Data-Augmented Deep Learning for Anatomical Completion and Dose-Accurate MR-Only Radiotherapy Planning
Abstract
Purpose
In MR-only radiotherapy planning (MROP), limited field-of-view (LFOV) acquisition and imaging artifacts can introduce truncated anatomy and density inaccuracies, reducing the reliability of dose calculation and preventing comprehensive plan evaluation due to incomplete visualization of OAR. This work develops and evaluates a data augmentation-driven (DAD) deep learning framework for recovering planning-appropriate images from truncated or corrupted inputs.
Methods
An image recovery framework was developed in which image degradation was synthetically introduced during training using a DAD strategy. The network was trained with full-length reference images as recovery ground truth, while degraded inputs were generated on-the-fly by applying masks with randomly sampled sizes and spatial locations to simulate LFOV effects. During inference, the trained recovery network was applied to truncated images to generate recovered images suitable for dose calculation. To demonstrate the framework for MROP, truncated brain MR images were first converted to truncated synthetic CT (sCT) using an MR-to-sCT (MR2sCT) model, followed by an sCT recovery step (sCT2sCTx) using the proposed image recovery network to generate extended sCT (sCTx) volumes including partial neck anatomy, making them suitable for treatment fields. The sCT2sCTx model was implemented as an auto-encoder adapted from nnU-Net and trained using longitudinal box masking to simulate truncation during data augmentation.
Results
The sCTx achieved reliable anatomical compensation and tissue contrast compared with the reference CT (rCT), including truncated regions. Dose calculation performed on sCTx demonstrated high agreement with that based on the rCT, as reflected by consistent DVHs and a 98.67% gamma passing rate using 3%/3 mm criteria between the dose distributions.
Conclusion
A DAD deep learning framework was developed to recover truncated anatomy in LFOV medical images. In MROP applications, the recovered images supported accurate dose calculation and more complete plan evaluation by restoring missing anatomical regions, enhancing robustness in the presence of LFOV constraints.