The purpose of this study was to systematically evaluate dose distributions computed with 5 different dose algorithms for patients with lung cancers treated using stereotactic ablative body radiotherapy (SABR). Treatment plans for 133 lung cancer patients, initially computed with a 1D-pencil beam (equivalent-path-length, EPL-1D) algorithm, were recalculated with 4 other algorithms commissioned for treatment planning, including 3-D pencil-beam (EPL-3D), anisotropic analytical algorithm (AAA), collapsed cone convolution superposition (CCC), and Monte Carlo (MC). The plan prescription dose was 48 Gy in 4 fractions normalized to the 95% isodose line. Tumors were classified according to location: peripheral tumors surrounded by lung (lung-island, N=39), peripheral tumors attached to the rib-cage or chest wall (lung-wall, N=44), and centrally-located tumors (lung-central, N=50). Relative to the EPL-1D algorithm, PTV D95 and mean dose values computed with the other 4 algorithms were lowest for "lung-island" tumors with smallest field sizes (3-5 cm). On the other hand, the smallest differences were noted for lung-central tumors treated with largest field widths (7-10 cm). Amongst all locations, dose distribution differences were most strongly correlated with tumor size for lung-island tumors. For most cases, convolution/superposition and MC algorithms were in good agreement. Mean lung dose (MLD) values computed with the EPL-1D algorithm were highly correlated with that of the other algorithms (correlation coefficient =0.99). The MLD values were found to be similar to 10% lower for small lung-island tumors with the model-based (conv/superposition and MC) vs. the correction-based (pencil-beam) algorithms with the model-based algorithms predicting greater low dose spread within the lungs. This study suggests that pencil beam algorithms should be avoided for lung SABR planning. For the most challenging cases, small tumors surrounded entirely by lung tissue (lung-island type), a Monte-Carlo-based algorithm may be warranted.

Purpose/Objective: To determine whether individualization of dose prescriptions based on tumor size and location is a feasible approach for SBRT-based treatment of non-small cell lung cancer.Materials and Methods: Treatment plans for 133 NSCLC patients treated using 12 Gy/Fxn x 4, and planned using a pencil-beam (1Dequivalent-path-length, EPL-1D) algorithm were retrospectively calculated with a Monte Carlo (MC)-based algorithm using EPL-1D plan parameters.4D imaging was performed to generate an ITV.PTV margin was 5 mm.For each plan, generalized equivalent dose (gEUD) and tumor control probability (TCP with SF=0.28) were computed.Tumors were stratified according to peripheral ('island', N=39), lungwall (attached to the rib-cage, N=44) and central locations (N=50), as well as tumor size and mean plan field size.Results: GEUD values for the EPL-1D algorithm were based on prescription doses and ranged from 48-49 Gy for all cases.MCcomputed gEUD decreased significantly with decreasing tumor size; values in Gy were 41.9+3.6 for average plan field widths (FW) between 3-5 cm; 43.8+3.2(5>FW>7 cm); 45.9+3.3 between (7>FW>10 cm).TCP's for the EPL-1D algorithm increased with decreasing tumor size with values of 0.88+0.04(3>FW>5 cm); 0.85+0.03(5>FW>7 cm); 0.80+0.05(7>FW>10 cm).On the other hand, MC-computed TCP's remained relatively constant as a function of tumor size; values were 0.77+0.07(3>FW>5 cm); 0.76+0.05(5>FW>7 cm); 0.76+0.05(7>FW>10 cm).Results were consistent for 'island', lung wall and central tumors.There was no correlation between tumor size and local control rate among the 92% of patients who were controlled locally at 2-years, which is consistent with the MC-calculated TCPs.Note, however, that MC-computed gEUDs were significantly lower for smaller tumor sizes.This suggests that although smaller tumors were receiving much lower doses (on average, 10.5 Gy vs. 12 Gy per fxn), they were being controlled.These findings are consistent with radiobiology; smaller tumors have fewer tumor cells and therefore require lower doses to be controlled relative to larger tumors.Results are also suggestive that dose could be tailored according to size and location of the tumor, i.e. one size does not need to fit all.We speculate as well that the concept of minimally ablative dose (MAD) is perhaps not an unreasonable approach to consider for smaller tumors.Dose deescalation for smaller tumors, based on the mimimum ablative dose is likely to be clinically relevant for tumors situated close to normal organs (e.g.ribs, centrally located airways) and for patients being treated for recurrent disease.

PURPOSE:We hypothesize that PTV margin dose is an important factor for local tumor control. We evaluated dose distributions for patients originally treated with pencil-beam (PB)-based plans and retrospectively calculated with Monte Carlo (MC) method, with emphasis on the spatial region between the ITV and PTV (PTV-margin), where the largest dose differences were expected. METHODS:Forty-six stage I-II lung cancer patients with 51 lesions treated with SABR were retrospectively analyzed (23 central and 28 peripheral tumors). All patients received 4DCT imaging, and an ITV was generated from the maximum intensity projection and subsequent review of four 4DCT phases. An isotropic 3mm ITV-to-PTV margin was used. The iPlan TPS was used to generate the original treatment plans using PB-based heterogeneity correction. MC doses were recalculated using the same MUs as in the PB plan. Dose distributions for the ITV, PTV-margin, and PTV were analyzed using generalized equivalent uniform dose (gEUD) with a = - 20. Student's paired t-test elucidated differences between PB and MC-based gEUD and the two different tumor locations. RESULTS:Mean ITV and PTV volumes were 24.2 cc (range: 2.2 to 99.3 cc) and 50.4 cc (range: 6.4 to 229.7 cc), respectively. The mean gEUDs of ITV, PTV-margin and PTV, normalized to PB-based 100% isodose were 1.02+/-0.04, 1.01+/-0.04 and 1.01+/-0.04 for PB-based plans, compared to 0.94+/-0.06, 0.88+/-0.08 and 0.90+/-0.08 (all p<0.05) for MC-based plans. The maximum overestimations with the PB algorithm in the PTV-margin average dose were 10.4% and 19.6% (p < 0.05) for peripheral tumor cases and central tumor cases, respectively. CONCLUSIONS:PB-based dose distributions showed the highest dose overestimation (relative to MC) in the PTV-margin spatial region. Analysis of spatial dose differences is an important precursor toward assessment of patterns-of-local failure, to be investigated in future work to explore possible association between dose and regions of failure. Acknowledgement: supported in part by grants from NIH R01 CA106770 and from Varian Medical Systems.

PURPOSE:Steep dose gradients and high dose per fraction in stereotactic ablative radiation therapy (SABR or SBRT) necessitate highly accurate tumor localization. This study evaluates inter-fraction shifts, as defined by couch correction analysis, and investigates the effect of tumor location and internal target volume (ITV) on these shifts. In addition, residual errors associated with post-CBCT correction and their dosimetric consequences were quantified.METHODS:Daily free-breathing (FB) CBCT images used for daily localization of 78 patients with non-small cell lung cancer were retrospectively evaluated. Among the population, 39 patients also received pre-treatment kV images after CBCT alignment. ITV inter-fraction displacement was evaluated by matching the CBCT and the FB helical CT images, and setup errors were quantified using orthogonal kV images. Associations between ITV location and inter-fraction motion were studied by categorizing tumors into the following locations: chest-wall seated (CWS) and island, peripheral, central, or upper, middle and lower. Dosimetric consequences for the patient with the largest setup error were explored.RESULTS:ITV inter-fraction motion included the mean of the systematic error, ?inter=(-1.4, 2.0, 1.6) mm, standard deviation (SD) of the systematic error, Σinter=(2.1, 4.2, 2.9) mm, and SD of random errors, sinter=(2.2, 3.2, 3.6) mm. No significant associations were observed between inter-fraction shifts and tumor location or volume. Using CBCT for image guidance reduced the observed errors to μsetup=(-0.3, 0.1, 0.0) mm, Σsetup=(0.6, 0.6, 0.4) mm and ssetup=(1.2, 0.7, 0.7) mm. Dosimetric consequences for the patient with the largest setup error were explored. It was shown that a 3.0 mm setup margin was sufficient to provide greater than 95% dose coverage to the ITV.CONCLUSION:CBCT image guidance reduced setup errors significantly such that 2-3 mm, population-based, setup margins provided proper dose coverage to the ITV. Further investigation of inter-and intrafraction error classification by tumor location is warranted.

PURPOSE:Given the differences in tumor size and location, encountered in lung SBRT, we hypothesize that 'one dose fractionation regimen does not fit all', i.e. that there is a role for patient-specific dose prescription based on optimization of biological models.METHODS:Sixty one NSCLC patients (tumor volume 46.5+/-47.3 cc) treated with stereotactic body radiotherapy (48 Gy in 4fx) were retrospectively studied. Clinically treated plans were generated using Brainlab's Pencil Beam (PB-BL), and then recalculated with fixed MUs using Anisotropic Analytic Algorithm (AAA), Pencil Beam (PB-EC), Monte Carlo (MC) and Collapsed-Cone-Convolution (CCC). DVHs were exported to calculate TCP (Poisson) and NTCP (Lyman-Kutcher-Burman). TCP/NTCP model parameters were utilized from published data. For each dose distribution two dose response curves were generated by scaling the prescription dose and assuming a linear relationship between the prescription dose and entire 3D dose distribution. In addition, associations were assessed between changes in each algorithm's TCP relative to PB-BL, target diameter, and local density (density of the 70% isodose covering the PTV).RESULTS:For PB-BL, mean TCP was 99.6%±0.9%, whereas for same MUs, mean TCP for PB-EC, AAA, CC and MC plans were 96.5±14.3%, 74.6±31.6%, 74.4±32.4% and 76.8±32.0%, respectively. With the same prescription dose for all plans, TCP values changed to 98.1±8.7%, 96.5±15.3%, 77.5±28.6%, 85.4±25.8% and 92.9±20.1% for PB-BL, PB-EC, AAA, and CCC, MC, respectively, indicating that AAA and CCC dose distributions are likely less homogeneous relative to MC. The TCP improvement was 12.3%, 8.9% and 4.4% for AAA, CCC and MC-based plans when the average NTCP before optimization was set as the upper limit for lung toxicity.CONCLUSIONS:This work supports patient-specific dose prescription strategies, based on biological optimization, for lung SBRT. However, further investigation is warranted. Acknowledgement: supported in part by a grant from Varian Medical Systems.

Purpose/Objective(s)To report on patterns of local failure as a function of dose computed with 5 algorithms in stereotactic ablative radiotherapy ("SABR" as coined by Loo and Timmerman et al., PRO, 2011) of Non-small cell lung cancers.Materials/MethodsFrom 2003 to 2009, 250 patients with NSCLC were treated with SABR at our institution. Dose regimen was 12 Gy x 4 in most cases; 12 Gy x 3 was used for some centrally located and recurrent disease sites. Twenty patients presented with PET-CT documented recurrence at different time points following radiation. All patients were planned with a 1-D pencil beam (PB) algorithm and 4D-CT simulation from which an ITV was generated to manage tumor motion. Using the same treatment plan and MU as in the clinical, 1-D PB case, dose was re-computed using the following algorithms: 3-D PB, anisotropic analytical algorithm (AAA), collapsed cone convolution (CCC), and Monte Carlo (MC). All algorithms were commissioned and calculations were found to be within 2%/2 mm of water phantom-based measurements. Follow-up CT scans were obtained for the 20 patients who recurred. The regions of recurrence (Vrecur) were contoured on follow-up CTs and fused with original treatment plans. For each patient, the dose received within Vrecur was quantified from the DVH of the recurrent volume. Regions of failure were classified following Chan et al. (JCO, 2002) according to the volume of Vrecur falling within the high-dose region (95% IDL on the original plan) as follows: "Central" (>95% Vrecur); "In-field" (80% > Vrecur >95%); "Marginal" (20% > Vrecur >80%); "Distant" (<20% Vrecur).ResultsThe 2-year local control rate was 92.6%. For patients who failed locally, regions of recurrence were quantified for all dose algorithms (Table). Patterns of failure were similar between the 1-D and 3-D PB algorithms; approximately 35% of the failures were central/in-field, and 65% were marginal/distant. Results were substantially different from those of the AAA, CCC, and MC algorithms, where patterns were consistent and showed a much smaller incidence of central/in-field failures (15%), with marginal/distant failure rate of ∼85%.ConclusionsTableNumbers of failures and regions according to the classification of Chan et al. (JCO, 2002)Regions and No. of failures1D-Pencil beam3D-Pencil beamAAA (conv/superpos.)CCC (conv/superpos.)Monte CarloCentral44111In-Field32222Marginal1010867Distant3491110 Open table in a new tab Purpose/Objective(s)To report on patterns of local failure as a function of dose computed with 5 algorithms in stereotactic ablative radiotherapy ("SABR" as coined by Loo and Timmerman et al., PRO, 2011) of Non-small cell lung cancers. To report on patterns of local failure as a function of dose computed with 5 algorithms in stereotactic ablative radiotherapy ("SABR" as coined by Loo and Timmerman et al., PRO, 2011) of Non-small cell lung cancers. Materials/MethodsFrom 2003 to 2009, 250 patients with NSCLC were treated with SABR at our institution. Dose regimen was 12 Gy x 4 in most cases; 12 Gy x 3 was used for some centrally located and recurrent disease sites. Twenty patients presented with PET-CT documented recurrence at different time points following radiation. All patients were planned with a 1-D pencil beam (PB) algorithm and 4D-CT simulation from which an ITV was generated to manage tumor motion. Using the same treatment plan and MU as in the clinical, 1-D PB case, dose was re-computed using the following algorithms: 3-D PB, anisotropic analytical algorithm (AAA), collapsed cone convolution (CCC), and Monte Carlo (MC). All algorithms were commissioned and calculations were found to be within 2%/2 mm of water phantom-based measurements. Follow-up CT scans were obtained for the 20 patients who recurred. The regions of recurrence (Vrecur) were contoured on follow-up CTs and fused with original treatment plans. For each patient, the dose received within Vrecur was quantified from the DVH of the recurrent volume. Regions of failure were classified following Chan et al. (JCO, 2002) according to the volume of Vrecur falling within the high-dose region (95% IDL on the original plan) as follows: "Central" (>95% Vrecur); "In-field" (80% > Vrecur >95%); "Marginal" (20% > Vrecur >80%); "Distant" (<20% Vrecur). From 2003 to 2009, 250 patients with NSCLC were treated with SABR at our institution. Dose regimen was 12 Gy x 4 in most cases; 12 Gy x 3 was used for some centrally located and recurrent disease sites. Twenty patients presented with PET-CT documented recurrence at different time points following radiation. All patients were planned with a 1-D pencil beam (PB) algorithm and 4D-CT simulation from which an ITV was generated to manage tumor motion. Using the same treatment plan and MU as in the clinical, 1-D PB case, dose was re-computed using the following algorithms: 3-D PB, anisotropic analytical algorithm (AAA), collapsed cone convolution (CCC), and Monte Carlo (MC). All algorithms were commissioned and calculations were found to be within 2%/2 mm of water phantom-based measurements. Follow-up CT scans were obtained for the 20 patients who recurred. The regions of recurrence (Vrecur) were contoured on follow-up CTs and fused with original treatment plans. For each patient, the dose received within Vrecur was quantified from the DVH of the recurrent volume. Regions of failure were classified following Chan et al. (JCO, 2002) according to the volume of Vrecur falling within the high-dose region (95% IDL on the original plan) as follows: "Central" (>95% Vrecur); "In-field" (80% > Vrecur >95%); "Marginal" (20% > Vrecur >80%); "Distant" (<20% Vrecur). ResultsThe 2-year local control rate was 92.6%. For patients who failed locally, regions of recurrence were quantified for all dose algorithms (Table). Patterns of failure were similar between the 1-D and 3-D PB algorithms; approximately 35% of the failures were central/in-field, and 65% were marginal/distant. Results were substantially different from those of the AAA, CCC, and MC algorithms, where patterns were consistent and showed a much smaller incidence of central/in-field failures (15%), with marginal/distant failure rate of ∼85%. The 2-year local control rate was 92.6%. For patients who failed locally, regions of recurrence were quantified for all dose algorithms (Table). Patterns of failure were similar between the 1-D and 3-D PB algorithms; approximately 35% of the failures were central/in-field, and 65% were marginal/distant. Results were substantially different from those of the AAA, CCC, and MC algorithms, where patterns were consistent and showed a much smaller incidence of central/in-field failures (15%), with marginal/distant failure rate of ∼85%. ConclusionsTableNumbers of failures and regions according to the classification of Chan et al. (JCO, 2002)Regions and No. of failures1D-Pencil beam3D-Pencil beamAAA (conv/superpos.)CCC (conv/superpos.)Monte CarloCentral44111In-Field32222Marginal1010867Distant3491110 Open table in a new tab

Radiation tolerance of the spinal cord has been studied in various animal models. The results seemed to be quite consistent among different animals. Clinical data for spinal cord complications have also been documented. However, these data were available only in the relatively low dose range. On the other hand, data points with complication probability between 0 and 15% were also sporadic in animal studies due to limited number of samples used in each dose group. Assuming that the radiobiological behaviors of the spinal cords for different species are similar, this study aimed to develop a normal tissue complication probability (NTCP) model of the spinal cord from the combination of these data, and apply it to treatment of patients with spine metastases. Various cord injury data from animal experiments in dog, monkey, pig, rat, and clinical data in humans were compiled. An NTCP model, expressed as NTCP = 1/(1+(D50/UBED)k), was used to fit the entire data set, where D50 and k are fitting parameters, and UBED are uniform and biologic equivalent dose. The linear quadratic model, with a fitting parameter "α/β", was used to convert doses of various fractionations to the biological equivalent dose. The power law, with a fitting parameter "a", was used to convert doses to UBED for the volume effect. The cord volume was normalized to a single vertebral section. For clinical data, we assumed that 4 vertebral sections were involved. The model was then used to compare the cord NTCP between the IMRT and AP/PA plans for various prescription doses and fractionations. The model seemed to fit the data quite well. The fitting parameters were: α/β = 3.4, a = 0.19, D50 = 145 Gy and k = 14. For a prescription dose of 18Gy in a single fraction, NTCP were 8.5E-9 and 2% for IMRT and AP/PA plans, respectively, for one vertebral section. The corresponding NTCP were 4.8E-10 and 2.1E-4, when 10 fractions were used with a prescription dose of same tumor control. When the prescription dose increased to 26 Gy in one fraction, the corresponding NTCP were 0.01% and 99.7%, respectively. They were 7.4E-6 and 86% when a 10-fraction scheme was used. Assuming the radiobiological behaviors of the spinal cord of all species are similar, this preliminary investigation suggests that the spinal cord NTCP is still low as the dose is increased above 18 Gy. A prescription delivered in a single fraction was found to have a higher NTCP than the one with same BED applied in 10 fractions.

Purpose: To use equivalent uniform dose (EUD) and tumor control probability (TCP) to retrospectively analyze the radiobiological effect of target volumes in patients with NSCLC planned and treated with Stereotactic Body Radiotherapy (SBRT). Methods: Eighty‐three stage I–II lung cancer patients with 86 lesions treated with SBRT were retrospectively analyzed. For each patient, a Pencil Beam (PB) algorithm‐based treatment plan with a dose regimen of 12 Gy/fraction in 4 fractions was generated. To overcome the known uncertainties of conventional PB algorithm in lung tissue, Monte Carlo (MC) treatment plans were also created in the iPlan (BrainLab) system using the same monitor units derived from the PB‐based plan. Niemierkoˈs EUD and TCP (Poisson model) were computed using different surviving fraction (SF) parameters for each dose calculation algorithm. The radiobiological effects of target volume were analyzed by correlating EUD and TCP with PTV volumes. Results: Mean PTV volume was 39.31 +/− 28.96 cc. The mean PB EUDs were 50.61, 50.60, 50.57 and 50.55 Gy for SF parameter values of 0.36, 0.34, 0.3 and 0.28, compared with MC EUDs 43.97, 43.84, 43.56 and 43.41 Gy. The mean PB TCP values were up to 18% higher than the mean MC TCP. EUDs calculated using both PB and MC were not sensitive to SF parameters, whereas they were for TCP calculation. Overall, MC EUDs were more sensitive to PTV volumes than PB EUDs, measured by Pearson correlation 0.46 vs. 0.07. Larger PTV volume decreased PB TCP values while this was not the case for MC TCP Conclusions: This work demonstrates encouraging evidence that radiobiological effect of target volume and dose calculation algorithm selection is significant in EUD and TCP estimations. Further studies confirming this relationship and relating to treatment outcomes are warranted.Work supported in part by NIH R01 CA106770

PURPOSE Tumor control probability (TCP) models have been proposed to evaluate and guide treatment planning. However, they are usually based on the dose volume histograms (DVHs) of the planning target volume (PTV) and may not properly reflect the substantial variation in tumor burden from the gross tumor volume (GTV) to the microscopic extension (ME) and to the margin of PTV. In this study, the authors propose a TCP model that can account for the effects of setup uncertainties and tumor cell density decay in the ME region. METHODS The proposed TCP model is based on the total surviving clonogenic tumor cells (CTCs) after irradiation of a known dose distribution to a region with a CTC distribution. The CTC density was considered to be homogeneous within the GTV, while decreasing exponentially in the ME region. The effect of random setup uncertainty was modeled by convolving the dose distribution with a Gaussian probability density function, represented by a standard deviation, sigma. The effect of systematic setup uncertainty was modeled by summing each calculated TCP for all potential offsets in a Gaussian probability, represented by sigma. The model was then applied to simplified cases to demonstrate the concept. TCP dose responses were calculated for various GTV volumes, DVH shapes, CTC density decay coefficients, probabilities of lymph node metastasis, and random and systematic errors. The slopes of the dose falloff to cover the CTC density decay in the ME region and the margins to compensate setup errors were also analyzed in generalized cases. RESULTS The sigmoid TCP dose response curve shifted to the right substantially for a larger GTV, while modestly for cold spots in DVH. A dose distribution with a uniform dose within the GTV, and a linear dose falloff in the ME region, tended to cause a minimal TCP deterioration if a proper dose falloff slope was used. When the dose falloff slope was less steep than a critical slope represented by kT, the D50, which is the prescription dose at TCP=50%, and gamma50, which is the TCP slope at TCP=50%, varied little with different dose falloff slopes. However, both D50 and gamma50 deteriorated fast when the slopes were steeper than kT. The random setup error tended to shift the TCP curve to the right, while the systematic error tended to compress the curve downward. For combined random and systematic errors, we demonstrated that based on the model, a margin of mean square root of (0.75 sigma)2 + (1.15 sigma)2 added to the GTV was found to cause a TCP change corresponding to 2% drop at TCP=90%, or 0.5 Gy shift in D50. CONCLUSIONS This study conceptually demonstrated that a TCP model incorporating the effects of tumor cell density variation and setup uncertainty may be used to guide radiation treatment planning.

Purpose/Objective(s)Heterogeneity-corrected (HC) Pencil beam (PB) calculations overestimate dose of lung tumors for stereotactic body radiation therapy (SBRT) in comparison to Monte Carlo (MC) calculations. The degree of difference between these calculations depends upon tumor size and tumor location. In this study, we explored the relationship between tumor/field size (FS), tumor location and this overestimation in dose calculation.Materials/MethodsTwenty-five stage I NSCLC patients treated with 48 Gy in 4 fractions using SBRT were retrospectively analyzed on this IRB approved study. PB and MC algorithms were commissioned on the iPlan treatment planning system (BrainLab). Plans for each algorithm without HC (unit-density) were also generated. The associated dose distributions and dose-volume-histograms (DVHs) were equivalent. HC-PB plans were then calculated with the same beam geometry. Using the PB-derived MUs, the HC-MC plans were re-calculated. In the Eclipse planning system (Varian Medical Systems), the 60% isodose volumes (i.e. isodose level volume minus primary tumor volume) from the iPlan HC-PB plans were exported to Pinnacle (Philips Radiation Oncology Systems) to determine average mass density (ρ60) for these volumes. Multiple isodose volumes were tested; however, the ρ60 was found to ideally quantify tumor location. FS (in cm) was obtained from the planning system. Clinically relevant dose-volume parameters were analyzed. Multivariate linear regression was employed.ResultsPTVs ranged from 4.9 to 202.5 cc (mean 41.9 cc). FS ranged from 2.1 to 7.3 cm (mean 3.9 cm). ρ60 ranged from 0.170 to 0.837 g/cc (mean 0.493 g/cc). For all patients, the average mean dose, minimum dose and dose to 95% of the PTV (D95) were 8.2% (SD 5.1%; range 0.6 - 21.6%), 18.5% (SD 8.6%; range 2.7 - 38.7%), and 14.6% (SD 6.9%; range 2.7 - 31.8%) higher for PB than MC, respectively. The largest differences were observed for “island” tumors (i.e. small tumors surrounded by lung tissue). Linear regression models were generated of the percent difference in D95 (ΔD95) as a function of either ρ60 or FS. As the ρ60 increased, these differences decreased. Similarly, as FS increased, these differences decreased. Using multivariate linear regression analysis, the ΔD95 between the PB and MC calculations was found to be a function of ρ60 and FS [ΔD95 = (4.46/ρ60) - (0.047* FS) +7.26, R2 = 0.83].ConclusionsThese results present a novel model using tumor location, as parameterized by the ρ60, and field size to predict the overestimation in dose calculations with PB in comparison to MC for SBRT lung tumor plans. In addition, these findings further our understanding of the dosimetric relationship between these algorithms. Future studies will extend to other algorithms. Purpose/Objective(s)Heterogeneity-corrected (HC) Pencil beam (PB) calculations overestimate dose of lung tumors for stereotactic body radiation therapy (SBRT) in comparison to Monte Carlo (MC) calculations. The degree of difference between these calculations depends upon tumor size and tumor location. In this study, we explored the relationship between tumor/field size (FS), tumor location and this overestimation in dose calculation. Heterogeneity-corrected (HC) Pencil beam (PB) calculations overestimate dose of lung tumors for stereotactic body radiation therapy (SBRT) in comparison to Monte Carlo (MC) calculations. The degree of difference between these calculations depends upon tumor size and tumor location. In this study, we explored the relationship between tumor/field size (FS), tumor location and this overestimation in dose calculation. Materials/MethodsTwenty-five stage I NSCLC patients treated with 48 Gy in 4 fractions using SBRT were retrospectively analyzed on this IRB approved study. PB and MC algorithms were commissioned on the iPlan treatment planning system (BrainLab). Plans for each algorithm without HC (unit-density) were also generated. The associated dose distributions and dose-volume-histograms (DVHs) were equivalent. HC-PB plans were then calculated with the same beam geometry. Using the PB-derived MUs, the HC-MC plans were re-calculated. In the Eclipse planning system (Varian Medical Systems), the 60% isodose volumes (i.e. isodose level volume minus primary tumor volume) from the iPlan HC-PB plans were exported to Pinnacle (Philips Radiation Oncology Systems) to determine average mass density (ρ60) for these volumes. Multiple isodose volumes were tested; however, the ρ60 was found to ideally quantify tumor location. FS (in cm) was obtained from the planning system. Clinically relevant dose-volume parameters were analyzed. Multivariate linear regression was employed. Twenty-five stage I NSCLC patients treated with 48 Gy in 4 fractions using SBRT were retrospectively analyzed on this IRB approved study. PB and MC algorithms were commissioned on the iPlan treatment planning system (BrainLab). Plans for each algorithm without HC (unit-density) were also generated. The associated dose distributions and dose-volume-histograms (DVHs) were equivalent. HC-PB plans were then calculated with the same beam geometry. Using the PB-derived MUs, the HC-MC plans were re-calculated. In the Eclipse planning system (Varian Medical Systems), the 60% isodose volumes (i.e. isodose level volume minus primary tumor volume) from the iPlan HC-PB plans were exported to Pinnacle (Philips Radiation Oncology Systems) to determine average mass density (ρ60) for these volumes. Multiple isodose volumes were tested; however, the ρ60 was found to ideally quantify tumor location. FS (in cm) was obtained from the planning system. Clinically relevant dose-volume parameters were analyzed. Multivariate linear regression was employed. ResultsPTVs ranged from 4.9 to 202.5 cc (mean 41.9 cc). FS ranged from 2.1 to 7.3 cm (mean 3.9 cm). ρ60 ranged from 0.170 to 0.837 g/cc (mean 0.493 g/cc). For all patients, the average mean dose, minimum dose and dose to 95% of the PTV (D95) were 8.2% (SD 5.1%; range 0.6 - 21.6%), 18.5% (SD 8.6%; range 2.7 - 38.7%), and 14.6% (SD 6.9%; range 2.7 - 31.8%) higher for PB than MC, respectively. The largest differences were observed for “island” tumors (i.e. small tumors surrounded by lung tissue). Linear regression models were generated of the percent difference in D95 (ΔD95) as a function of either ρ60 or FS. As the ρ60 increased, these differences decreased. Similarly, as FS increased, these differences decreased. Using multivariate linear regression analysis, the ΔD95 between the PB and MC calculations was found to be a function of ρ60 and FS [ΔD95 = (4.46/ρ60) - (0.047* FS) +7.26, R2 = 0.83]. PTVs ranged from 4.9 to 202.5 cc (mean 41.9 cc). FS ranged from 2.1 to 7.3 cm (mean 3.9 cm). ρ60 ranged from 0.170 to 0.837 g/cc (mean 0.493 g/cc). For all patients, the average mean dose, minimum dose and dose to 95% of the PTV (D95) were 8.2% (SD 5.1%; range 0.6 - 21.6%), 18.5% (SD 8.6%; range 2.7 - 38.7%), and 14.6% (SD 6.9%; range 2.7 - 31.8%) higher for PB than MC, respectively. The largest differences were observed for “island” tumors (i.e. small tumors surrounded by lung tissue). Linear regression models were generated of the percent difference in D95 (ΔD95) as a function of either ρ60 or FS. As the ρ60 increased, these differences decreased. Similarly, as FS increased, these differences decreased. Using multivariate linear regression analysis, the ΔD95 between the PB and MC calculations was found to be a function of ρ60 and FS [ΔD95 = (4.46/ρ60) - (0.047* FS) +7.26, R2 = 0.83]. ConclusionsThese results present a novel model using tumor location, as parameterized by the ρ60, and field size to predict the overestimation in dose calculations with PB in comparison to MC for SBRT lung tumor plans. In addition, these findings further our understanding of the dosimetric relationship between these algorithms. Future studies will extend to other algorithms. These results present a novel model using tumor location, as parameterized by the ρ60, and field size to predict the overestimation in dose calculations with PB in comparison to MC for SBRT lung tumor plans. In addition, these findings further our understanding of the dosimetric relationship between these algorithms. Future studies will extend to other algorithms.

To develop a web-based relational database for analysis of outcomes for lung patients treated with hypo-fractionated stereotactic body radiotherapy (SBRT). An outcomes-study database was designed, under IRB-approval, to collect and analyze medical information for lung cancer patients undergoing SBRT. Web Programming languages, such as ASP, HTML and Java, were used to develop the database. The database management system on the server is MS Access (Microsoft, Richmond, WA). The outcomes-study database incorporates a variety of patient information, including demographic, social, diagnostic, treatment and follow-up data. A software tool with a graphical user interface was developed to calculate biological dose indices, including TCP, Biologically Effective Dose, Equivalent Uniform Dose (Niemierko) and NTCP (Lyman-Kutcher-Burman (LKB) and relative seriality models). Medical records for approximately 250 lung cancer patients treated with SBRT have been populated into the database. Different parameters related to the outcomes of these patients, evaluated using the database, are summarized here. Analysis of results for patients with early stage disease, the majority of whom were treated with 12 Gy x 4 fractions, revealed that approximately 92% of patients were free of 2-year local failure. No grade II or greater lung toxicities (including radiation-induced pneumonitis) were observed. To study the impact of dose distribution on local failure and normal tissue complications, 100 patients were retrospectively re-planned using the Monte Carlo (MC) method. MC calculations predicted a lower minimum PTV dose by 21% on average vs. the conventional pencil beam algorithm (PB) (range, of -2.5 to 41.5%, p < 0.001). There was a strong inverse correlation between the size of the tumor and the magnitude of the underdosage; minimum target dose reductions of up to 41.5% were observed for smaller lesions, particularly those located peripherally and surrounded by lung tissue. Mean Lung Dose (MLD) and the NTCP (LKB model) were lower in the MC calculations by 10% (p < 0.001) and 15% (p < 0.001) respectively. However, given the average MLDs were only 4.57 Gy vs. 5.09 Gy with MC and PB respectively, the impact on healthy lung tissue complications is limited. One of the main goals of the outcomes database is to facilitate investigation of dose, volume and effect relationships for lung cancer patients treated with SBRT, using hypo-fractionated doses. The database has been demonstrated to be an effective tool for collecting and analyzing various aspects of patient outcome, including follow-up data and treatment planning information. As such the database can also be used to establish the efficacy of the hypo-fractionated SBRT treatment modality for early stage lung cancers.

Purpose: To use generalized equivalent uniform dose (gEUD) to evaluate the dose distributions of patients with non-small-cell lung cancer (NSCLC) planned for stereotactic body radiotherapy (SBRT). Methods and Materials: Thirty-eight stage I–II lung cancer patients with 41 lesions treated with SBRT were analyzed. For each patient a Pencil Beam (PB) algorithm-based treatment plan was produced. Monte Carlo (MC)-based treatment plans were calculated using the same number of monitor units, for each beam, as in the PB-based plan. Spearman rank correlation coefficients between gEUD with different parameter a values and dose-volume endpoints Dx and Vx were computed and compared. The effects of PTV diameter on dose differences between PB-based plans and MC-based plans were also analyzed using gEUD. Results: The Spearman rank coefficient as a function of the a value in the gEUD produced functions with unique maxima for the PTV (0.995) and normal lung tissue (0.989). For PB-based plans, the parameter a correlated best with V5, V10, V20 and D95 for a values of 0.7, 0.9, 2.2 and −24 respectively. For the MC-based plans, these values of a were 0.8, 1.1, 2.4, and −22, respectively. gEUD differences between MC- and PB-based plans were found to be inversely proportional to the PTV diameter. These differences were found to increase with decreasing values of a, as the gEUD converges to the minimum PTV dose. Conclusion: Spearman rank analysis showed good correlation between gEUD and dose-volume endpoints as a function of the a parameter in the gEUD model. Further investigation is needed to correlate these a values with biologically meaningful results for tumors and normal tissues in the context of lung SBRT in the clinical setting. Work supported in part by NIH_R01-CA106770

Purpose: To reconstruct surface models from delineations on CT images, a simple two-step algorithm is implemented. Method and Materials: Stacks of contours with CT images are extracted from DICOM files which are exported from the treatment planning system. The surface reconstruction algorithm contains two steps. The first step is to segment the CT images according to delineations via determining volumetric pixels inside or outside the contours via a simple scan line algorithm. The second step is to classify the six facets of each volumetric pixel of regions of interest, which is on or inside the contours, into inner facets and outer facets via determining a facet shared by two volumetric pixels or not. All the outer facets represent the boundary of delineated regions of interest on the CT images. The surface model is then generated from all the outer facets. Results: In comparison to other surface reconstruction algorithms, such as “Marching Cubes”, the surface reconstruction algorithm we describe here can be used to generate ‘interpolating’ surfaces other than ‘approximate’ surfaces. The surface model can also be opened or closed if outer facets depending on whether the first and last layers of a CT image are included or not. A lung case and a prostate case were tested. The largest surface model is the body surface model from the lung case which has roughly 400K triangular facets. For targets, the surface models have approximately 2K triangular facets and 20K triangular facets for the lung and prostate PTVs, respectively. Conclusions: We have implemented a simple method to reconstruct surfaces from contours. All surface models can be efficiently operated on a low-end desktop computer. The algorithm is efficient and accurate, and can be used to generate surface models for applications such as IGRT and dose-surface histogram analysis. Work supported in part by NIH_R01-CA106770

Purpose: To investigate the clinical issues associated with the use of Monte Carlo‐based prospective planning for lung SBRT and spine SRS patients. Methods: Experimental verification of the iPlan v.4.1 MC photon beam algorithm (BrainLab) was performed using film and ion‐chamber in water phantoms and solid‐water slabs containing bone and lung‐equivalent materials for a 6 MV photon beam from a Novalis linac. MC dose verification was performed for 5 spine and 7 lung patients using an anthropomorphic phantom. Treatment plans of prospectively treated patients were examined to investigate the influence of statistical uncertainties, MC‐based dose‐to‐water (Dw) and medium (Dm) and calculation speed. Results: Agreement between calculations and measurements in the water phantom verification tests was, on average, within 2%/1 mm (high dose/high gradient), and was within ±4%/2 mm in the heterogeneous slab geometries. For spine SRS, the agreement between PB and Dm calculations were within 4 % of IC measurements, however, the difference between MC_Dw and Dm was on average 9 %. For the lung SBRT tests the average difference between calculated PB doses and IC reading was 13%, and both MC calculations agreed within 1.5 % of measurements. The use of reduced uncertainty below 2% increases dose calculation time significantly but has an insignificant effect on the dose volume histograms (DVH). Conclusion: Prospective treatment planning with a well‐commissioned MC algorithm provides improved dose coverage and can be introduced for efficient calculations in the routine clinical setting. However, further investigation is warranted on issues such as dose‐to‐medium and water, the impact of statistical uncertainties on serial and parallel organs, and proper approaches for reliable dose measurements at small field sizes, under non‐equilibrium conditions. Acknowledgement: Supported in part by a grant from the NIH/NCI (R01CA106770)

Purpose: To investigate an efficient method for assessing the impact of target deviations detected by daily image‐guidance on planning volumes. We hypothesize that, using this method, on‐line daily correction may not be warranted. Method and Materials: Geometrical surfacemodels of the CTV and PTV were generated from the respective planning volumes via a surface reconstruction algorithm. Given a set of target deviations, represented by 6‐degree positioning shifts (translation, pitch, roll, yaw) from IGRT systems (e.g. Varian's OBI/CBCT), collision detection was performed to find geometric distance between CTV and PTV surfacemodels when the shifts are applied to the CTV. Since the method is based on shifts, which do not incorporate intra‐fraction motion, a criterion was defined such that on‐line shifts are not applied if the CTV and PTV surfaces are greater than a distance defined by a population‐based margin for intra‐fraction motion. Results: A software module was developed to: import IMRT plans, generate surfacemodels, communicate with the IGRT system, and calculate CTV/PTV spatial geometric distances. Surfacemodel generation is performed off‐line. Collision detection between CTV and PTV surfacemodels from daily shifts is done on‐line within minutes on a desktop PC. In the examples shown, “collisions” are defined by minimum distances of 3 mm in any direction (intrafraction motion margin) between the CTV and PTV. If a collision is detected, an on‐line shift is applied; otherwise the treatment proceeds as planned. Conclusions: The proposed method represents a quick and quantitative way to manage target deviations determined during image‐guidance, without the need to apply on‐line corrections. The approach is not limited to target analysis, but may also include normal tissues, such as the bladder and rectum. Adaptive and on‐line treatment planning studies are under way to determine the validity of the hypothesis, and to improve the collision detection criteria.

Purpose: To develop a software framework which allows analysis and evaluation of outcomes for patients treated with hypo-fractionated stereotactic body radiotherapy (SBRT) Method and Materials: An outcomes-study database was designed, under IRB-approval, to collect and analyze medical information for lung cancer patients undergoing SBRT. The outcomes-study database incorporates a variety of patient information, including demographic, social, diagnostic, treatment and follow-up data. A software tool with a graphical user interface was developed to calculate biological dose indices, including TCP, EUD (Niemierko) and NTCP (Lyman-Kutcher-Burman and relative seriality models). Biologically effective doses are determined from dose distributions and DVHs, automatically imported from the treatment planning systems. Results: Approximately 200 SBRT patients' medical records have been populated into the outcomes database. To demonstrate the functionality of the software framework, 11 lung cancer patients treated with SBRT (12 Gy/fraction × 4 fractions) for central lesions (tumors within a 2 cm zone of the proximal bronchial tree) were reviewed. The linear-quadratic (LQ) adjusted mean lung dose was 13.38 Gy (range: 9.24 Gy to 16.50 Gy) with an α/β ratio of 3.0 Gy. Observed lung toxicities were reviewed using the database and showed the regimen to be well tolerated: none of the patients developed grades 3–5 toxicities and only 2 lower-grade toxicities were noted. The calculated NTCP ranged from 1.73% to 8.22% using the LKB model, with parameters n=1, m=0.33, TD50=30.5 Gy. Conclusions: One of the goals of the outcomes database is to enable the investigation of correlation between dose, volume and effect for patients treated with SBRT, using non-standard, hypo-fractionated RT doses. Parameters specific to biological dose models are being updated using maximum likelihood analysis of calculated probabilities vs. observed outcomes for this large cohort of patients. Such outcomes studies are important in understanding the efficacy of new treatment paradigms.

Purpose: To develop an automatic re‐contouring approach for tumor volumes via surface reconstruction and smoothing. The method overcomes the limitations of traditional methods used in 3D volume reconstruction which can be quite irregular, leading to non‐physical beam intensities during inverse‐planning. Method and Materials: Stacks of contours are extracted from DICOM RS files which are exported from the treatment planning system. Surface models are generated from the contours via a surface reconstruction algorithm. The surface reconstruction algorithm contains two steps. First, contours are triangulated and medial axes are computed. Then, contours on adjacent cross sections are joined to create a surface. Gaussian and other surface smoothing operators are applied to the surface models. The smoothed models are re‐sliced with CT planes to produce new contours. Results: The Gaussian surface smoothing operator is an excellent smoothing tool but suffers from the volume shrinkage problem. To circumvent this problem, a volume‐preserving surface smoothing operation is implemented (Taubin's surface smoothing algorithm), which uses a non‐shrinking variant of the Gaussian smoothing algorithm. As noted in the figures, the method, applied to prostate volumes, shows good correspondence between the smoothed and original volumes. The mean 3D vector distance between the smoothed and original surfaces was 0.6 mm (RMS = 0.8 mm) for an example case. Conclusion: The proposed surface‐based reconstruction algorithm produces more physically realistic, smooth target volumes in 3D while maintaining the overall structure of the original volume. Future work will include the implementation and evaluation of other surface smoothing operators as well the incorporation of the smoothed 3D volumes into treatment planning to assess the impact on inverse‐planned fluence maps and IMRT dose distributions.

Purpose: Outcome analysis is an important and challenging task in radiation oncology. We have developed an integrated software platform that facilitates evaluation of outcomes for patients treated with stereotactic radiosurgery (SRS) and hypo‐fractionated stereotactic body radiotherapy (SBRT). Method and Materials: An outcomes‐study database was designed, under IRB‐approval, to store medical information of patients who undergo SRS or SBRT. The outcomes‐study database is integrated with a record‐and‐verify system (ARIA, Varian Medical Systems) and with an in‐house developed hospital database. A software package with GUI was developed using Visual Studio. Net (Microsoft, USA) to evaluate treatment plans. Biological dose models including TCP, EUD (Niemierko's model), NTCP (Lyman, F dam , relative seriality models, etc.) as well as biologically effective doses are determined from dose distributions and DVHs, automatically imported from the treatment planning systems. Patient clinical data (from physician on‐treatment and followup visits) and image data (including planning CT scans and followup CT scans) are imported for outcomes analysis. Results: Thus far approximately 700 SRS/SBRT patients medical records have been populated into the outcomes database. The database contains all medical information, including demographic, social, diagnostic, treatment and follow‐up information. As it is linked to the hospital database, treatment information encompasses multi‐disciplinary‐based treatments including surgery and chemotherapy. The web‐based interface enables access and information management remotely. The calculation of biological dose indices provides useful means to correlate dose distributions with clinical outcome with respect to tumor control and healthy tissue complications assessed from followup physical examinations as well as image data used to inspect possible recurrent disease or radiation‐induced damage. Conclusion: The goal of the database is to study the various factors of significance related to the outcomes of patients treated with SRS and SBRT, using non‐standard, hypo‐fractionated RT doses. Parameters specific to biological dose models will be updated using maximum likelihood analysis. Acknowledgement: NIH‐R01CA106770.