Introduction: the quality of radiotherapy treatment plans varies across institutions and depends on the experience of the planner. For the purpose of intra- and inter-institutional homogenization of treatment plan quality, we present an algorithm that learns the organs-at-risk (OARs) sparing patterns from a database of high quality plans. Thereafter, the algorithm predicts the dose that similar organs will receive in future radiotherapy plans prior to treatment planning on the basis of the anatomies of the organs. The predicted dose provides the basis for the individualized specification of planning objectives, and for the objective assessment of the quality of radiotherapy plans. Materials and method: one hundred and twenty eight (128) Volumetric Modulated Arc Therapy (VMAT) plans were selected from a database of prostate cancer plans. The plans were divided into two groups, namely a training set that is made up of 95 plans and a validation set that consists of 33 plans. A multivariate analysis technique was used to determine the relationships between the positions of voxels and their dose. This information was used to predict the likely sparing of the OARs of the plans of the validation set. The predicted doses were visually and quantitatively compared to the reference data using dose volume histograms, the 3D dose distribution, and a novel evaluation metric that is based on the dose different test. Results: a voxel of the bladder on the average receives a higher dose than a voxel of the rectum in optimized radiotherapy plans for the treatment of prostate cancer in our institution if both voxels are at the same distance to the PTV. Based on our evaluation metric, the predicted and reference dose to the bladder agree to within 5% of the prescribed dose to the PTV in 18 out of 33 cases, while the predicted and reference doses to the rectum agree to within 5% in 28 out of the 33 plans of the validation set. Conclusion: We have described a method to predict the likely dose that OARs will receive before treatment planning. This prospective knowledge could be used to implement a global quality assurance system for personalized radiation therapy treatment planning.
- dose prediction algorithm
- IMRT/VMAT treatment plan optimization
- normal tissue sparing knowledge based radiation therapy treatment planning
- personalized radiotherapy treatment planning