br Fig Differences in estimated

Fig. 2. Differences in estimated actually given dose for a PTV optimized and a robustly optimized VMAT treatment plan: typical case (a) Difference in estimated actually given dose distribution. (b) Dose–volume histogram of the estimated actually given dose for both treatment plans.
Other studies, albeit on different treatment sites, show similar results with CMRO in terms of OAR dose sparing. It was shown for CMRO in breast cancer patients that target coverage improved at identical or reduced OAR dose [6,23]. The agreement between the planned nominal dose and dose evaluated on the 4DCT was better for the CMRO-plans than the PTV-plans [6]. Archibald-Heeren et al. compared CMRO VMAT of lung cancer on a thorax phantom [24]. The authors found fewer maximum and minimum dose variations compared to other current treatment techniques such as internal target volume based planning as evaluated on a 4DCT [24]. Zhang et al. investigated the benefit of CMRO for five prostate cancer patients in which the CTV and OAR delineations were shifted inside the patient to create different scenarios [25]. Their method did not include a dose recalculation per scenario and accounted for internal target motion rather than patient posi-tioning errors. They reported a mean dose reduction of 6.4% and 19.7% for the rectal and EPZ031686 walls, respectively. Previous stud-ies on CMRO of proton therapy found larger benefits for CMRO, which was expected due to the inherent lack of dose invariance of the treatment modality [4,13,26].
In terms of toxicity probabilities, the various NTCPs decreased approximately 1–3% for the CMRO plans as compared to the PTV-plans. The NTCP models were derived from planned dose distribu- 
tions, whereas the NTCPs in this study were derived from the EAGD. Since the EAGD is potentially more directly related to toxi-city than the planned nominal dose, this could lead to a slight underestimation of the improvement in toxicity probability. How-ever, the NTCP values of the planned dose distribution were very similar to those from the EAGD (Table 1). Therefore, we expect these results to be representative.
We demonstrated that CMRO led to improved robustness com-pared to PTV-based optimization, at no expense of OAR dose. Other factors affecting the plan robustness are the magnitude of robust-ness shifts (or PTV margin) and the number of plan adaptations. In this study, we found, albeit on 4/10 patients, that CMRO led to an improved dose distribution when using plan adaptations after three weeks of treatment, indicating that CMRO can be used in combination with adaptive radiotherapy. Furthermore, taxonomy can be argued that CMRO may lead to fewer plan adaptations than PTV-based optimization. This was however not investigated.
In this study, the EAGD was derived from dose calculations on daily acquired CBCTs that were mapped to the reference CT and accumulated. We acknowledge that the deformable image regis-tration (required for dose mapping) introduces new unknown errors [27]. The dose calculation was performed using a segmenta-tion of the CBCT based on six tissue substitutes. Based on the
The average dosimetric parameters and NTCP values of the PTV-optimized and CMRO treatment plans.
PTV-plan CMRO-plan
PTV-plan CMRO-plan
PTV-plan CMRO-plan Dmean of OARs (Gy)
Abbreviations: CI = conformity index; HI = homogeneity index; CMRO = Composite minimax robust optimization; PCM = pharyngeal constrictor muscle; EAGD: estimated actually given dose. The 95% confidence interval values are given between the brackets. * Statistically significant difference between de PTV-plan and cmRO-plan, using the Wilcoxon signed-rank test adjusted using Bonferroni’s correction for multiple testing.
Fig. 3. Estimated actually given dose for PTV optimized and robustly optimized VMAT treatment plans (n = 10). Each scatter plot represents a different dose parameter. The x position of each data point corresponds to its value of the PTV optimized plan and its y position to its value of the robustly optimized plan. Therefore, data points below the diagonal indicate a lower value in the robustly optimized plan than in the PTV optimized plan. Data points shown as circles represent the planned nominal dose and diamond shapes represent the estimated actually given dose.
76 Robust optimization of VMAT
Fig. 4. Estimated normal tissue complication model outcome for PTV optimized and robustly optimized VMAT treatment plans (n = 10). Each scatter plot represents a NTCP model. The x position of each data point corresponds to its value of the PTV optimized plan and its y position to its value of the robustly optimized plan. Therefore, data points below the diagonal indicate a lower value in the robustly optimized plan than in the PTV optimized plan. Data points shown as circles represent the planned nominal dose and diamond shapes represent the estimated actually given dose.