Dose calculation of Acuros XB and Anisotropic Analytical Algorithm in lung stereotactic body radiotherapy treatment with flattening filter free beams and the potential role of calculation grid size

The International Commission on Radiation Units (ICRU) recommended an overall accuracy of 5% in the delivery of absorbed dose [22], and the American Association of Physicists in Medicine (AAPM) advised that the uncertainty in the computed dose distribution should be less than 2% [23]. Both of the AXB and AAA algorithms exhibit similar dose prediction in homogeneous media. However, AXB was reported to be more accurate in heterogeneous medium [2,5-7]. Accordingly, it is essential for us to accurately determine the dose delivered to lung tumor in which the water density target is always surrounded by low density lung tissue. For lung radiotherapy treatment, PTV in the low density range is originated either from the tumor motion uncertainty [24] or uncertainty in patient positioning and alignment of the therapeutic beams during the treatment [25]. We determined that the mean dose of PTV_lung predicted by the AXB algorithm was up to 3.34% lower compared to the AAA by using the 10XFFF beam energy. Although the dose to PTV_lung was reduced, the PTV_soft dose was not affected. We acknowledged this as it was an interesting finding. For lung SBRT treatment, the PTV_lung dose changes were of positive significance because they could be associated with significant changes in tumor control probability (TCP) and normal tissue complication probabilities (NTCP) according to the AAPM Report No. 85 [23]. Therefore, determination of the dosimetric effect with AXB on TCP will be an interesting work for our future study.

The dosimetric impact of the AXB algorithm has been investigated in several studies. Fogliata et al. [12] compared the AXB to AAA in patients with stage III NSCLC and concluded that a lower mean dose to soft tissue and a slightly higher mean dose in low-density lung tissue were obtained by the 6 MV beam with the AXB algorithm. Kan et al. [4] addressed this issue using IMRT and RapidArc techniques in nasopharyngeal carcinoma patients and found that AXB presented a 1% higher dose in air compared with AAA. The same mean target dose between the two algorithms was demonstrated by Liu et al. [15] in SBRT treatment for NSCLC patients. Kroon et al. [16] reported that D_98% of PTV was lower using AXB, whereas the mean dose of PTV was comparable for the lung SBRT treatment plan. Our finding indicating that the mean dose of PTV_lung was obviously lower with the AXB algorithm differs from the results of these studies. These contradictions are reasonable because the differences in lung doses calculated by AAA and AXB were influenced by many factors, such as beam energy, field size, field number and the density of lung tissue [4,6]. Another possibility that may lead to the inconsistency was the use of FFF beam in our research. Unlike flattened beams, FFF ones can provide softer energy and a cone-like profile. Moreover, some differences in the electron contamination parameters were acquired by removal of the flattening filter, which was not observed in the flattened beams [26]. These differences may influence the dose differences between the two algorithms and further investigation is necessary.

Additionally, we also determined that AXB estimated a lower dose for most of the OARs, whereas V₅ and V₁₀ of the ipsilateral lung were significantly higher with AXB. The higher dose prediction in ipsilateral lung could be explained by the shape of the profiles obtained by the AXB algorithm. Kan et al. [4] reported that the dose profile in air indicated a lower dose inside the field but a higher out-of-field dose predicted by AXB. When the gantry rotated around the patient, all OARs were inside the treatment field at some degrees of gantry rotation during the entire treatment process except for the ipsilateral lung. This helped to explain why V₅ and V₁₀ of the ipsilateral lung increased with the AXB algorithm. V₅ was reported to be a dosimetric factor that was predictive of radiation-induced pneumonitis (RIP) [27], however, as the dose difference was minor (0.6-1.6% higher) with the AXB and its clinical effect could be neglected.

Our finding that the CGS also contributed to the dose difference between AAA and AXB is interesting. CGS was reported to be associated with dose estimation and calculation accuracy. Chung et al. [13] and Mittauer et al. [14] found that CGS influenced dose estimation in head and neck treatment. Ong et al. [28] found that 1.0 mm CGS improved the accuracy of dose calculation compared with 2.5 mm for the AAA algorithm. Kan et al. [9] also reported that AXB with 1.0 mm CGS resulted in a significant improvement in the dose accuracy (within 3%). Smaller grid resolution reduces the averaging effect and results in better sampling of the structure voxels to the calculation grid [9]. Interestingly, our result demonstrated that CGS also influenced the relative dose difference between the AAA and AXB algorithm. By using 1 mm CGS, dose difference in PTV_lung between the two algorithms was greater compared with 2.5 mm CGS, regardless of whether the 6XFFF beam or 10XFFF beam was used. That’s due to the ability of AAA and AXB algorithms to adapt the CGS to ensure the dose accuracy in the axis perpendicular to the image slices. If the slice thickness is larger than the defined grid size, AAA and AXB may calculate the dose on dose planes between the image slices. And if the slice spacing is smaller than the defined grid size, AAA and AXB may skip calculating the dose on some slices. Based on the fact that the AXB algorithm predicts lower dose in each image slice, the dose difference between the two algorithms will undoubtedly become larger with 1 mm CGS because it involves more image slices in dose prediction. Compared with previous research, our result can more truly reflect the dosimetric difference between the two algorithms with 1 mm CGS due to its better calculation accuracy than 2.5 mm CGS. This was another factor that led to the inconsistency with other previous studies. To our knowledge, the potential impact of CGS on the dose difference between AAA and AXB has not been reported and analyzed and this study is the first to address this issue.

Our study has certain limitations. Although some recent publications have demonstrated the accuracy of AXB algorithm, few studies have verified the dosimetric performance of AXB algorithm in a SBRT plan produced from FFF beams. Our further work will focus on lung SBRT plans where small fields and air cavity are involved. Additionally, Kan et al. [9] recommended the use of 1 mm CGS instead of 2.5 mm whenever possible for stereotactic plans using a slice spacing of 1.25 mm, particularly when PTV contains relatively large areas of low density tissue. Whether different slice spacing will influence the relative dose difference between the two algorithms will be another area of interest for us.