ORIGINAL ARTICLE Year : 2011  Volume : 36  Issue : 4  Page : 220229 Dose volume histogram analysis and comparison of different radiobiological models using inhouse developed software Arun S Oinam^{1}, Lakhwant Singh^{2}, Arvind Shukla^{1}, Sushmita Ghoshal^{1}, Rakesh Kapoor^{1}, Suresh C Sharma^{1}, ^{1} Department of Radiotherapy, PGIMER, Chandigarh, India ^{2} Department of Physics, Guru Nanak Dev University, Amritsar, Punjab, India Correspondence Address: The purpose of this study is to compare LymanKutcherBurman (LKB) model versus Niemierko model for normal tissue complication probability (NTCP) calculation and Niemierko model versus Poissonbased model for tumor control probability (TCP) calculation in the ranking of different treatment plans for a patient undergoing radiotherapy. The standard normal tissue tolerance data were used to test the NTCP models. LKB model can reproduce the same complication probability data of normal tissue response on radiation, whereas Niemierko model cannot reproduce the same complication probability. Both Poissonbased and Niemierko models equally reproduce the same standard TCP data in testing of TCP. In case of clinical data generated from treatment planning system, NTCP calculated using LKB model was found to be different from that calculated using Niemierko model. When the fractionation effect was considered in LKB model, the calculated values of NTCPs were different but comparable with those of Niemierko model. In case of TCP calculation using these models, Poissonbased model calculated marginally higher control probability as compared to Niemierko model.
Introduction Evaluation of treatment plans for the determination of best plan among the different plans is done by analysis of dose volume histogram (DVH) as well as twodimensional and threedimensional spatial dose distributions. These plans are further evaluated by calculating the tumor control probability (TCP) and normal tissue complication probability (NTCP) of the treatment plans to determine the radiobiological ranking of different plans amongst them. [1],[2],[3] This final evaluation of treatment plans can be done by using both cumulative and differential DVH generated by treatment planning system (TPS). Various researchers developed their own inhouse evaluation software for TCP and NTCP calculation. [3],[4],[5],[6] Some of these models do not consider the fractionation effect in NTCP calculations, [2],[7] whereas some of the models use the effect of fractionation in radiotherapy treatment in NTCP [5],[6],[8] and TCP [5],[6],[9] calculation. The purpose of this study is to develop a comprehensive and userfriendly inhouse computer program for DVH analysis and clinical implementation of NTCP calculation based on LymanKutcherBurman (LKB) model and Poisson distribution model based TCP calculation and its comparison with those of calculations based on Niemierko models. Further, the incorporation of different dose per fraction sizes in LKB model will be tested. Materials and Methods Software requirements and data input/output A MATLAB® software version 7.1 was used to develop this program. The differential and cumulative DVH data were exported from Eclipse TPS of Varian Medical System, Palo Alto, USA, in ASCII format in dose bin size of 20 cGy and these DVH data were imported in MATLAB software for the estimation of different DVH parameters, TCP and NTCP. Normal tissue complication probability based on dose response curve LymanKutcherBurman model Lyman's formula models the sigmoidshaped dose response curve of NTCP as a function of dose (D) to a uniformly irradiated fractional reference volume (v ref ). The parameters used in this model are TD 50/5 (dose at which probability of complication becomes 50% in 5 years), m (tissuespecific parameter inversely proportional to the slope of response curve) and n [parameter to find the equivalent uniform dose (EUD) of inhomogeneous irradiation using DVH reduction method proposed by KutcherBurman model]. [10],[11] The expression of this NTCP is given as [INLINE:2] The volumedependent parameter TD 50/5 (v) for fractional volume v can be expressed in terms of TD 50/5 (1) of full volume irradiation as [INLINE:3] where 0 < n < 1 for all tissues fitted by Burman et al. [7] The above parameters are fitted on this model by Burman et al. [7] for the normal tissue tolerance data of highgrade complications associated with full or partial organ irradiation, compiled by Emami et al. [12] The values used in this study are summarized in [Table 1]. The dose response data of Emami et al. are derived mostly from nominal doses near to 2 Gy per fraction. Dose fractionation effects are not explicitly taken into account in this LKB model calculation of NTCP.{Table 1} Application of fractionation in LymanKutcherBurman model The raw data obtained from TPS are of different dose per fraction. The radiobiological effects of such different dose per fraction size are different from those of 2 Gy per fraction for the same radiation dose distribution in an organ. In order to consider the effect of fractionation and size of dose per fraction in the NTCP calculation, the DVH of different doses per fraction is converted into biologically equivalent physical dose of 2 Gy per fraction (EQD 2 ) using the linear quadratic (LQ) model as [INLINE:4] where nf and df = D/nf are the number of fractions and dose per fraction size of the treatment course, respectively. α/β are the tissuespecific LQ parameters of the organ being exposed. EQD 2 DVH data obtained using equation (4) were used to calculate the LKB model based NTCP by Kuperman et al., [13] using equations (1)(3). Niemierko model In this model also, the raw data from TPS are converted into the biological equivalent physical EQD 2 DVH using equation (4). Then, this DVH is converted into the DVH of the whole volume of the organ receiving an EUD, using the DVH reduction method proposed by Kutcher et al. [10] The EUD, obtained from the conversion of inhomogeneous dose distribution of different partial volume vi receiving the dose Di, is given as [INLINE:5] The value of a is equal to 1 when EUD is equal to mean dose. The "a" is a large negative value for tumor as the tumor control depends on the minimum dose received by the tumor. In case of normal tissues such as serial and parallel architectures, the values of a are large positive and small positive values depending on small and large volume effects, respectively. Then, the NTCP of such organ is determined using a logistic function as [INLINE:6] where γ50 is the slope of sigmoidal dose response curve of normal tissue at 50% complication probability [INLINE:7] The tissuespecific parameters given in [Table 1] are used for NTCP calculation based on Niemierko model. [8] Tumor control probability Poisson's model of cell killing Survival of cell killing by radiation exposure follows the Poisson's distributions. [14] The probability of survival of clonogenic cell that does not receive any hit (N = 0) after the exposure of radiation dose D is given by [INLINE:8] where Nc.p(D) is the average number of hits on Nc (clonogenic cells) due to small p(D) (probability of hit per cell). If it is assumed that cell killing follows twotarget model of singletrack and multiple track events, the probability of survivability of clonogenic cells after Nf, number of exposures of dose per fraction d, is given by [INLINE:9] Similarly, TCP depends on the number of survived clonogenic cells Ns and small survival fraction ps(D), and TCP is given by the probability of average number of clonogenic cells survived (Ns = 0 = Nc.ps(D)) as [INLINE:10] The number of clonogenic cells can be found out using the following relation according to Niemeirko and Goiten [9] and Stavrev et al. [15],[16] [Appendix A]. If the clonogenic cell data are not available, the expression of TCP in equation (10) can also be rewritten in terms of sigmoidal dose response parameters according to Warkentin et al. [4] and Stavrev et al. [15],[16] as [INLINE:11] where TCD 50 is the tumor dose required to produce 50% TCP and γ50 is the slope of dose response at 50% TCP, which is [INSIDE:1]. For a heterogeneous irradiation of independent subvolumes vi of tumors with dose Di, the overall TCP is given by [INLINE:12] Using equation (11), the TCP using Poissonbased model is given by [INLINE:13] The tumorspecific parameters used in this study are given in [Table 2]. [17]{Table 2} Niemierko model based on equivalent uniform dose After converting the tumor DVH data of different dose per fraction size into equivalent physical DVH of 2 Gy per fraction using equation (4), the inhomogeneous dose distribution of different dose bin Di irradiating small volume vi is reduced into an EUD of the whole volume of the tumor using equation (5). The TCP of the tumor [9] of such dose distribution is given by [INLINE:14] Results and Discussion Dose volume analysis The flow charts of MATLAB programs for dose volume analysis and TCP and NTCP calculation software are shown in Appendix B. For DVH analysis and LKB model based NTCP calculation software, the required input data is cumulative DVH, whereas Poisson's model based TCP calculation software uses the differential DVH. In NTCP and TCP calculation using Niemierko model, differential DVHs are used. This software is compatible for any type of DVH. It can convert any type of DVH into a particular DVH depending on the requirement of model to be used for analysis. [Figure 1] shows the output of cumulative DVH of this software, reconstructed from the differential DVH of TPS. This figure also shows percentage difference of calculated cumulative DVH from that of original DVH. The deviations of both the calculated DVHs from those of original DVHs are less than 0.006% except a small spike of −2.35% at an initial dose bin of differential DVH, reconstructed from cumulative DVH. The output of DVH analysis parameters is shown in [Table 3]. The DVH parameters calculated using this software are D 100 , D 95 , D 2/3 , D 1/2 , D 1/3 , and D 1cc , (dose to 100%, 95%, 66.66%, 50%, 33.33%, 1 cc volume of the organ, respectively). Mean, maximum and minimum dose (D mean , D max and D min ) to the organ are also calculated by it.{Table 3}{Figure 1} Validity checking of NTCP calculation using Emami et al.'s data The output of this comprehensive inhouse developed mfile program of MATLAB software for NTCP and TCP calculation is shown in [Figure 2]. [Table 4] shows the validity checking of normal tissue tolerance data of Emami et al. [12] NTCP calculation of LKB model for bladder using normal tissue tolerance dose data points of TD 5/5 (normal tissue tolerance dose of 80 Gy and 65 Gy to 2/3 volume and whole volume of bladder, respectively, for 5% complication occurrence within 5 years of radiation exposure) of Emami et al. [12] was found as 4.76%, which is approximately equal to 5%, whereas Niemierko model calculated it to be 40.96%. 50% complication of bladder for the data of tolerance dose (TD 50/5 of 85 Gy and 80 Gy to 2/3 volume and whole volume of bladder, respectively) was found exactly as NTCP of 50% was obtained using LKB model, whereas Niemierko model calculated higher complication probability of 70.12%. In selecting the data point of dose corresponding to volume lesser than 2/3 volume of bladder, it is assumed that any volume lesser than 2/3 volume of bladder receiving the dose corresponding to 2/3 volume of bladder will produce the same radiobiological effects.{Table 4}{Figure 2} Similarly, in case of rectum, NTCP calculation of LKB model using TD 5/5 and TD 50/5 of rectum was found to be 4.78% and 50%, respectively, whereas Niemierko model underestimated NTCP to be 0.99% for TD 5/5 and 50% complication was exactly calculated for TD 50/5 . The complication probability calculation of spine using LKB model corresponding to TD 5/5 and TD 50/5 was found as 6.63% and 45.44%, respectively, whereas those calculated using Niemierko model were found to be 0.40% and 48.17%, respectively. Both TD 5/5 and TD 50/5 were overestimated by Niemierko model in case of lung and heart. The overestimations of lung were 86.48% and 99.12% for TD 5/5 and TD 50/5 , respectively. Similarly, for heart, TD 5/5 and TD 50/5 are calculated as 49.60% and 90.29%, respectively. The values for parotid, [18],[19] spine and rectum were found to be underestimated as 0.25%, 0.40% and 0.99%, respectively, for TD 5/5 . However NTCPs of bladder and brainstem were overestimated as 37.91% and 13.58%, respectively, corresponding to TD 5/5 . In all cases of normal tissues, LKB model reproduces the same complication probabilities of normal tissue tolerance data of Emami et al. [12] NTCP calculation output using TPS data A few brief examples of the output of this software are shown in [Table 5]. These consist of DVH parameters and NTCPs calculated by this software using the normal tissue dose response and tumor dose response parameters. [Figure 3] and [Table 5] show DVH calculation of rectum for four different prostate adenocarcinoma treatment plans. According to relative cumulative DVHs and DVH parameters of rectum, plan (2) has to produce the minimum NTCP and NTCP has to be increased in the order: plan (2), plan (4), plan (3) and plan (1). The NTCPs calculated using LKB models are 2.73%, 2.88%, 3.07% and 4.66%, respectively, in the increasing order for the above order of plans, whereas Niemierko models calculated the NTCPs to be 16.48%, 15.84%, 3.17% and 5.76%, respectively. The first two plans used the number of fractions of 27, whereas the last two plans used the number of fractions of 37. Since the fractionation effect is not taken into account in the LKB model, the first two plans produce lesser NTCPs as compared to the last two plans. But when the fractionation effect is taken into account using equation (4) by converting the cumulative dose of different dose per fraction into the equivalent dose of 2 Gy per fraction, the first two plans predict NTCPs of 12.10% and 12.25%, respectively, whereas the last two plans are estimated to predict 3.04% and 4.95%, respectively, which are approximately predicted by Niemierko model.{Table 5}{Figure 3} NTCPs of different organs at risk (bowel and bladder for four different cases of prostate adenocarcinoma treatments; spinal cord, brainstem, ipsilateral parotid, larynx and contralateral parotid for two cases of a head and neck radiotherapy treatment, and spinal cord, heart, liver and lung for two treatment cases of squamous cell carcinoma of esophagus) are also shown in [Table 5]. Similarly, the DVH parameters [Table 5] and cumulative DVH [Figure 4] and [Figure 5] of these organs can predict NTCPs in LKB model. Considering the fractionation effect by Niemierko and LKB models, the calculated NTCPs were comparable to each other in all organs at risk.{Figure 4}{Figure 5} The complication probabilities of ipsilateral parotids were found to be 94.38%, 99.84% and 93.55% for plan (1) and 54.87%, 98.44% and 38.90% for plan (2), when they were calculated using LKB, Niemierko and LKB fractionation models, respectively. DVH parameters also showed relatively larger volume of parotid irradiated to higher dose in plan (1) as compared to plan (2). The magnitudes of NTCPs amongst these three models were comparable in plan (1) but varied largely. Both the LKB models calculated smaller than that of Niemierko model in case of plan (2). This is occrred due to the use of a single value of EUD for the whole volume of organ, derived using KutcherBurman DVH reduction method in Niemierko model. [5] The decrease in the values of NTCPs of both the LKB models [2,7] is due to the considerations of every fractional volumes and the corresponding equivalent doses reduced from physical doses using the same DVH reduction method. Complication probability calculated using LKB fractionation model was lesser than that of LKB model due to the reduction of the physical dose of parotid into a relatively lesser dose of EQD 2 . But in case of plan (1), EQD 2 dose of different fractional volumes of DVH was approximately equal to the physical dose. TCP calculation [Table 6] presents the calculated values of TCPs of TPS DVH data using Poissonbased model and Niemierko model for the treatment of adenocarcinoma of prostate, squamous cell carcinoma of oropharynx and esophagus tumor. The TCP values calculated using Poissonbased model were marginally larger than those of Niemierko model (within 5%), except the lesser estimation (within 0.6%) for macroscopic tumor of oropharynx and esophagus tumor and microscopic tumor of adenocarcinoma of prostate (within 0.06%). When the DVHs reconstructed from Okunieff et al.'s data [17] were used for TCP calculation, 50% TCPs of microscopic [20] and macroscopic tumor of adenocarcinoma [17],[21] of prostate were reproduced as 48.31% and 50.25% by both Poisson and Niemierko models. Both the TCP calculation models calculated the same TCP values of 49.32% for all microscopic tumors of squamous cell carcinoma. [17] In case of squamous cell carcinoma of macroscopic tumor of head and neck [22],[23],[24],[25] and esophagus, [26],[27],[28] TCP values were calculated as 49.86% and 49.32%, respectively, by both Poisson and Niemierko TCP calculation models.{Table 6} Conclusion This software developed using MATLAB platform can be used as a userfriendly program to estimate the DVH parameters, TCP and NTCP values, for the ranking of different plans. From the above discussion, it can be concluded that Niemierko model cannot predict the same normal tissue complication data of Emami et al., [12] whereas LKB model can predict the same complication data. Both Poissonbased model and Niemierko model for TCP calculation equally reproduced the same TCP of Okunieff et al.'s data. [17] But in case of clinical data generated from TPS, NTCPs calculated using LKB model were found to be different from those of Niemierko model. When the fractionation effect is considered in LKB model, the calculated NTCPs were lesser than those of LKB model which does not take into account fractionation, but both LKB and Niemierko models were comparable to each other. In case of TCP calculation using these models, Poissonbased model calculated marginally higher control probability as compared to Niemierko model. [INLINE:1] References


