Section: New Results

Medical robotics

In order to track the motion of a tumour or cyst during needle insertion, we developed a first approach to track a deformable target within a sequence of 2D ultrasound images. It is based on a dedicated hierarchical grid interpolation algorithm (HGI) that is typically used for real-time video compression purposes. This approach provides a continuous motion representation of the target by using a grid of control points that models both their global displacement and local deformations. The motion of each control point is estimated by a hierarchical and multi-resolution local search method in order to minimize the sum of squared difference of the target pixel intensity between successive images. This approach was validated from 2D ultrasound images of real human tissues undergoing rigid and non-rigid deformations [48] and was recently adapted for tracking 3D deformations.

A second approach for automatically tracking deformable target within 2D ultrasound images has been developed [50] . It combines dense information with a physically-based model and has therefore the advantage of not using any fiducial marker. The physical model is represented by a mass-spring damper system driven by external and internal forces. The external forces are obtained by maximizing an image similarity metric between a reference target and the deformed target along the time. The internal forces of the mass-spring damper system constrain the deformation to be physically plausible and therefore efficiently reduce the sensitivity to the speckle noise. This second approach was validated on simulated and real data, both for rigid and non-rigid motions of soft tissues [49] . It was recently extended for deformable target tracking in 3D ultrasound volumes.

The objective of this work is to provide robotic assistance during needle insertion procedures such as biopsy or ablation of localized tumor. A method has been developed for steering a beveled-tip flexible needle actuated by a robotic arm in such a way to control the needle curvature in 3D space [34] . It is based on the design of a new duty-cycling control strategy that makes possible to control both the 2 lateral angular velocities and the insertion velocity of the needle tip (3 DOF). An image-based visual servoing approach has then been developed to automatically position the needle tip on a 3D target indicated by the user. It is based on the use of geometrical visual features extracted from 2 images provided by 2 orthogonal cameras observing a translucent gelatin phantom where the needle is inserted. Preliminary results of this automatic targeting task demonstrate the feasibility of this new concept and its robustness to needle kinematic model errors [35] . We recently extended this approach to automatically steer the needle toward a target by an image-based visual servoing that uses geometrical features extracted from images provided by a 3D ultrasound probe.