synthetic imagery is first presented, confirming the hypothesis that the errors are largely systematic and that a relationship between these errors and the ground truth displacement exists. A novel method to correct the estimations is thus proposed, based on the learning of this relationship using the Relevance Vector Machine algorithm. Experiments on synthetic imagery demonstrate the validity and reliability of our method : the results show a significant error reduction in comparison with the conventional method.

Date	24 Aug 2012
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Catherine Laporte (Supervisor)