Purpose: Precision radiation therapy requires managing geometric variations from physiologic motion. While real-time respiratory motion tracking is reasonably mature, little attention has been paid to non-breathing intra-fractional motions (including gastric contraction motion). The purpose of this study is to develop a cyclic gastric contraction motion prediction model to support real-time management during radiotherapy.
Methods: The short-term reproducibility of phases permitted development of a prediction model that a) extracts motion states from the first couple of minutes of golden angle stack-of-stars scanning (at patient positioning) and b) aligns newly acquired data during treatment delivery to these reconstructed states. Reconstructed volumes from 5 and 10 spokes were aligned to motion states. Taking advantage of the temporal sparsity of stomach contraction, alignments were restricted to a narrow window of the next 3 motion states (4.1-7.2s depending on the patient) and projected forward, using periodic linear extrapolation of the measured cyclic phases to account for processing and system latency times.This model was tested on 20-minute data samples from 25 scans of 15 patients acquired under an institution approved protocol, with tests of 2, 3 and 5 minutes data for motion state extraction and the remaining times for testing. The 95% Hausdorff Distance (HD95) of the stomach was calculated to quantify the prediction error and geometric variation from uncorrected gastric contraction.
Results: The mean prediction error with 10 spokes and 2 minutes training was 0.3mm (0.1 – 0.7mm) with 3.4s latency, slowly rising to 0.6mm (0.2 – 1.5mm) for 6.8s and then increasing rapidly for longer forward predictions, for an average 3.6mm (2.8 – 4.7mm) HD95 of gastric motion. Using 5 spokes increased prediction errors.
Conclusion: Taking advantage of the intra-fractional stability of gastric contraction allows prediction of future stomach contractile motion with accuracy and times that are compatible with motion monitoring during treatment.
Funding Support, Disclosures, and Conflict of Interest: supported by NIH R01 EB016079
Image-guided Therapy, Modeling, MRI