Purpose: The goal of our method is to reconstruct the 4D‐CT from single 3D free‐breathing image acquired at the treatment day, for improving accuracy in patient setup and dose delivery during radiotherapy of lung cancer. Methods: Our method consists of two steps. First, we propose a new spatiotemporal registration algorithm to build a 4D lung motion model by establishing temporal correspondences across different respiratory phase images of the 4D‐CT acquired in the planning day. Second, we use the obtained 4D lung motion model to reconstruct a new 4D‐CT for the treatment day, using just a free‐breathing 3D‐CT acquired in the treatment day. Specifically, in this step, we first de‐interlace each slice of the free‐ breathing 3D‐CT w.r.t. the optimal phase and couch position to obtain a sequence of incomplete 3D‐CT images, each with image information only in certain slices. Then, we warp the 4D lung motion model (built in the planning day) to the sequence of incomplete 3D‐CT images for reconstruction of a new complete 4D‐CT for the treatment day. Results: The spatiotemporal registration algorithm in the first step was tested on five lung 4D‐CT datasets with manual landmarks. Our algorithm achieves the best registration accuracy, compared to several other start‐of‐the‐art deformable registration algorithms. The 4D‐CT reconstruction algorithm in the second step was evaluated on a simulated free‐breathing dataset. Our algorithm is able to reconstruct a high‐quality 4D‐CT with clear anatomical structures from single free‐breathing 3D‐CT. Conclusions: We have developed a novel two‐step method to reconstruct a complete 4D‐CT from single free‐breathing 3D‐CT acquired in the treatment day, by using a 4D lung motion model built from the 4D‐CT acquired in the planning day. Promising results have been obtained in both simulated and real‐patient datasets, indicating great potential of this method in improving the quality of image‐guided lung radiotherapy.
ASJC Scopus subject areas
- Radiology Nuclear Medicine and imaging