Section: New Results

Medical robotics

Visual servoing based on dense ultrasound information

Participants : Caroline Nadeau, Alexandre Krupa.

In the context of the ANR USComp project (see Section  8.2.3 ), we pursued our works on the development of ultrasound image-based visual servoing methods that directly use pixel intensities of the ultrasound image as control inputs. In opposite with methods based on geometrical visual features, this new approach does not require any image segmentation step that is difficult to robustly perform on ultrasound images. By coupling our method with a predictive control law based on the periodicity of physiological motion, we propose a solution to stabilize the ultrasound image by actively compensating the physiological motions of the patient. The principle consists in automatically synchronizing the 6 DOF motion of a 2D or 3D probe with the rigid motion of a soft tissue target. First ex-vivo results obtained on animal tissues demonstrated the validity of the concept [39] .

In collaboration with Prof. Pierre Dupont from Harvard University at Boston, we also addressed the motion tracking of a target that can consist of either the tip of a robot inserted on a beating heart or cardiac tissues. Unlike the previous work, where the motion compensation task was realized physically by moving the probe attached to a robotic arm, we propose here to track the motion of the target using a 3D region of interest (ROI) which is automatically moved within the whole volume observed by a 3D probe thanks to our intensity-based ultrasound visual servoing method. In vivo animal experiments were conducted in Children's Hospital at Boston and validated this tracking approach [38] .

Autonomous control modes for ultrasound probe guidance

Participants : Tao Li, Alexandre Krupa.

In the context of the ANR Prosit project (see Section  8.2.2 ), we proposed several autonomous control modes in order to assist a doctor during a robotized and teleoperated ultrasound examination (tele-echography). This year we developed an assistance functionality that automatically maintains the visibility of an anatomic element of interest while the doctor teleoperates the 2D ultrasound probe held by the medical robot. The method is based on a multi-task controller that gradually activates an ultrasound visual servoing in case some geometrical features leave a pre-defined safe area of the image in order to bring them back inside the view [33] . With this approach the DOFs of the robotized probe are not exclusively constrained by the visibility task but also available for the tele-operation. This new assistance functionality was implemented on the ANR Prosit robotic platform and first in vivo results obtained on a human volunteer validated the concept.

Real-time soft-tissue deformation tracking in 3D ultrasound

Participant : Alexandre Krupa.

We proposed a dense ultrasound tracking algorithm that estimates in real time both rigid and non-rigid motions of a region of interest observed in a sequence of 3D ultrasound images. The deformation is modeled by 3D thin-plate splines (TPS) whose parameters are estimated online from intensity difference measured in successive volumes. To increase the robustness of this approach to image noise, we proposed two solutions to mechanically constrain the deformable model. The first is based on the addition of a regularization term in the TPS model and the second consists in coupling the TPS with a mass-spring system. These methods were validated on simulated sequences of deformed 3D ultrasound images.

Needle detection and tracking in 3D ultrasound

Participant : Alexandre Krupa.

We designed an algorithm able to detect a needle inserted manually in a 3D ultrasound volume from an arbitrary point, and able to robustly track this needle in real-time. We also experimentally demonstrated the possibility to guide the ultrasound probe to keep the needle visible and aligned, using visual servoing. Such a system could assist an operator during manual insertions, which are currently performed under free-hand ultrasound monitoring. In addition, we plan in future works to combine this method to a needle steering robotic system for guiding accurately the needle toward a target while optimizing its visibility.