Optimising the verification of patient positioning in proton beam therapy

A new patient positioning system incorporating a robotic arm is currently being designed for proton beam therapy. This requires aligning a treatment image with a pre-defined reference image. This is achieved by the alignment of the radiation and reference field boundaries, followed by registering the patient's anatomy relative to the boundary. This paper compares and tests methods for both field boundary and anatomy alignment. For field boundary alignment, it is proposed to use a powerful object detection algorithm known as the generalised Hough transform (GHT). It was found that the GHT algorithm, followed by an optimisation routine, worked successfully and overcame problems in existing solutions. For anatomical-body alignment, a number of intensity-based similarity measures and optimisation routines are tested. It was found that the genetic algorithm, minimising the correlation coefficient similarity measure worked the best.