3054 - A Phase-Sensitive Frequency-Domain EPID Analysis Framework for Detecting Sub-Millimeter Target Variations Beyond Conventional QA

Purpose/Objective(s): High-precision stereotactic radiotherapy achieves sub-millimeter mechanical accuracy; however, imaging-based verification remains constrained by conventional EPID quality assurance (QA) methods, including Winston–Lutz testing and gamma-based spatial comparison. These approaches rely primarily on geometric coincidence or intensity agreement and may lack sensitivity to subtle sub-millimeter perturbations. We hypothesized that phase-sensitive frequency-domain analysis enhances detection of small localization variations beyond conventional EPID QA metrics.

Materials/Methods: Two cubic phantoms containing embedded metal spheres were imaged using a 6 MV photon beam on Trilogy and TrueBeam linear accelerators. Controlled setup variations simulated sub-millimeter positional shifts while preserving a fixed coordinate reference. EPID images were acquired using 2 cm × 2 cm multileaf collimator apertures. Unlike conventional QA, which evaluates spatial-domain intensity agreement, images were transformed into the frequency domain using Fast Fourier Transform (FFT). Spatial shifts produce frequency-dependent phase modulation, enabling detection of sub-pixel localization differences even when magnitude similarity remains high. Structural Similarity Index Measure (SSIM) and gamma index (1 mm/1%) pass rates were calculated in spatial, frequency, and inverse-transformed domains to assess sensitivity differences.

Results: For different phantoms imaged on TrueBeam, SSIM/gamma pass rates (spatial, frequency, reconstructed domains) were 0.99/66%, 0.12/25%, and 0.98/66%, respectively. For Trilogy, corresponding values were 1.00/65%, 0.34/28%, and 1.00/67%. Cross-linac comparison demonstrated markedly reduced agreement (0.65/4%, 0.15/8%, and 0.59/3%). Frequency-domain analysis consistently amplified detectable discrepancies relative to spatial-domain metrics, while inverse transformation preserved structural fidelity.

Conclusion: Conventional EPID QA methods primarily assess geometric coincidence and intensity similarity and may underestimate sub-millimeter localization deviations. By leveraging Fourier phase information, this framework introduces a next-generation EPID analytic paradigm that enhances sensitivity to subtle spatial perturbations. With further validation, this approach may complement existing QA tools and support improved localization confidence, potentially enabling tighter treatment margins in stereotactic radiotherapy.