MLC quality assurance using EPID: A fitting technique with subpixel precision

Abstract

Amorphous silicon based electronic portal imaging devices (EPIDs) have been shown to be a good alternative to radiographic film for routine quality assurance (QA) of multileaf collimator (MLC) positioning accuracy. In this work, we present a method of acquiring an EPID image of a traditional strip-test image using analytical fits of the interleaf and leaf abutment image signatures. After exposure, the EPID image pixel values are divided by an open field image to remove EPID response and radiation field variations. Profiles acquired in the direction orthogonal to the leaf motion exhibit small peaks caused by interleaf leakage. Gaussian profiles are fitted to the interleaf leakage peaks, the results of which are, using multiobjective optimization, used to calculate the image rotational angle with respect to the collimator axis of rotation. The relative angle is used to rotate the image to align the MLC leaf travel to the image pixel axes. The leaf abutments also present peaks that are fitted by heuristic functions, in this case modified Lorentzian functions. The parameters of the Lorentzian functions are used to parameterize the leaf gap width and positions. By imaging a set of MLC fields with varying gaps forming symmetric and asymmetric abutments, calibration curves with regard to relative peak height (RPH) versus nominal gap width are obtained. Based on this calibration data, the individual leaf positions are calculated to compare with the nominal programmed positions. The results demonstrate that the collimator rotation angle can be determined as accurate as 0.01 degrees. A change in MLC gap width of 0.2 mm leads to a change in RPH of about 10%. For asymmetrically produced gaps, a 0.2 mm MLC leaf gap width change causes 0.2 pixel peak position change. Subpixel resolution is obtained by using a parameterized fit of the relatively large abutment peaks. By contrast, for symmetrical gap changes, the peak position remains unchanged with a standard deviation of 0.05 pixels, or 0.026 mm. A trial run of 36 test images, each with gap widths varying from 0.4 to 1.4 mm, were used to analyze 8640 abutments. The leaf position variations were detected with a precision of 0.1 mm at a 95% confidence level, with a mean of 0.04 mm and a standard deviation of 0.03 mm. The proposed method is robust and minimizes the effect of image noise and pixel size and may help physicists to establish reliable and reasonable action levels in routine MLC QA.

Dosimetric impact of systematic MLC positional errors on step and shoot IMRT for prostate cancer: a planning study

“…Our study has demonstrated the need for accuracy in both the modelling of the MLC system in the TPS and in calibration as well as quality assurance (QA) of the MLC in order to minimize systematic errors of IMRT treatment delivery. Various MLC QA for IMRT techniques have been described in the literature [12,16,25,26] and recommendations for modelling of MLC have been proposed by the AAPM Task Group No. 50 [27].…”

Task Group 142 report: Quality assurance of medical acceleratorsa)

“…Recent development of tools utilizing EPID devices allow for subpixel precision to detect changes in discrete locations of an acquired image. 69,70 As treatment planning parametrization seeks a global value for leaf transmission, it is important that the leaf body, side, and end characteristics do not change over time, the most vulnerable being the leaf side rigidity due to leaf inderdigitation, as it may affect interleaf leakage, hence the close attention needed. Leaf position repeatability, MLC spoke shot, and coincidence of light field and x-ray field all are tests intended to check the alignment of the MLCs.…”

A novel phantom and procedure providing submillimeter accuracy in daily QA tests of accelerators used for stereotactic radiosurgery*

“…Since pixels of kV and MV images are about 0.4 mm or larger and CBCT slices are 1 mm thick, small BBs cannot provide one‐tenths‐mm accuracy. However, it has been demonstrated that with computer aid the position of regular‐shaped objects that extend over many pixels can be found with accuracy substantially better than pixel size 9 , 12 . The center of a sphere having a radius of about 10 times the pixel dimensions, e.g., can be determined with accuracy of one‐tenth pixel size (13) .…”