Simulating the breathing of lung cancer patients to estimate tumor motion and deformation at the time of radiation treatment

Abstract

The motion caused by breathing influences the precision of radiation treatment (RT) delivery to the tumors in the thoraco-abdominal region. The limitations associated with currently implemented motion management techniques expose healthy tissue to unnecessary radiation toxicity. Surrogate-based motion models (SMMs) offer a solution to this problem. Commonly, to build and train these models we correlate an easy-to-measure external surrogate to the simultaneously acquired internal anatomical motion- measured from respiratory-correlated CT (4DCT) scans. However, ever-changing nature of the internal-external correlation demands routine updates of the correlation for maintaining the accuracy of the SMM. Additionally, clinical implementation of these models is pending for their thorough in-situ validation.In this work we present the methodology of incorporating fluoroscopic projections (FL) acquired at the time treatment delivery for SMM construction and update. To develop the methodology, we designed and implemented a fully deformable lung phantom. We collected 4DCT data, photogrammetry surfaces, and orthogonal fluoroscopy while the phantom simulated respiration-induced motion by playing patient-derived tumor trajectories. The collected data was used to develop and evaluate methodologies for data processing and analysis. Under a prospective Institutional Review Board approval, 4DCT data, photogrammetry surfaces of chest and abdomen, and orthogonal FLs were collected from five lung cancer patients. A simulated annealing optimization scheme was used to estimate optimal lung deformations by maximizing the mutual information (MI) score between digitally reconstructed radiographs (DRRs) of the model-estimated DRRs and FLs. Subsequently, we used mathematical methods such as partial-least-regression and Principal Component Analysis to train the SMM to enable transient estimation of lung surface deformation (boundary conditions) with sub-millimeter accuracies. However, evaluation of SMM performance in estimating the tumor shape and position was not possible due to the inherent lack of contrast inside the lungs in fluoroscopy data. We generated a finite element (FE) model of lung and tumor in ABAQUS and used the SMM-generated lung surface BCs to simulate the respiration process as a function of time. The 95th percentile absolute difference between FE-estimated and 4DCT-delineated tumor centroid in the superior-inferior direction was 2.7 mm. Sub millimeter median errors were observed for all patients. Lung tumor motion was simulated during radiotherapy session for five patients.