Using needle detection and tracking for motion compensation in abdominal interventions

In this paper, we present a method of using the needle detection and tracking to compensate breathing motion in 2D fluoroscopic videos. The method can robustly detect and tracking needles, even with the presence of image noises and large needle movements. The method first introduces an offline learned needle segment detector that detects needle segments at individual frames. Based on detected needle segments, a needle is interactively detected at the beginning of an intervention, and then is automatically tracked based on a probabilistic tracking framework. A multi-resolution kernel density estimation is applied to handle large needle movements efficiently and effectively. Experiments on phantom and clinical sequences demonstrate that the method can successfully track needles in fluoroscopy, and can provide motion compensation for abdominal interventions.