Premium
Accuracy of patient dose calculation for lung IMRT: A comparison of Monte Carlo, convolution/superposition, and pencil beam computations
Author(s) -
Vanderstraeten Barbara,
Reynaert Nick,
Paelinck Leen,
Madani Indira,
De Wagter Carlos,
De Gersem Werner,
De Neve Wilfried,
Thierens Hubert
Publication year - 2006
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.2241992
Subject(s) - monte carlo method , pinnacle , nuclear medicine , dosimetry , computation , radiation treatment planning , convolution (computer science) , imaging phantom , mathematics , pencil (optics) , algorithm , physics , radiation therapy , computer science , statistics , medicine , optics , radiology , artificial intelligence , artificial neural network
The accuracy of dose computation within the lungs depends strongly on the performance of the calculation algorithm in regions of electronic disequilibrium that arise near tissue inhomogeneities with large density variations. There is a lack of data evaluating the performance of highly developed analytical dose calculation algorithms compared to Monte Carlo computations in a clinical setting. We compared full Monte Carlo calculations (performed by our Monte Carlo dose engine MCDE) with two different commercial convolution/superposition (CS) implementations (Pinnacle‐CS and Helax‐TMS's collapsed cone model Helax‐CC) and one pencil beam algorithm (Helax‐TMS's pencil beam model Helax‐PB) for 10 intensity modulated radiation therapy (IMRT) lung cancer patients. Treatment plans were created for two photon beam qualities (6 and 18 MV ). For each dose calculation algorithm, patient, and beam quality, the following set of clinically relevant dose‐volume values was reported: (i) minimal, median, and maximal dose ( D min , D 50 , and D max ) for the gross tumor and planning target volumes (GTV and PTV); (ii) the volume of the lungs (excluding the GTV) receiving at least 20 and 30 Gy ( V 20 and V 30 ) and the mean lung dose; (iii) the 33rd percentile dose ( D 33 ) and D max delivered to the heart and the expanded esophagus; and (iv) D max for the expanded spinal cord. Statistical analysis was performed by means of one‐way analysis of variance for repeated measurements and Tukey pairwise comparison of means. Pinnacle‐CS showed an excellent agreement with MCDE within the target structures, whereas the best correspondence for the organs at risk (OARs) was found between Helax‐CC and MCDE. Results from Helax‐PB were unsatisfying for both targets and OARs. Additionally, individual patient results were analyzed. Within the target structures, deviations above 5% were found in one patient for the comparison of MCDE and Helax‐CC, while all differences between MCDE and Pinnacle‐CS were below 5%. For both Pinnacle‐CS and Helax‐CC, deviations from MCDE above 5% were found within the OARs: within the lungs for two ( 6 MV ) and six ( 18 MV ) patients for Pinnacle‐CS, and within other OARs for two patients for Helax‐CC (for D max of the heart and D 33 of the expanded esophagus) but only for 6 MV . For one patient, all four algorithms were used to recompute the dose after replacing all computed tomography voxels within the patient's skin contour by water. This made all differences above 5% between MCDE and the other dose calculation algorithms disappear. Thus, the observed deviations mainly arose from differences in particle transport modeling within the lungs, and the commissioning of the algorithms was adequately performed (or the commissioning was less important for this type of treatment). In conclusion, not one pair of the dose calculation algorithms we investigated could provide results that were consistent within 5% for all 10 patients for the set of clinically relevant dose‐volume indices studied. As the results from both CS algorithms differed significantly, care should be taken when evaluating treatment plans as the choice of dose calculation algorithm may influence clinical results. Full Monte Carlo provides a great benchmarking tool for evaluating the performance of other algorithms for patient dose computations.