Room: AAPM ePoster Library
Purpose: The tuning for the weighting factors in the objective function for an IMRT plan is driven by humans to achieve an ideal dose distribution. This is done manually in a gradual and time-consuming process based on trial and error and user experience. In the past fuzzy inference system (FIS) relying on membership functions were used to tuning the weighting factors with promising results. Nevertheless, FIS is too slow due to the number of iterations. This work aims to propose a method to predict the weighting factors using an unsupervised learning method based on data generated by a FIS.
Methods: A dynamic membership function generator was implemented to translate linguistic humans tag (i.e. High or low dose) for different types of organs (target volume, the organ at risk and normal tissue) into a degree of truth. The membership function was generated using an iterative algorithm implemented in MATLAB using the central point for a sigmoid membership function. Then, FIS guided inverse planning system was implemented to generate data finding optimal combinations of weighting factors in the objective function for an IMRT plan. This data was used to train a neural network to predict the optimal weighing factors.
Results: The performance of the algorithm was examined utilizing the C-Shape TG119 IMRT phantom using the variations of weighting factors and mean dose versus the iteration number as well as dose-volume histograms. Fuzzy logic guided inverse planning system is capable of finding the optimal combination of mean dose and weighting factors for different anatomical structures involved in treatment planning.
Conclusion: It is possible to use a feasible way to automatically predict the weighting factors for an IMRT plan under the guidance of FIS using a neural network without human intervention other than providing the treatment plan parameters and set of constraints.
Not Applicable / None Entered.
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation