Gridline Suppression in X-Ray Imaging Using Global Feature-Augmented U-Nets – MT Final Talk by Shadi Khamseh

Join us for the final presentation of a Master’s thesis on “Gridline Suppression in X-Ray Imaging Using Global Feature-Augmented U-Nets.” In digital radiography, anti-scatter grids are used to reduce scattered radiation, but because modern detectors have very small pixel sizes, the high-frequency grids needed to avoid visible gridlines are not available. Clinical systems therefore use lower line-density grids, which produce gridline artifacts that overlap with anatomical structures. As traditional filtering cannot remove these patterns effectively, algorithmic correction is required.

This thesis proposes a deep-learning approach that incorporates global feature representations into a FiLM-modulated U-Net to improve gridline suppression. Using paired acquisitions, synthetic gridline generation, and residual learning, the model learns to remove gridlines while maintaining the underlying anatomical structures. The pipeline is evaluated against a standard U-Net baseline, showing that the feature-augmented models achieve clearer and more stable artifact suppression across different detector systems.