Author: Weixing Cai, Laura I. Cervino, Yabo Fu, Xiuxiu He, Tianfang Li, Xiang Li, Hao Zhang 👨🔬
Affiliation: Department of Medical Physics, Memorial Sloan Kettering Cancer Center 🌍
Purpose:
This work aims to develop an innovative technique to evaluate patients’ daily respiratory pattern using three-dimensional (3D) deformation vector fields (DVF) derived from a free-breathing (FB) cone beam CT (CBCT) scan. The goal is to embed this technique as a daily pretreatment quality assurance (QA) step to confirm positioning and motion of target(s) and normal tissues treated with stereotactic body radiotherapy (SBRT). A deep learning-based approach was developed to extract motion pattern as DVF since the information is inherently included in the raw projections.
Methods:
The model was trained on 123 patients and tested on 14 patients. The 3D object space is divided into an array of narrow “columns” with long sides parallel to the superior-inferior (SI) direction. Each column contains the full range of respiratory motion of its anatomy that has a large SI component but small left-right (LR) and anterior-posterior (AP) components. In the acquired FB projections, the projected regions of each column are identified, stacked, and fed into a neural network, which was trained to calculate on the motion of the anatomies that always stay in the stacked frames. The ground truth is defined as the DVF that transforms the max-inspiration phase to the end-expiration phase. The DVF of each column can be inferred from the trained network, which together constructs the 3D DVF in the patient volume.
Results:
For the test patients, the mean average error (MAE) at liver dome is 1.2mm ± 0.6mm SI, 0.3mm ± 0.2mm LR and 0.3mm ± 0.2mm AP. Maximum DVF discrepancies are noticed below diaphragm, where the motion pattern is different from lung.
Conclusion:
The model predicted DVF faithfully represented each patient’s respiratory motion in the FB scan. The DVF can be used to improve QA for assessing a patient’s daily breathing pattern before lung SBRT treatment.