Automated Analysis of Leaf Position Accuracy for Multi-Leaf Collimator Quality Assurance
W Du*, K Casey, R Kudchadker, University of Texas MD Anderson Cancer Center, Houston, TXSU-E-T-134 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
The traditional QA method for checking the multi-leaf collimator (MLC) positional accuracy is to expose a picket-fence (PF) pattern on a piece of film. Evaluation of the film is often cumbersome or qualitative. We have developed a method that is simpler, more accurate, and capable of analyzing images from both static gantry and rotational gantry (i.e., VMAT) PF tests.
PF tests with static gantry or VMAT delivery were performed on a Varian 21EX linac. The resulting EPID images were analyzed with an in-house Matlab program. First, the PF lines were detected with a Hough transform algorithm. Then the leaf pairs were identified by fitting the image intensity profile of interleaf leakage with the ideal profile. For each leaf pair, the positions and widths of the PF gaps were computed by fitting the individual gaps with a Lorentzian function. Finally, the VMAT PF was overlaid and compared with the static gantry PF.
In the static gantry PF images, the distances between adjacent MLC PFs were <0.1 mm from the expected values. The distances between the MLC PFs and the jaw PF were <0.3 mm from expectation. The maximum angular difference between all PFs was 0.03 degrees. In one occasion the software detected a 0.34 mm gap position error. The corresponding gap width was 0.40 mm smaller than the neighboring gap widths. The misaligned leaf was identified automatically. For the VMAT PF tests, the distances between the VMAT PFs and static gantry PFs were <0.5 mm, while the angular differences were <0.2 degrees.
Software was developed to analyze the static gantry and VMAT PF images. The software is sensitive and is able to detect leaf position errors of <0.4 mm. The overall leaf position accuracy is excellent in both static gantry and VMAT delivery.