Author: Grant Evans, Maxwell Arthur Kassel, Charles Shang, Michael H. Shang, Stephen Shang, Timothy R Williams 👨🔬
Affiliation: South Florida Proton Therapy Institute, SFPRF, Department of Radiation Medicine, MedStar Georgetown University Hospital 🌍
Purpose:
Daily image guidance for head and neck intensity-modulated proton therapy (IMPT) presents significant challenges due to large target volumes and anatomical changes. Geometric deviations along beam projectiles can directly impact proton dose distribution. This study aimed to develop and evaluate a novel algorithm for removing fixation devices, such as masks, to enable objective surface mapping and quantify patient setup deviations.
Methods:
An automated algorithm was developed to remove immobilization masks from head and neck CTs using binary image processing, followed by identifying the largest inner contour containing the centroid and adjacent to the patient exterior. Processed contours were overlaid with simulation CTs and reconstructed into 3D surfaces using Poisson reconstruction. The beam’s eye view (BEV) portions of these surfaces were used to calculate surface mapping (SM) scores, which quantified geometric deviations between daily CBCTs and simulation CTs. SM scores were analyzed before and after resimulation events across treatment fractions. A statistical interaction test was used to assess slope differences in SM score trends, and a Student’s t-test evaluated changes in SM scores pre- and post-resimulation.
Results:
The algorithm successfully removed immobilization structures without error in over 95% of cases, enabling accurate surface contour analysis. Across a subset of head and neck cases undergoing resimulation, SM scores quantified geometric deviations over time. A trend shift in SM scores was observed after resimulation, although the interaction test (p = 0.171) indicated no significant slope difference. However, a Student’s t-test (p = 5.73e-5) demonstrated a statistically significant reduction in SM scores after resimulation, suggesting improved patient setup accuracy.
Conclusion:
This study introduces a robust algorithm for mask removal and surface mapping in head and neck IMPT, offering a quantitative framework to assess and enhance patient setups. Future work will focus on dosimetric validation and incorporating machine learning for resimulation feature detection.