Automation in radiation therapy treatment planning for prostate cancer

Abstract

Cancer of the prostate (Ca prostate) is one of the most common cancers in men. One of the main treatment options for this disease is radiation therapy. The treatment planning of radiation therapy starts with image acquisition and is followed by image registration, delineation of the targets and organs at risk, and creation of a computer treatment plan. Both the contouring process and computer planning process are time consuming and suffer from operator dependency. Automation of these elements is therefore desirable to solve such problems. In this thesis, we aimed to study the feasibility and benefits of automating the treatment planning process. The contouring process was automated by a commercial Atlas Based Auto-Segmentation (ABAS) system. The performance of the system was studied and the parameters used were optimized. Thirty Ca prostate cases were recruited to evaluate the auto-contouring system, and the performance was quantified by using contour similarity measures. A novel perimeter-based index for evaluating the similarity of structure contours was proposed, while the relationships of the proposed index and other commonly used indexes with the contouring time saved were compared. Both our proposed index and the Dice Similarity Coefficient indicated that the use of ABAS with our recommended setting can produce structure contours with reasonable accuracy that can reduce the contouring time and the workload of staff. To achieve automation in computer treatment planning, the paramount step is to accurately predict the amount of normal tissue sparing according to the geometric relationships with the treatment targets. A statistics-based model was developed to predict the achievability of the planning criteria for intensity modulated radiation therapy (IMRT) planning. The model was validated by retrospectively studying 200 Ca prostate cases and 200 nasopharyngeal carcinoma cases that were treated in our centre with the IMRT technique. A high degree of accuracy was achieved with the prediction model and it was further developed to be adopted to achieve automation in computer treatment planning. The automation script for the entire computer planning process was developed with the use of the Varian Eclipse Scripting Interface. This script can produce treatment plans without any manual input, covering beams selection, creation of virtual structures, dose-volume histogram estimation, optimization and dose calculation. Fifty Ca prostate cases that were treated with the IMRT technique were selected as the validation cases. Automation plans were created with the script and were compared with the manual plans. It was found that the automation script can produce the same or better treatment plans than those produced by our human planners. In conclusion, although the contouring process still cannot be fully automated and requires manual editing, our results showed that the use of ABAS can reduce the contouring time. Also, the computer planning process can be fully automated by the automation script. The resultant plans are of equal or better quality and are more consistent than those produced by our human planners.