Unencrypted login | home

Program Information

4D Ultrasound Calibration for Radiotherapy Guidance Using Automatic Intramodality Image Registration

no image available
J Schlosser

J Schlosser1*, C Kirmizibayrak1, V Shamdasani2, S Metz2, D Hristov1, (1) Stanford University, (2) Philips Medical

SU-D-213CD-1 Sunday 2:15:00 PM - 3:00:00 PM Room: 213CD

Purpose: In prior work we developed a robotic system providing real-time soft-tissue ultrasound (US) volumes during radiotherapy beam delivery. For image guidance, the US volumes must be transformed to the linear accelerator reference frame. In this work we propose and characterize a new method of calibrating 4D US volumes based on automatic intramodality image registration.
Methods: A dynamic navigation link was used to port 3D US volumes from a Philips iU22 xMatrix machine to a PC in real-time. Sixty volumetric (3D) US images of a pelvic phantom were collected from various probe positions while the transducer's pose was monitored by an optical tracking system. US volumes were automatically registered to the first US volume using normalized mutual information. A system of equations was formulated and solved for the US probe-to-image transformation using the registration transformations and the optical tracking information. Accuracy of the US calibration was assessed on eight additional US volumes with two separate methods. In the first method, a set of three fiducial markers implanted in the phantom was manually selected in each volume by three individual readers. Selected marker locations were reconstructed in the stationary camera frame, and for each marker, mean distance to the reconstructed centroid was measured. In the second method, a bladder structure was semi-automatically segmented in each image volume. Mean distance between bladders segmented in a reference volume and the other seven volumes was computed. Calibration accuracy was also investigated as a function of the number of calibration images used.
Results: Mean error for the fiducial marker reconstruction was 2.3 mm. Mean distance error between segmented structures was 1.1 mm. The proposed calibration method typically converged with less than 20 images.
Conclusion: Automatic image registration facilitates fast and simple US spatial calibration with accuracy under 2.3 mm using any US phantom.

Contact Email