Gain correction for variable imaging geometry

Gain (flat-field) correction is a standard preprocessing step in X-ray imaging that compensates for detector fixed-pattern response, electronic offset, and spatial non-uniformity in incident fluence. Conventional correction relies on a geometry-matched air reference acquired at the same source-detector relative position as the clinical image. In variable-geometry systems, changes in source-detector distance (SDD) and relative source-detector rotation (SDR) alter the illumination field through inverse-square dilution and orientation-dependent anode heel shading, making geometry-matched reference acquisition impractical.

To develop and evaluate a geometry-aware gain correction method that synthesizes a geometry-matched reference field from a compact calibration dataset and recorded acquisition geometry, enabling gain correction for variable geometry without geometry-matched reference acquisitions.

The measured signal was modeled as a function of the dark-current, detector response, geometry-dependent dilution, and a source fluence map that includes heel-effect gradients. Rigid transformations were used to map between the detector and source domains, for any acquisition geometry. Calibration consisted of a dark acquisition to estimate electronic offset, an air acquisition at a large SDD to estimate detector response, and a set of air images at a moderate SDD to construct a source fluence map. For each raw image, a geometry-specific synthetic reference was generated, followed by standard gain correction.

Synthetic reference images closely matched measured air references. In air-only images, residual non-uniformity after correction remained below 0.5% of the mean signal, comparable to conventional matched-reference correction across varying SDDs and SDRs. In a chest phantom study, images corrected using synthetic and matched references demonstrated high agreement (multiscale structural similarity index = 0.995; Lin's concordance correlation coefficient = 0.999) with no visually structured residual artifacts.

The proposed synthetic-reference approach enables accurate gain correction under variable imaging geometry with performance comparable to conventional matched-reference methods, eliminating the need for geometry-specific air reference acquisitions.