Real-time parametric curved needle segmentation in 3D ultrasound images

Abstract

3D Ultrasound-guided needle insertion has recently received much attention in minimally invasive percutaneous procedures such as image guided needle steering. In this paper we present a novel technique for real-time detection of curved needles in three-dimensional ultrasound images. A new parameterization for the generalized Radon transform is introduced and implemented on a graphics processing unit using NVIDIApsilas Compute Unified Device Architecture (CUDA) programming. The effectiveness of the algorithm is demonstrated on reconstructed 3D-US images of a brachytherapy needle carried by a laparoscopic needle holder mounted on a minimally invasive surgical robotic test-bed. Results of experiments for insertion of three different needles into a tissue mimicking gelatin agar cube are presented.