Poster - Thur Eve - 37: Improved clustering MLC leaf-sequencing algorithm for step-and-shoot IMRT

This study examines how to reduce the complexity of fluence map generated using an improved clustering leaf-sequencing method, and evaluates such method in step-and-shoot intensity-modulated radiotherapy (IMRT). Based on the current equal-space grouping algorithm for multileaf collimator (MLC) leaf sequence, we proposed an improved K-means grouping algorithm which can replace the stratification routine in the program of existing leaf sequence. The improved algorithm can be thought of as a gradient descent procedure, which begins at starting cluster centroids, and iteratively updates these centroids to decrease the objective function. The K-means always converge to a local minimum depending on the starting cluster centroids. The K-means algorithm continuously updates cluster centroids until the local minimum is reached. We compare the leaf-sequencing results in term of numerical values from the improved K-means and equal-space grouping algorithm. A representative 1D intensity map from a clinical treatment plan was investigated and optimized by the K-means and equal-space grouping algorithm. It was found that the K-means algorithm decreased the dose square error from 53.67 to 35.27. Moreover, the normalized square differences of the equal-space and K-means algorithm are equal to 2881 and 1244, respectively. The results demonstrated an improvement in accuracy achievable by allowing the fluence map to be defined in an optimal way rather than using the pre-defined criteria. Therefore, the K-means leaf-sequencing algorithm can better simulate the distributions of the input fluence data compared to the traditional grouping algorithm in step-and-shoot IMRT.