Needle tracking in 3D ultrasound

This work is part and continuation of our previous work on RF 3D Imaging.

We have previsouly developped several techniques and algorithms to detect and identify the precise position of a straight, or slihlty curved needle, in 3D ultrasound volumes UHER-13 . Wheras some techniques adress this problem by developping specific technologies and materials (active needles, Sonix GPS, highly echoic needles etc...) we have made the choice to work only with an image processing approach. Therefore the only specificty compared to clinical practive, which involves usually only 2D imaging, is that we consider 3D ultrasound imaging.

The method we have proposed recently  (called ROI-RK ZHAO-13) is based on a combination of Line filtering, RANSAC and Kalman filtering. The implementation consists in a two step algorithm: 1) Line filter and ROI Initialisation and 2) RANSAC-Kalman loop.

1) As described in the figure below, when the first ultrasound volume is acquired it must be processed entirely since there is no priori knowledge of the position of the needle. In order to improve the image quality and increase the probability to find the needle in this volume, Line Filtering (Frangi's method) is implemented in order to enhance line-like structures. Then RANSAC model fitting is applied to find the needle.

This approach is quite efficient but also time consuming. Using it for each new incoming volume would avoid the possibility to meet real-time requirements which are a must in ultrasound imaging. As a result at the end of this first step, a ROI is initialized around the needle axis. This ROI is a cylinder.

2) Once the ROI is initialized, when the next volume is acquired, the folowing steps are performed. A speckle tracking algorithm estimated the needle tip displacement. RANSAC algorithm is run in the ROI to find the new needle axis. Based on these two measurments, a Kalman filter is used to predict and update the ROI position. This is done in an infinite loop.