AbstractID: 9382 Title: EPID Based Quality Assurance for Collimation Verification in External Beam Radiotherapy A methodology for automatic EPID based quality assurance (QA) for multileaf collimation (MLC) verification (i.e. leaf position and collimator angle) is presented. The clinical implementation consists of 3 features. (1) EPID image acquisition is done independently of treatment verification software (TVS) and linac. Images are acquired continuously regardless of ‘beam on’ condition, enabling an image sequence (consisting of blank, incomplete and full exposure images) to be continuously and asynchronously acquired without operator intervention. (2) An automatic image classifier is used to group the images based on collimation for comparison to the treatment planning fields. The classifier is self-organizing, allowing images acquired at time of first treatment to be grouped without input from the TVS. Images from subsequent fractions are automatically sorted into the initial groupings. Area and centroid position are used as indices. For similarity measure, cross-correlation, entropy and newly formulated ‘weighted Hausdorff’ distance are considered. (3) Leaf position is computed using a novel ‘supervised Laplacian of Gaussian’ edge detector. Collimation angle is determined using a ‘dual orientation transform’ algorithm (hybridized truncated Hough and Radon transforms). Both algorithms are not significantly affected by patient anatomy, and accuracies of 0.5mm and <0.5º are obtained. This methodology was clinically implemented, and used for pre-treatment verification of MLC and micro-MLC fields for both conformal radiotherapy and IMRT. With the resulting software, one can automatically obtain precise collimation verification of beam segments associated with a modulated field. For 60 fields, image acquisition and analysis each required 10min using a 2.0GHz PC.