Ultrasound tracking of organs or target volumes is a promising means to correct the displacement caused by respiration and errors from repositioning in medical applications e.g. in radiation therapy. However, one major problem of ultrasound images is their inherent low contrast and clutter which often makes standard algorithms instable for this purpose. In this work we present the adaption and application of a probabilistic tracking approach based on conditional density propagation (condensation) for real-time tracking on ultrasound images. This approach promises to facilitate robust real-time tracking with 5 degrees of freedom (translation and scaling in x-/y- direction, rotation) of anatomic structures on noisy and low contrast ultrasound images. The real-time performance and precision of the algorithm are investigated with ultrasound data from the liver. The tracking results of the algorithm are compared with results obtained from image registration. It is shown that this algorithm is real-time capable with processing time less than 5 ms per frame and robust on low contrast target structures with a precision below 1.6 mm in translation. Compared with an independent image co-registration method, this method leads to a superior displacement correction in pre-delinated target structures.
展开▼
