Consistent Multi-Modal Ventilation Imaging for Online Adaptive Radiotherapy Using Biomechanically Constrained Deformable Image Registration
Abstract
Purpose
Computed tomography ventilation imaging (CTVI) allows generating ventilation maps from CT data. CTVI provides spatially resolved information on regional lung function that can support functional avoidance and adaptive treatment strategies, but its current reliance on CT limits its wider use. The purpose of this work is to investigate the extension of CTVI to magnetic resonance (MR) and cone-beam CT (CBCT).
Methods
Three datasets were used: 46 4DCT/nuclear medicine images (21 SPECT, 25 PET), 7 matching 4DCT/4DCBCT and 10 4DCT/4DMR image sets. Ventilation maps were generated using a deformable image registration (DIR) approach, co-registering end-inhale to end-exhale images. Two DIR methods were used: 1) a baseline model and 2) a task-specific model which imposes a biomechanical constraint on non-lung tissue. The similarity between CT/nuclear medicine, CT/CBCT and CT/MR was measured in terms of the Spearman correlation coefficient for both DIR methods.
Results
The baseline and biomechanically constrained models achieved a similar accuracy when evaluated against PET/SPECT maps (Spearman of 0.50 ± 0.15 for baseline and 0.5 ± 0.13 for biomechanically constrained). Their performance was also very similar in terms of CT/MR reproducibility (Spearman of 0.69± 0.10 for baseline and 0.68± 0.10 for biomechanically constrained). A significant difference was observed in the CT/CBCT comparison (Spearman of 0.46± 0.30 for baseline and 0.66± 0.12 for biomechanically constrained) with a p-value of 0.04.
Conclusion
MR-based and CBCT-based ventilation imaging can produce adequately consistent ventilation maps, but this is only true when appropriate DIR models are employed. The use of a biomechanically constrained DIR model was more important for CBCT, due to the severe artifacts in the latter. Overall, enabling ventilation imaging from MR and CBCT image sets allows frequent evaluation of 3D spatially resolved pulmonary function, opening the way for function-based online adaptive treatments.