SU-E-J-84: Investigating the Use of Deformable Algorithms to Register and Accumulate Dose to the Spinal Cord for...

Mega-voltage cone-beam CT (MVCBCT) imaging can be used to monitor changes in neck flexion and calculate the true spinal cord dose during radiotherapy. To sum the dose over multiple fractions and compare with the planned dose requires deformable image registration. This study investigates the accuracy of deformable registration of the spine in kilovoltage CT (kVCT) to MVCBCT images for head-and-neck patients.

Head-and-neck kVCT and MVCBCT images were selected from five patients who exhibited large changes in neck flexion. The spinal canal was contoured in each image. MVCBCT images were rigidly registered to the kVCT image by aligning the C1-C2 vertebrae. Variations of a commercial deformable registration algorithm using the B-spline method were applied to a volume of interest surrounding the spine to deform the spinal cord structure from the kVCT to the MVCBCT. The Dice similarity index (DSI) was calculated between rigidly registered structures (DSI1), and compared against the DSI calculated between the MVCBCT structure and the structure deformed from kVCT to MVCBCT (DSI2). The difference between DSI2 and DSI1 was calculated to investigate the ability of the deformation algorithm to match the change in neck flexion.

In this data set, a single-pass B-spline deformation algorithm was the most effective in deforming the kVCT spinal canal to the MVCBCT spinal canal, with a mean improvement in DSI (DSI2 - DSI1) of 0.134. The mean change in DSI for a single-pass algorithm with a contrast correction was 0.111; for a multiple-pass algorithm was 0.105; and for a multiple-pass algorithm with contrast correction was 0.075.

Several versions of a commercially implemented B-spline deformation algorithm were found to improve the registration of the spinal canal in kVCT and MVCBCT head-and-neck images. The resulting accuracy in the calculation of the cumulative dose to the spinal cord is under investigation. We acknowledge Velocity for providing the registration software.