A COMPARISON OF CONVENTIONAL FORWARD PLANNING WITH INVERSE PLANNING FOR 3D CONFORMAL RADIOTHERAPY OF THE PROSTATE

A radiotherapy treatment plan optimisation algorithm has been applied to 48 prostate plans and the results compared with those of an experienced human planner. Twelve patients were used in the study, and 3-, 4-, 6- and 8-field plans (with standard coplanar beam angles for each plan type) were optimised by both the human planner and the optimisation algorithm. The human planner 'optimised' the plan by conventional forward planning techniques. The optimisation algorithm was based on fast simulated annealing using a cost-function designed to achieve a homogenous dose in the 'planning-target-volume' and to minimise the integral dose to the organs at risk, 'Importance factors' assigned to different regions of the patient provide a method for controlling the algorithm, and it was found that the same values gave good results for almost all plans. A study of the convergence of the algorithm is presented and optimal convergence parameters are determined. The plans were compared on the basis of both dose statistics and 'normal-tissue-complication-probability' (NTCP) and 'tumour-control-probability' (TCP). The results of the comparison study show that the optimisation algorithm yielded results that were at least as good as the human planner for all plan types, and on the whole slightly better. A study of the beam-weights chosen by the optimisation algorithm and the planner revealed differences that increased with the number of beams in the plan. The planner was found to make small perturbations about a conceived optimal beam-weight set. The optimisation algorithm showed much greater variation, in response to individual patient geometry, frequently deselecting certain beams altogether from the plan. The algorithm is shown to be a useful tool for radiotherapy treatment planning. For simple (e.g., three-field) plans it was found to consistently achieve slightly higher TCP and lower NTCP values. For more complicated (e.g., eight-field) plans the optimisation also achieved slightly better results with generally less numbers of beams, unfavourable beams being deselected from the plan. Probably the greatest benefit is the reduced time taken by the optimisation to compute optimised beam-weights. This time was always less than or equal to 5 min; a factor of up to 20-times faster than the human planner.