Author: Resat Aydin, Joseph Barbiere, Brett Lewis, Roland Teboh 👨🔬
Affiliation: HUMC, Hackensack University Medical Center 🌍
Purpose:
Accurately compensating for respiratory-induced tumor motion is critical in BgRT, where precise delineation of volumes ensures effective dose delivery. We propose an integrated approach that combines Mean Squared Error (MSE) and structural similarity (SSIM) measures to robustly quantify and rank for the best match 4D-CT phase for contouring.
Methods:
Each phase of the 4D-CT was aligned with the untagged CT for geometric consistency. The intensities in each phase volume were normalized to standardize the intensity range. MSE comparison was aimed to focus on regions of intensity mismatch. SSIM was subsequently computed to capture local structural and textural similarities. These two metrics were combined through a weighting scheme to yield a single integrated similarity score. Phases were then ranked in descending order based on this composite score to find the “best match” phase.
Results:
Our approach was tested on a motion phantom for which 4D-CT was performed. The ten phases (from 0% to 90% with 10% increment) were compared against an untagged CT image. The top three ranked phases were 70%, 30% and 50%, which were consistent with the findings upon a visual determination made to delineate volume. Further validation was performed choosing the 0% phase as a reference, this time, and the other phases including the 0% phase were then compared. The latter comparison revealed that the 0% phase had the best match which was trivial, but was important to prove the algorithm works as intended. The next best three rankings were 90%, 10% and 80%, respectively, and the worst three rankings were 40%, 50% and 60%, respectively, which would be expected from a sinusoidal respiratory cycle.
Conclusion:
This approach reduces phase selection uncertainties and advances planning accuracy for BgRT while offering a reproducible framework for managing respiratory motion and improving radiotherapy outcomes.