Accurate needle localization in two-dimensional ultrasound images

Ultrasound imaging is commonly used to guide needle interventions. However, needle localization in ultrasound images is often challenging, particularly when the needle visibility is poor and the image includes strong linear structures. This paper presents a new method to accurately localize the needle. The method employs two features. The first feature quantifies the needle-induced motion and the second feature provides intensity-invariant detection of the edges that match the needle direction. These features are analyzed using Radon transform to approximately estimate the needle trajectory. The needle axis is localized by defining a region around the initial needle trajectory and employing the ranklet and Radon transforms to analyze the two features within this region. The needle tip is determined by analyzing the local phase, speckle variations, and image intensity along the needle axis. The proposed method is used to accurately localize the needle in both ex vivo and in vivo ultrasound images.