Room: Exhibit Hall | Forum 8
Purpose: Quantitative assessment of radiotherapy response in the treatment of cholangiocarcinomas across longitudinal images is challenged by complex deformation present in the images due to tumor response and patient positioning. Advanced deformable image registration (DIR) is required in this complex setting. Here we compare four commercially available DIR techniques.
Methods: For fifty-one choloangiocarcinoma patients, the liver was segmented on pre and post-radiotherapy CT images, using a convolutional neural network. The post-radiotherapy image was registered to the pre-treatment image using four DIR techniques available in a commercial treatment planning system: (1) intensity-based registration (IR); (2) intensity and structure-based hybrid registration with focus on the liver (HRL); (3) a finite element-based biomechanical registration with focus on the liver (BRL) and (4) BRL with additional focus on a single manually-contoured internal structure (BRLI). Registration accuracy was evaluated using target registration error (TRE). The results were analyzed with a one-way ANOVA followed by the Tukey multiple-comparison (TMC) test. For the ANOVA, the TREs were first normalized by dividing them by the corresponding voxel diagonal length to compensate for the different resolutions of images.
Results: Twenty patients have been analyzed to date. Of the four DIR techniques, BRLI achieved the best average registration error (5.6mm, SD 2.0mm) and resulted in the best registration accuracy in 9 out of 20 cases. Moreover, for 12/20 patients the performance was within the image resolution. The one-way ANOVA found DIR technique to be a statistically significant factor (p-value=0.003). However, the TMC test showed that only IR (average TRE = 8.1mm, SD 3.0mm) was statistically different (poorer performance) from the other three methods.
Conclusion: Our study demonstrates the utility of biomechanical DIR and the potential limitation in standard intensity-based DIR. Automated vascular segmentation is needed to further improve the biomechanical technique with internal boundary conditions and enable an automated workflow.