Evaluating plan robustness for organ deformation and set-up uncertainties in head and neck cancer

Abstract

Purpose or Objective In radiotherapy, a planning CT (pCT) is used to plan treatment, however the pCT only shows a snapshot of the patient anatomy at a particular point in time. As the treatment is delivered in multiple fractions several days after the pCT is acquired, it is expected that the patient’s internal anatomy will change throughout the treatment course as organs may deform. Additionally, patient positioning on the couch may vary for each fraction, e.g. due to set-up uncertainty. These two factors cause uncertainties in the dose absorbed by the patient. In this study we model these uncertainties and simulate the joint effect they have on treatments. Material and Methods Data from 20 patients were used. Organ deformations were modelled using 6 independent weekly principal component analysis (PCA) models based on the pCT and 6 weekly conebeam CT scans from 13 patients. To account for time trends in organ deformations, 5 daily vector fields were simulated from each of the 6 weekly PCA models. Systematic and random set-up uncertainties were modelled as Gaussian distributions with standard deviations of Σ = 0, 1, 2 mm and σ = 1, 2 mm respectively. We simulated 100 30-fraction treatments using the clinical plan for each patient, accounting for: 1) only organ deformations, 2) only set-up errors, and 3) both organ deformations and set-up errors. The difference in the maximum brainstem dose and mean dose to each parotid from the planned dose to the simulated doses was evaluated for the training and validation (n = 7) datasets separately. Results Figure 1 shows the difference in the resulting DVH parameters from the planning value for all simulated scenarios. As expected, the DVH parameter differences with both set-up uncertainties and organ deformations were always larger than where each uncertainty was considered separately. Considering only organ deformations resulted in the smallest ranges, except for (Σ, σ) = (0, 1) mm, and similar distributions for (Σ, σ) = (1, 1) mm considering only set-up errors. For simulations considering set-up uncertainties, the range of the DVH parameter differences increases most as Σ increases and to a lesser extent as σ increases for both the training and validation data sets. The overall distributions are similar for both the 13 training patients and the 7 validation patients. Conclusion We evaluated the effect of time-varying organ deformation and set-up uncertainties on clinical head and neck cancer plans. Our data suggests that the effect of organ deformation may be modelled by considering (Σ, σ) = (1, 1) mm for this dataset. We found that for small set-up uncertainties, (Σ, σ) = (0, 1) mm, organ deformations are the dominant contribution to DVH parameter differences for the parotids and the brainstem. This highlights the need to correct for organ deformations in modern image-guided radiotherapy workflows, where patient set-up errors are being considerably reduced.