Vector-model-based case retrieval approach for improving and expediting optimization in intensity modulated radiotherapy

Abstract

A similarity reference database is a database of previous radiotherapy treatment cases. Currently there is no such database for intensity modulated radiotherapy (IMRT) or volumetric arc radiotherapy (VMAT). Every IMRT/VMAT plan is based on an individual oncologist's preferences of prescription dose and an individual planner's experience. Without a similarity reference database, it is impossible to manually identify, retrieve and assess all similar IMRT/VMAT cases from thousands of patient records. This forces a planner to use the trial-and-error method to search for optimization parameters and overcome the dose difference between final accurate dose calculation and fast optimization dose calculation which normally takes several days. Because of the long planning time, this means that adaptive IMRT/VMAT planning is unfeasible without a reduction in planning time. With a similarity reference database, a vector model could retrieve previously successful radiotherapy cases that share various anatomical/physiological features with the current case. Using the optimization parameters from those references as a template for the current case, the IMRT/VMAT optimization time can be reduced and the plan quality can be guaranteed. In this study, two similarity reference databases were first created from 100 previous static field IMRT cases and 100 previous VMAT cases. A vector model was then created using a combination of features extracted from the cases' CT images and structure contours in the DICOM-RT files. After the vector model was completed, the similarity between a present case and each reference case was measured by the direction cosine between their feature vectors. Planning parameters were retrieved from the selected most similar reference case and applied to the present case to bypass many gradual adjustments of optimization parameters. Prostate cases were replanned with both the conventional manual optimization and the vector-model-based optimization based on the oncologists' clinical dose prescriptions. A total of 360 plans (30 cases of IMRT, 30 cases of 1-arc VMAT, and 30 cases of 2-arc VMAT plans including first optimization and final optimization with/without the vector-model-based optimization) were compared using the two-sided t-test and paired Wilcoxon signed rank test with a significance level of 0.05 with a false discovery rate of less than 0.05. For IMRT, 1-arc VMAT and 2-arc VMAT prostate plans, there was a significant reduction in the planning time with the vector-model-based optimization by 2 hours (p = 3.4x10ˉ⁶), 5.44 hours (p = 4.6x10ˉ⁷) and 2.77 hours (p = 1.7x10ˉ⁶), respectively. Similarly, the number of iterations was significantly reduced with the vector-model-based optimization. From the first optimization plans comparison, CTV D99 of IMRT and 1-arc VMAT with the vector-model-based optimization was 0.7 Gy higher than that of the conventional manual optimization. The volume receiving 35 Gy in the femoral head for 2-arc VMAT plans was reduced by 10% with the vector-model-based optimization compared to the conventional manual optimization approach. Otherwise, the quality of plans from both approaches was comparable. From the results, the vector model approach of former cases retrieval was shown to expedite the optimization of IMRT/VMAT while maintaining the plan quality.