Purpose: Deformable image registration (DIR) is increasingly used in radiotherapy applications and provides basis for previous treatment evaluation and dose accumulation. We examine the accuracy of DIR algorithms available on commercial planning platforms and evaluate their clinical applicability for the abdominal site.
Methods: Following the guideline of AAPM TG-132, we implemented a patient-specific QA process to evaluate the accuracy of DIR for the abdomen site using digital phantoms created via ImSimQA (Shrewsbury UK). To simulate the impact of respiratory motion, we deformed the planning CT image at the end of exhalation to the end of inhalation using the biomechanical model embedded in ImSimQA. Consequently, the respiratory deformations were transferred onto the GTV in pancreas in the abdominal region, and we obtained the deformation vector field (DVF) as ground truth. Subsequently, we exported the deformed CT to Eclipse and MiM, performed DIR, and derived DVF_Eclipse and DVF_MiM for evaluation. Quantitative differences between DVFs, such as mean, max, and the distribution were tabulated. Geometrically, dice coefficient between the DVF-propagated and ground truth tumor was calculated. We generated a comprehensive report to illustrate the performance for the patient of interest as QA record.
Results: The biomechanical model-derived DVF covers an FOV from the mid lung to the inferiormost pancreas boundary, with displacements ranges from 0 to 6.8mm. Data analysis comparing DIRs using DVFs shows the mean ± standard deviation 3D discrepancy is 1.8±0.8 mm for Eclipse, and -1.2±0.9 mm for MIM, respectively. Discrepancies near the diaphragm are larger than the mid lung or the pancreas. MIM tends to underestimate while Eclipse tents to overestimate the effect of deformation.
Conclusion: DIR performed by Eclipse and MIM are evaluated, with a realistic biomechanical model derived DVF serving as ground truth. Further investigations on the impact of DIRs on dose distribution for radiation therapy is warranted.
Deformation, Registration, Image Fusion