This paper deals with the development of a new 6 degrees-of-freedom (DOF) vision-based controller for robot-assisted medical applications. The main objective is to use visual information extracted from Ultrasounds (US) images to control an US probe held by the robot. Instead of the conventional use of visual features e.g., geometric features (points, lines, moments, etc.) in the visual control law design, the described method use the shearlet coefficients. More precisely, the time-variation of the coarsest level of shearlet decomposition coefficients are linked to the US probe (respectively to the robot) spatial velocity and then related to the task-function control law. The proposed control law was experimentally tested and validated using a realistic abdominal phantom. The obtained results demonstrated promising performances in terms of accuracy, repeatability, robustness and convergence behavior.