Accurate longitudinal assessment of imaging information and correlation with delivered dose often requires deformable image registration (DIR) to resolve anatomical movement. Although many DIR algorithms have been developed, current DIR techniques may be compromised by dose-dependent volumetric changes within the liver.

Researchers at the University of Michigan in Ann Arbor have created a DIR technique to improve the correlation of subvolumes within the liver. They developed a new, modified biomechanical modelling method for hepatic tissue that includes independent dose-volume response deformation forces, and combined it with a validated spatially-constrained biomechanical DIR algorithm (Morfeus). When evaluated on CT images of seven patients, the combined DIR technique improved target registration and deformed segment volume accuracy (Int. J. Radiat. Oncol. Biol. Phys. doi 10.1016/j.ijrobp.2017.06.2455).

Improved DIR techniques are needed to account for potentially significant volume changes observed following radiotherapy, the authors explain. Currently available DIR methods often cannot accurately account for large volumetric changes with localized mass loss or gain. Since many DIR techniques are primarily driven by external organ boundaries rather than intra-organ anatomy, these algorithms may result in inadequate registration of liver subvolumes and intrahepatic anatomy, even when the external liver boundary appears accurately registered.

The research, led by Kristy Brock and colleagues in the department of radiation oncology, included the development and application of a population-based normal liver tissue dose-response model and application of the dose-volume response within a biomechanical DIR algorithm. The researchers used CT data from 33 patients who received conformal high-dose radiotherapy for either a single unresectable intrahepatic malignancy or liver metastases from colorectal cancer. They created a population model to relate liver volume response to dose, in terms of a linear expansion coefficient. Seven stratifications of the response model were investigated based on the hypothesis that dose response is correlated to tumour type and spatial location of the tumour.

DIR validation

The team validated the new technique on CT images of an additional seven patients. They compared the model-predicted volumetric change within the liver to the observed change, tracked vessel bifurcations within the liver through the deformation process, and determined the target registration error (TRE) between the predicted and identified post-treatment bifurcation points. TRE analysis was performed on both an individual patient level and a population average.

Use of the Morfeus algorithm combined with dose boundary conditions demonstrated significant improvements in both volumetric modelling and TRE, compared with the use of either alone. For segments with volume changes exceeding the 95% confidence intervals (CI) for manual contouring, the combined technique deformed 100% of these segments to within the 95% CIs. By comparison, use of dose boundary conditions or Morfeus alone resulted in 81% and 63% of segments, respectively, modelled within the 95% CIs.

These results demonstrated that the combined technique can perform volumetric modelling to within the variability of manual contouring. The combined technique also represented a significant TRE improvement of 44% compared with rigid registration and 30% compared with Morfeus alone.

"Evaluation of the proposed combined DIR technique showed that all lobes were volumetric deformed to within the respective contour variability of each lobe. The average TRE achieved was 7.3 mm, representing a significant improvement over the application of Morfeus, which only achieved 10.9 mm," the authors wrote.

Lead author Daniel Polan told medicalphysicsweb that members of the Morfeus Lab are continuing to investigate innovative biomechanical techniques to improve the accuracy of deformable registration. These include using additional biomechanical properties optimized using patient classifications and clinical criteria, as well as novel boundary conditions based on imaging features. The overall goal is to achieve a voxel-level accuracy of approximately 2 mm.

"This work functions as a proof-of-concept and is currently still in the research and development stage," said Polan. "Further research will be conducted to determine if the development of these biomechanical models to determine a priori the volumetric response and resolve the spatial discrepancies of an organ to a local therapy will help advance the safe utilization of radiotherapy for liver cancer. Use of these biomechanical modelling techniques could enable improved correlation of functional imaging with delivered dose and more accurate mapping of the delivered dose from a previous treatment onto the planning images for a subsequent treatment."

Related stories

• Shrinkage predictions boost lung tumour dose
• Assessing uncertainties in image registration
• Plan incorporates tumour shrinkage
• Deformable registration: multisite scrutiny