A Method to Generate CT Scanner-Independent Synthetic Internal Target Volumes for a Dynamic Thorax Phantom with an Embedded Target: Single Waveform, Fixed-Period Motion
Abstract
Purpose
The purpose of this study is to present a novel quality assurance framework that provides a reproducible gold standard for evaluating 4DCT volume reconstruction performance. This framework establishes a gold standard ITV by providing an independent, quantitative reference for evaluating how accurately a 4DCT system acquires and reconstructs the motion excursion of an internal target.
Methods
The geometry of a commercially available dynamic thorax phantom was modeled in Python to generate a synthetic CT volume. Nine known respiratory motion profiles were programmed into the synthetic dataset, and a spherical target was embedded and repositioned across respiratory phases. The resulting internal target volumes (ITV) were generated on a slice-by-slice basis from the phase-specific positions. The datasets were saved as DICOM files. Fiducial markers were embedded in the synthetic CT to enable rigid registration with user-acquired phantom scans in Velocity AI. Using Python, quantitative comparison between acquired and synthetic ITVs was performed using metrics such as volume difference, Dice Similarity Coefficient, Jaccard Index, and Hausdorff distance.
Results
Across nine motion profiles, synthetic and acquired ITVs demonstrated strong volumetric and geometric agreement. Absolute volume differences were ≤0.33 cc, with relative volume differences between 4.2% and 0.1%. Dice similarity coefficients ranged from 0.95-0.97, with Jaccard indices of 0.90-0.93 and precision and recall ≥0.94 across all profiles. Surface agreement was high, with mean surface distances of 0.16-0.33 mm and 95th-percentile Hausdorff distances ≤1.56 mm, demonstrating sub-voxel spatial concordance across all motion conditions. The conventional analytical ITV formulation commonly used in 4DCT QA systematically overestimated ITV volume for multi-axis motion, with relative differences up to 23.0%, whereas the straight-line swept-volume formulation closely matched the synthetic and acquired ITVs within 0.19 cc.
Conclusion
This framework establishes a scanner-independent ground truth for 4DCT ITV validation, enabling direct, reproducible assessment of volumetric and geometric reconstruction accuracy.