Real-time needle shape prediction in soft-tissue based on image segmentation and particle filtering

Prostate brachytherapy is a current technique used to treat cancerous tissue in the prostate by permanently implanting radioactive seeds through the use of long flexible needles. This paper proposes a real-time method to predict the shape of a flexible needle inserted into soft tissue using axial Transrectal Ultrasound (TRUS) image segmentation and a non-holonomic bicycle model informed via particle filter. The needle location is tracked in TRUS images to capture the needle shape up to a specified depth. Through the use of a particle filter the noisy tracked needle shape is used to update the parameters of a kinematic bicycle model in a robust manner to predict the shape of the entire needle after it is fully inserted. The method is verified in both ex-vivo beef phantom tissue and in-vivo clinical images, yielding an average tip prediction error of less that 0.5 mm in both the ex-vivo and in-vivo image sets with a peak processing time of less than 9.5 ms per image frame.