We are studying the use of deformable image registration as a tool in image-guided radiation therapy of the prostate. As modern techniques for conformal therapy attempt to fit the high-dose region as closely as possible to the target volume, it becomes more essential to ensure that the tumor is in the expected location, and to assess the dose actually delivered to each organ.
We use deformable image registration in two ways. First, we use it for automatic segmentation of images acquired at treatment time, by deforming manual segmentations of the planning day image. Second, we use it to estimate the dose actually delivered, by deforming calculated dose distributions from each treatment day. In each case, the deformations we apply are computed by registering a treatment image to a planning image.
We perform our registration in two stages. We first perform a translation based on bony landmarks, and then a deformable registration in a region of interest. The images to the right illustrate the effects of the stages. Only the conturs differ; the images are all the same slice from the planning CT of a patient. The red contour is the location of the prostate in the planning CT, and the yellow contour is the rectum. The green contours indicate the locations of the prostate in the daily images. The top image shows the prostate locations as treated. The middle images shows the prostate locations after translation to align bony landmarks. And the bottom image shows the locations of the prostate after deformable registration is performed. In the latter case, the image deformation was applied to 3D prostate models derived from the manual segmentations, which were then resliced to get the deformed contours.
The effect of the registration steps on the actual images can be seen in animations of the daily images in sequence. A figure assessing the accuracy of our automatic segmentation method can be found below.
Papers
- Davis B, Foskey M, Rosenman J, Goyal L, Chang S, and Joshi S. 2005. Automatic segmentation of intra-treatment CT images for adaptive radiation therapy of the prostate. Proc. Medical Image Computing and Computer-Assisted Intervention.
- Foskey M, Davis B, Goyal L, Chang S, Chaney E, Strehl N, Tomei S, Rosenman J, and Joshi S. 2005. Large deformation 3D image registration in image-guided radiation therapy. Phys. Med. Biol. 50, 5869-5892.
Posters
These are Powerpoint files with one 48x36-inch slide. You may be able to print them on on a letter-size sheet of paper by choosing a "Reduce to fit" option from the print dialog.
- Foskey M, Rosenman J, Goyal L, Chang S, and Joshi S. Calculating biological effective dose in the presence of organ deformation. AAPM 2005.
- Bechtel J, Joshi S, Chang S, Goyal L, Cullip T, Tracton G, Rosenman J, Foskey M. Motion Causes Prostate Underdosage when a Common Planning Treatment Volume Definition is Used in IMRT Treatment Planning. ASTRO 2004.
Talks
- Powerpoint presentation giving an overview of the work.
 |
 |
 |
Comparison of Automatic and Manual Segmentation
These are maps, over the surface of the prostate, of the average radial distance between paired segmentations. For both of the maps, one of the segmentations is manual.
Top The alternate segmentation is our automatic method, in which a manual segmentation from the planning day is warped to the treatment day using a deformation determined by image registration. The planning-day segmentation is by the same segmenter as the daily ones to which we are comparing.
Bottom The alternate segmentation is by another manual segmenter.
|
|
