A statistics-based model for prediction of achievability of the planning criteria for IMRT planning

The purpose of this study is to develop and evaluate a statistics-based model to predict the achievability of the planning criteria for intensity-modulated radiation therapy planning. A statistics-based model was proposed to predict the achievability of the planning criteria based on the structure set. A retrospective study was performed to validate the proposed model. A total of 160 prostate cases and 134 nasopharynx (NP) cases were used as the training set to build the model while 200 cases for each treatment site were used to validate the proposed model. An overlapping ratio and the minimum distance between organ at risks (OARs) and planning target volumes were used to predict the achievability of the planning criteria for serial organs. Since both prostate and NP cases were treated with simultaneous boost using multiple targets with different prescription doses, the effect of each parameter on the OARs was studied by binary logistic regression. For parallel organs, average distance to dose level was introduced and it described the dose falloff gradient within each OAR. By studying the distribution of average distance to dose level in the training set for each OAR, the result was used to predict the dose volume and average dose criteria. The accuracy of the proposed model was evaluated by comparing the predicted proportion and the actual proportion of the validation group that can achieve the clinical planning criteria. For prostate cases, the differences between the actual and predicted proportions were small for all criteria of rectum and bladder. The maximum deviations were 7% for bladder, 9% for rectum, and 5% for femur. For NP cases, the difference ranged from 0% to 7% with the largest difference found at the criterion D50% < 30 Gy of the parotids. In conclusion, the proposed statistics based plan prediction model was presented and the results showed the model could achieve acceptable prediction accuracy. The proposed tools could be used to calculate the probability of the OARs to achieve the clinical goals. This statistics based model can be further developed to adapt into the planning objective estimation in the auto-planning process.