Purpose: To improve the sensitivity and specificity of EPID-based measurements of picket fence tests for MLC QA by accounting for small translational and rotational differences between the MLC and imager coordinate systems.
Methods: A custom picket fence (PF) pattern containing four additional 1x1 cmÂ² fiducial openings near each corner of the field was developed. A single script acquired four integrated EPID images at 100-cm SID at each cardinal angle though a 40x40 or 40x22 cmÂ² field size for standard definition (SD) or high definition (HD) MLCs, respectively. In-house software used the 1x1 cmÂ² openings to align the imaging panel and MLC coordinate systems, effectively correcting the images for translational and rotational shifts from a variety of sources. Individual leaf edges were detected and used to represent the MLC position. The reference MLC positions were derived from a PF image acquired at the zero gantry angle. ROC analysis used uncorrected images with known MLC errors to aid in the selection of the optimal error threshold that maximized the true positive fraction and minimized the false positive fraction. Analysis was repeated on images with corrections applied to investigate any benefit.
Results: Without corrections, an optimal threshold of 0.9 mm maximized the sensitivity/specificity of the SD MLC at 90.5%/90.0% and whereas a 0.95 mm threshold yielded 94.6%/93.8% for the HD MLC. With corrections applied, an optimal threshold of 0.95 mm maximized the sensitivity/specificity of the SD and HD MLCs at 99.9%/99.9% and 99.6%/98.8%, respectively. The AUC improved from 0.9654 to 1.0000 for the SD MLC and from 0.9814 to 0.9996 for the HD MLC.
Conclusion: The sensitivity and specificity of EPID-based MLC QA can be improved when test images are corrected for residual translational and rotational shifts. More benefit was observed for the large field of the SD MLC.