The Shape of Parotid DVH Predicts the Entity of Gland Deformation During IMRT for Head and Neck Cancers

Transcription

1 Technology in Cancer Research and Treatment ISSN Volume 14 Number 6 December June 16. Epub ahead of print. The Shape of Parotid DVH Predicts the Entity of Gland Deformation During IMRT for Head and Neck Cancers DOI: /tcrt The Jacobian of the deformation field of the registration between images taken during Radiotherapy is a measure of compression/expansion of the voxels within an organ. The Jacobian mean value was applied to investigate possible correlations between parotid deformation and anatomical, clinical and dosimetric parameters. Data of 84 patients were analyzed. Parotid deformation was evaluated through Jacobian maps of images taken at the start and at the end of the treatment. Several clinical, geometrical and dosimetric factors were considered. Correlation between Jacobian mean value and these parameters was assessed through Spearman s test. Univariate and multivariate logistic analyses were performed by considering as the end point the first quartile value of the Jacobian mean value. Parotid dose volume histograms were stratified according to gland deformation, assessing the most predictive dose-volume combination. At multivariate analysis, age (p ), overlap between tumor volume and parotid gland (p ) and the parotid volume receiving more than 10 Gy (p ) were found as the best independent predictors, by considering Jacobian mean value first quartile as the end point. By comparing the average dose volume histogram of parotids with Jacobian mean value first quartile and fist quartile, the parotid volume receiving more than 10 Gy and 40 Gy were found as the most predictive dosimetric parameters. Parotid glands were divided in three different sub-groups (bad-, medium- and good dose volume histogram). The risk to have Jacobian means value lower than first quartile was 39.6% versus 19.6% versus 11.3% in these three groups. By including in the multivariate analysis this dose volume grouping parameter, age and bad dose volume histogram were found as the most predictive parameters for large shrinkage. The pattern of parotid deformation may be well predicted by some pre-treatment variables; a bad dose volume histogram seems the most important predictor. Key words: Dose volume histogram, Head-and-neck cancer, Intensity modulated radiation therapy, Parotid gland deformation. S. Broggi, M.Sc. 1 E. Scalco, Ph.D. 2 C. Fiorino, M.Sc. 1 M. L. Belli, M.Sc. 1 G. Sanguineti, M.D. 3 F. Ricchetti, M.D. 3 I. Dell Oca, M.D. 4 N. Dinapoli, M.D. 5 V. Valentini, M.D. 5 N. Di Muzio, M.D. 4 G. M. Cattaneo, M.Sc. 1 * G. Rizzo, Ph.D. 2 1 Medical Physics Department, San Raffaele Scientific Institute, Milan, Italy 2 Istituto di Bioimmagini e Fisiologia Molecolare, CNR, Segrate, Milano, Italy 3 Radiation Oncology and Molecular Radiation Sciences Department, The Johns Hopkins University, Baltimore, MD, USA 4 Radiotherapy Department, San Raffaele Scientific Institute, Milan, Italy 5 Radiotherapy Department, Università Cattolica S. Cuore, Roma, Italy Abbreviations: ASD: Average Symmetric Distance between Contours; AUC: Area Under the Curve; CI: Confidence Interval; CT: Computerized Tomography; CTV: Clinical Target Volume; D_mean: Mean dose; DVH: Dose Volume Histogram; DSC: Dice Similarity Coefficient; FFD: Free-form Deformation; FDG-PET: Fluorodeoxyglucose Positron Emission Tomography; HN cancer: Head-and- Neck cancer; IMRT: Intensity Modulated Radiation Therapy; IPV: Initial Parotid Volume; Jac: Jacobian of the Deformation Field of the Registration; Jac_mean: Jacobian Mean Value; JVH: Jacobian Volume Histogram; KVCT: KiloVoltage-CT; MVA: Multivariate Analyses; MVCT: MegaVoltage- CT; OR: Odds Ratio; OVPPTV: Overlap between the Parotid Gland and the PTV; PTV: Planning Target Volume; Q1: First Quartile Value of the Population; ROC: Receiver Operating Characteristic; SIB: Simultaneous Integrated Boost; V X : Volume receiving more than x Gy. *Corresponding author: G. M. Cattaneo, M.Sc. Phone: Fax: cattaneo.mauro@hsr.it 683

2 684 Broggi et al. Introduction Head and neck cancer patients may undergo significant anatomic changes and deformations over the course of a radiation treatment. Shrinkage of the primary tumor, weight loss and parotid gland shrinkage with a shift of the centre of mass towards the midline were widely reported (1-6). The advent of on-board volumetric imaging devices provided patient pre-treatment volumetric images that allow the better quantification of the dosimetric consequences of the interfraction anatomical modifications. Several studies (2, 8, 9) have reported that the dose deliv ered to healthy tissues significantly increased during head-and-neck IMRT. Adaptive radiotherapy is a conceptually attractive approach to correct and compensate interfraction anatomical and dosimetric uncertainties, in order to limit the deviations between planned and cumulative delivered dose distributions. A few studies have reported that replanning improved both target coverage and healthy tissue sparing (2, 7-10). Frequent replanning may also result in better sparing of parotid glands; however improvement could quickly saturate beyond one-two replannings (7, 10). The manual trial-and-error approach to the fine-tune of planning parameters is time-consuming and is usually considered impractical, especially for online adaptive RT. Li et al. (11) defined an automatic re-planning algorithm to generate, starting from original fluence patterns, a plan with similar, or possibly better, DVH curves compared with the original plan. Due to the potential adaptation of the treatment, in terms of both potential escalation of tumor dose and modification of critical structures, the modeling and the prediction of anatomical deformations during intensity modulated radiation therapy (IMRT) for HN cancer are of high interest. In the case of modifications of critical structures, adaptive planning may reduce the risk of acute and late effects (1, 2, 7), or if adaptation is not feasible, the early assessment of anatomy modifications may permit to better predict and manage the toxicities with supportive therapies. Although several studies have attempted to demonstrate the advantage of adaptive radiotherapy, its clinical implementation remains limited mainly because this approach still requires demanding and time consuming procedures. Precise practical guidelines for identifying patients who would benefit more from an adaptive approach and/or for defining the best adaptive strategy have jet to be defined. The assessment of pre-treatment parameters predicting significant anatomic variations could provide useful information either to better guide planning optimization and to a more careful selection of patients with a higher probability of experiencing significant deformations, thus potentially candidate to re-planning. In a previous study (12), the possibility to assess a predictive model for parotid shrinkage, estimated in terms of absolute and relative parotid volume variation, was investigated, including clinical, dosimetric and geometric parameters. In this paper, parotid deformation was quantitatively assessed (in terms of compression and expansion of each single voxel) by the Jacobian of the deformation field of the elastic registration between images taken during treatment and correlated to anatomical, clinical and dosimetric parameters. Materials and Methods Patient Data and Imaging Procedure Data of 168 parotid glands of 84 patients from three institutions treated for different HN cancers, both with radical and adjuvant intent, were pooled. Median patient s age was 59 years (range: years). Most of the patients (79/84, 94%) didn t undergo upfront surgery. 75/84 (89.2%) patients received chemotherapy: in 18 patients as neoadjuvant treatment while concomitant to radiotherapy in 70 cases. 12 patients received both neoadjuvant and concomitant chemotherapy. Based on treatment intent, different clinical target volumes (CTVs) were defined: elective node chains (CTV2) and tumor volume including possible nodes or tumor bed in case of adjuvant intent (CTV1). In one of the centers, the highrisk node s CTV was also defined and for some radical cases the FDG-PET positive volume was boosted in another center. In all the institutions, for all considered CTVs an isotropic expansion of 5 mm was applied for planning target volume (PTV) definition. 66 patients were treated with a simultaneous integrated boost approach (SIB), delivering Gy ( Gy/day) on elective node chains (PTV2) and Gy ( Gy/day) on tumor or tumor bed (PTV1). The rest of the patients were treated with a sequential approach delivering Gy ( Gy/day) on PTV2 and Gy ( Gy/day) on PTV1. All patients were treated with IMRT: 41 patients with the helical tomotherapy Unit, 43 with the conventional Linac IMRT, in dynamic or step-shoot modality. An inverse planning optimization approach was considered with the goal to deliver at least 95% of the prescribed dose to at least 95% of each PTVs volume, while keeping the dose as homogeneous as possible and sparing parotids without compromising PTVs coverage.

3 Parotid Deformation Prediction 685 Parotid deformation was evaluated through images taken at start and at the end of the treatment. MVCT images were generally acquired with 4-6 mm slice thickness; KVCT images with 3-5 mm slice thickness. End Point: Parotid Deformation Parotid deformation (in terms of compression or expansion of each single voxel) was estimated as the Jacobian of the deformation field between the start (MVCT/KVCT_1) and the final image (MVCT/KVCT_2). Very briefly, KVCT (or MVCT) acquired at the first fraction was registered to KVCT (or MVCT) acquired at the end of the treatment, using a non rigid approach based on free-form deformation (FFD) and B-splines (13, 14). Preliminarily, accuracy of image registration was assessed both for MVCT and KVCT images, using the same approach described in (13, 14). Parotid contours manually drawn on each final image were compared with parotid contours automatically generated on the same CT by a dedicated contour propagation algorithm (14). This correspondence analysis gave similar results for both acquisition techniques: registration error, measured as average symmetric distance, was 1.72 mm for MVCT images and 1.78 mm for KVCT images. The parotid gland volume was defined using manual contours, delineated on the first fraction images. In this volume, the entity of the deformation of each voxel of the parotid gland volume was assessed by the determinant of the Jacobian (Jac) of the deformation field calculated by image registration. Jac(x) is defined as: [Jac(x) 5 det(i 1 T(x))] where T(x) is the global spatial transformation between the two images, I denotes the identity matrix and the transformation gradient. Jac quantitatively assessed the shrinkage or the expansion of each image voxel: Jac 5 1 identifies voxels whose volume doesn t change; Jac 1 shrinking voxels and Jac 1 expanding voxels. Similarly to the classical DVH, the Jacobian volume histogram (JVH) was then introduced (13) in order to assign at each bin the fraction volume with a compression or expansion larger than that one indicated by the bin itself. The Jacobian is a less used index in this context, compared to more standard contour coefficients such as the dice similarity coefficient (DSC) (15) or the average symmetric distance (ASD) between contours (16, 17). JVH Variation and Correlations with Clinical/Geometric and Dosimetric Data For each parotid gland the Jacobian map was calculated and a number of parameters were assessed such as: mean Jac value (Jac_mean), fraction of voxels with Jac 1 (compression), fraction of voxels with Jac 1 (expansion). The JVH was computed and the fraction of parotid with Jac X with X ranging between 0 and 1 (compression, step: 0.1) and the fraction of parotid with Jac Y with Y 1 (expansion, step 0.1) were recovered. The main focus of this paper was to investigate possible correlation between parotid shrinkage and pre-treatment parameters. For this aim the Jac_mean was considered as a synthetic and a robust parameter quantifying parotid shrinkage. A number of clinical data were collected; age, chemotherapy (y/n), neoadjuvant (y/n) and concomitant (y/n) chemotherapy, surgery (y/n) and primary tumor site (oropharynx, nasopharynx, hypopharynx and larynx) were considered as clinical variables. A number of anatomic/geometrical and dosimetric parameters were collected, including: initial parotid volume (IVP) (average: cc; range: cc), tumor volume (T) (average: cc; range: cc), overlap between the parotid gland and the lymphnode chain volume (OVPPTV2) (average: 4.10 cc; range: cc) and between the parotid and the high dose PTV (OVPPTV1) (average: 1.11 cc; range: cc), prescribed dose and daily dose, parotid planning mean dose (Dmean) (average: Gy; range: Gy) and V10-V40 dose volume histogram (DVH) values at planning. In addition to these pre-treatment parameters, the absolute and the percentage body thickness variation (ΔT) between the start and the end of the treatment were also considered. Thickness variation was measured on the available CT images as the difference between the half-thickness between the end and the start of the therapy at the level of C2, taking the line tangent to the vertebral body on CT, selecting the side opposite to the disease and/or to previous neck surgery. Data Analysis Several statistic tests were performed to assess correlation between parotid shrinkage and all considered parameters. First, the correlation between Jac_mean and all the recovered parameters was assessed through Spearman s test. An univariate and a stepwise logistic multivariate analysis (MVA, selecting variables with p-value 0.1) were performed by considering as the end point the parotid Jac_mean value smaller than the first quartile value of the population (Q1).

4 686 Broggi et al. To better investigate the correlation between parotids deformation and dosimetric parameters, the average DVHs of parotid gland with Jac_mean Q1 (large shrinkage) and Jac_mean Q1 (small shrinkage) were compared though a two tails t-test in order to define the dosimetric parameters (DVH points) more significantly different (lowest p-value). Parotid glands DVHs were stratified in three different groups based on DVH shape and according to their degree of deformation, trying to assess the most predictive dose-volume combination in the low and medium dose region. Logistic MVA analyses including the shape of DVHs were also performed. Analyses were carried out with the MedCalc software (v , MedCalc software bvba). Results Correlation between Jac_mean and Clinical/Dosimetric Parameters On average 82.6% (median value: 86.6%; range: 19.02%- 100%) of the voxels of parotid glands are affected by a shrinkage effect (Jac 1) and on average 13.7% (median: 8.5%; range: 0%-84.1%) of voxels presents a compression 50% (Jac 0.5). The average value of Jac_mean was 0.77 (median value: 0.76; range: ). The first quartile value (Q1) of Jac_mean was Only 10/168 parotid glands showed a Jac_mean 1. Jac_mean was compared with other standard contour indices; when all data were considered the correlation was statistically significant for both DSC and ASD (absolute Spearman s coefficient ; p ), but in the Q1 group of data the correlation decreased, and for the ASD index the co efficient was equal to with a bordeline p (0.052) value. body thickness variation were found significantly correlated (p ) with parotid deformation. Interestingly (Figure 1) the correlation between dosimetric/ geometric parameters and ΔVcc or ΔV% was generally poorer compared to the Jac_mean. These results suggest that JVH based parameters add significant information compared to the mere volume reduction. Table I Results of Spearman s correlation test between parotid Jac_mean and the pre-treatment variables. Variable p-value Institute Chemotherapy (y/n) Chemo-neoadjuvant (y/n) Chemo-concomitant (y/n) Surgery (y/n) Age Tumor site Tumor volume Initial parotid volume (IVP) Overlap parotid-lymphonodal tumour Overlap parotid-tumour Prescribed lymphonodal tumour dose Prescribed tumour dose 0.03 Daily lymphonodal tumour dose Daily tumour dose Parotid mean dose (Dmean) V V V V V Results similar to our previous study (12) were found for parotids volume variation: an average absolute (ΔVcc) and a percentage (ΔV%) volume variation equal to 7 cc and equal to 28% were respectively measured. Results of correlation tests for Jac_mean and all clinical, geometric, anatomic and dosimetric variables are reported in Table I. Based on correlation tests, OVPPTV1 (p ), OVPPTV2 (p ), age (p ) were found as the pre-treatment anatomical/clinical/geometrical variables mostly correlated with Jac_mean; Dmean (p ) and all DVH parameters, V10 (p ), V15 (p ), V20 (p ), V30 (p ) and V40 (p ) were also found significantly correlated with Jac_mean. In addition to pre-treatment parameters, absolute and percentage Figure 1: Parotid shrinkage and treatment related parameters. p-values ( 0.2) of the correlation (Spearman s test) between clinical/geometrical/ dosimetric parameters and parotid shrinkage in terms of Jac_mean, absolute (ΔVcc) and percentage (ΔV%) parotid volume variation.

5 Parotid Deformation Prediction 687 Modeling the Risk of Large Deformation: Pre-treatment Variables In Table II the results of the logistic univariate analysis were reported when considering Jac_mean 0.67 (i.e., Q1) as the end point. Focusing on pre-treatment parameters, OVPPTV1 (p ; OR , CI: ) and OVPPTV2 (p ; OR , CI: ) were found as the most predictive geometrical variables; V10 (p ; OR , CI: ), V15 (p ; OR , CI: ) and V40 (p ; OR , CI: ) the dosimetric ones. At MVA analysis, age (p ; OR , CI: ) and OVPTV1 (p ; OR , CI: ) were found as the best independent pre-treatment predictors (p ; AUC ( ). If we exclude geometrical parameters (being correlated to dosimetric variables), V10 (p ; OR , CI: ) was found as the best predicting variable (p ; AUC ( ). Shape of Parotid DVH and Risk of Large Deformation In order to better understand the correlation between parotid deformation (in terms of Jac_mean) and dosimetric parameters, the impact of the DVH shape was better investigated in the following way: the average DVHs of parotid gland with Jac_mean Q1 (large shrinkage) and Jac_mean Q1 (small shrinkage) were compared through a two-tails t-test (Figure 2). Table II Results of univariate logistic analysis. End point: Jac_mean < 0.67 (first quartile value of the population). Variable Chemotherapy (y/n) Chemo neoadjuvant (y/n) Chemo concomitant (y/n) p-value 0.647; OR ( ) 0.192; OR ( ) 0.892; OR ( ) Surgery (y/n) 0.097; OR ( ) Age 0.070; OR ( ) Tumor site ; OR = 1.75 ( ) Tumor volume 0.2; OR = ( ) Initial parotid volume (IVP) 0.56; OR = ( ) Overlap parotid-lymphonodal tumour 0.005; OR ( ) Overlap parotid-tumour 0.002; OR ( ) Prescribed lymphonodal tumour dose 0.824; OR ( ) Prescribed tumour dose 0.557; OR ( ) Daily lymphonodal tumour dose 0.416; OR ( ) Daily tumour dose 0.647; OR ( ) Parotid mean dose (Dmean) 0.042; OR ( ) V ; OR ( ) V ; OR ( ) V ; OR ( ) V ; OR ( ) V ; OR ( ) Figure 2: DVH and parotid deformation. (A) DVHs of all considered parotid glands. In black parotid glands with small deformations (Jac_mean Q1) and in grey with large deformations (Jac_mean Q1). (B) p-value (t-tests) between the average DVH of parotid gland with small and large deformations for all DVH dose points. As result, V10 and V40 were assessed as the most predictive dosimetric parameters, corresponding to the lowest p-values. A further ROC analysis found V % (sensitivity: 41.4%, specificity: 82.5%) and V % (sensitivity: 68%, specificity: 52.5%) to be the best cut-off values discriminating large and small shrinking parotids. Parotid glands were then separated according to their DVH shape as: bad- DVH (V10 93% and V40 36%), intermediate-dvh (V10 93% and V40 36%), good-dvh (V10 93%). The risk to have Jac_mean Q1 was 39.6% vs 19.6% vs 11.3% in the three risk groups respectively (p ) (Figure 3). When adding the DVH grouping variable in the MVA, patient s age (p ; OR , CI: ), OVPPTV1 (p ; OR , CI: ), OVPPTV2 (p ; OR , CI: ), V30 (p ; OR , CI: ) and bad-dvh (p ; OR: 0.19; CI: ) were found as the clinical, geometrical and dosimetric variables more significantly correlated with large parotid deformation (p ; AUC , CI: ). If we exclude geometrical parameters correlated with dosimetric variables, patient s age (p ; OR ; CI: ) and bad-dvh (p ; OR ;

6 688 Broggi et al. CI: ) were found as the best independent predicted variables (p ; AUC , CI: ) of the risk of large shrinkage (Jac_mean Q1). Including Patient Modification During Treatment in the Model When considering not only pre-treatment parameters, the percentage (p ; OR , CI: ) and the absolute (p ; OR , CI: ) body thickness variation were found significantly correlated with Jac_mean. Figure 3: Average parotid DVH. Parotid gland DVH was stratified based on V10 and V40 values: bad-dvh (V10 93% and V40 36%), intermediate-dvh (V10 93% and V40 36%), good-dvh (V10 93%). When including these variables in the backward MVA, the final models (including or excluding geometrical parameters, i.e., OVPTV1 and OVPTV2) the predictive power of the model increases, with AUC passing from 0.70 and 0.78 to 0.79 and 0.87, respectively. In Table III, a summary of the Table III Results of four logistic regression models. Body Thickness (no) OVPTV1/OVPT2 (yes) Body Thickness (no) OVPTV1/OVPT2 (no) Body Thickness (yes) OVPTV1/OVPT2 (yes) Body Thickness (yes) OVPTV1/OVPT2 (no) Predictive variable Age (p ; OR ; 95%CI: ) OVPPTV1 (p ; OR ; 95%CI: ) OVPPTV2(p ; OR ; 95%CI: ) V30 (p ; OR ; 95%CI: ) Bad-DVH (p ; OR ; 95%CI: ) Age (p ; OR ; 95%CI: ) Bad-DVH (p ; OR ; 95%CI: ) Body thickness % (p ; OR ; 95%CI: ) OVPPTV1 (p ; OR ; 95%CI: ) OVPPTV2 (p ; OR ; 95%CI: ) V20 (p ; OR ; 95%CI: ) V30 (p ; OR ; 95%CI: ) Bad-DVH (p ; OR ; 95%CI: ) Body thickness % (p ; OR ; 95%CI: ) Bad-DVH (p ; OR ; 95%CI: ) Significance level/ ROC curve analysis p AUC: %CI: p AUC: %CI: p AUC: %CI: p AUC: %CI: four logistic models (including or not body thickness variation and/or OVPTV1/OVPTV2) is shown: in any case, bad- DVH grouping always results to be highly correlated to the risk of large deformations. Discussion and Conclusion Head and neck cancer patients are known to be subject to significant anatomy modifications resulting from both tumor response and changes in normal tissues as a consequence of acute reactions, including weight loss. Several recent papers focused on anatomic variations of the parotid glands during fractionated radiotherapy, generally reporting a medial shift of the glands with a concomitant volume reduction and a consequent increase of the delivered dose compared to the planned one (1-6, 18). In the study of Barker et al. (1), the volume of parotid glands decreased trough a treatment course of fractions at a median daily rate of 0.6% with a medial shift of 3.1 mm at treatment completion. Hansen et al. (2) performed planning CTs scans before treatment and after an average dose of 36 Gy. A mean reduction in the parotid volume of 21.5% and 15.6% was observed for the left and right gland, respectively. In a more recent study, Han et al. (18) estimated a daily reduction of 1.2% in both parotid glands. Robar et al. (3) calculated 3D displacements for several critical structures

7 Parotid Deformation Prediction 689 including parotid glands by considering diagnostic weekly CT during the RT course. They found that the superficial regions of both parotids showed a medial translation of mm per week and mm per week respectively for the left and the right gland with a global shrinking of 4.9% per week. Few studies investigated how parotid shrinkage may be influenced by the position of the primary tumor and by the dose distribution within the parotids. Vásquez Osorio et al. (5) investigated 3D anatomic changes during IMRT, separately for irradiated and spared parotids and submandibular glands by analyzing CT scans at the begin of the treatment and after 46 Gy. They found a significantly (p 0.001) larger reduction for the irradiated parotid gland compared to the spared one (17 7% vs 5 4%); a similar trend (20 10% vs 11 7%) was also found for irradiated and spared submandibular glands. As suggested by several authors (4, 5, 7, 18-20), all these observed anatomic changes may cause significant dosimetric effects when highly tailored dose distributions are delivered; in particular, parotid volume reduction and medial migration can lead to the delivery of an higher parotid dose compared to the planned one. Hunter KU et al. (21) analysed the parotid gland dose-effect relationship based on planned and actually delivered doses in a group of 18 PTs. The median difference from planned to delivered mean gland doses was only 2.3 Gy; both planned and delivered doses were significantly correlated with post-treatment salivary outputs, without statistically significant differences in the correlations. Our investigation had the primary aim of assessing pretreatment predictors of parotid shrinkage with the potential to better guide planning optimization and to select patients that could significant take advantage from adaptive strategies. Compared to other investigations where anatomical variations were estimated in terms of volume variation or in terms of position differences, in this paper parotid deformations were quantified by using the Jacobian of the deformation field of an elastic registration. This is a simple and intuitive method able to condense the information dealing with single voxel deformation in one number. This approach is not novel; the Jacobian has been recently suggested to quantitatively assess local changes of pulmonary functions during radiotherapy (22, 23), by considering expansion/compression of lung voxels during normal breathing. In addition the introduction (13) and the application in current study of the Jacobian Volume histogram (JVH) permits to condense the information pattern into a single curve, providing the possibility to quantify the deformation as an organ-effect. JVH-based parameters were found to add information compared to the mere volume reduction, enhancing the correlation between the type of the deformation and pre-treatment variables. In particular the correlations with Jac_mean were enhanced for parotid mean dose and V10. Despite less predictable, absolute parotid volume variation can be also used in the clinical practice; in fact, a significant correlation (p 0.01) with DVH values in the range V15-V40 was found and a similar correlation (p ) with patient s age was estimated. Similarly to a previous study (12), we found patient s age to be the most predictive clinical variable of parotid shrinkage. It was confirmed that the parotid shrinkage seems more significant in younger patients, suggesting that massive apoptosis, leakage of granules and subsequent lysis of acinar cells, responsible for the acute radiation-induced function loss of salivary glands, could be more active for younger patients. The low sensitivity of old patients compared to the younger ones could also be correlated to possible interactions with drugs (i.e., anti-depressive and anti-hypertensive) mostly used by older patients and/or a likely different level of parotid hydration between old and young patients. Obviously, this result should be reconfirmed by considering a larger and different head-neck cancer population; no other studies have investigated a correlation between parotid shrinkage and patient age. Without any doubt, our Jacobian-based method showed a clear correlation between parotid deformation and dosimetric variables; in particular, as confirmed in the multivariate analyses there is a strong correlation between parotid shrinkage and the planned DVH shape, with the low dose bath effect being predominant. In according to other papers (24-27) in which the low dose bath has been reported as a potentially detrimental factor when considering salivary function impairment, also in our study V10 and V15 were found as the dosimetric pre-treatment parameters more significantly correlated to parotid shrinkage. Although parotid shrinkage mainly involved the external portion of the glands, our results do not exclude that the fraction of parotid receiving higher doses (i.e., V40 Gy) could play some role; neither a possible interaction between the internal parotid gland region irradiated at medium-high dose and the external region reactions could be excluded. In any case, our results confirm the importance of an accurate planning optimization approach in order to minimize as much as possible the parotids DVH in the whole dose range, including the fraction of parotid receiving doses as low as 10 Gy (5, 19). The significant correlation found with the percentage body thickness variation, when not only including pre-treatment variables in the MVA analysis, seems to indirectly suggest that parotid deformation may be only partially explained by the planned dose distribution. A subjective patient radiation reaction should also be considered. On the other hand, the thickness variation should be associated to an increase of the dose effectively delivered during the treatment, due to the consequent shift of parotids toward the midline; thus, we cannot exclude the possibility that this correlation is partially due to this effect.

8 690 Broggi et al. In conclusion, the clear correlation between parotid deformation and clinical/geometrical/dosimetric parameters found in this study together with the possibility to use quantitative image-based scoring information from images taken during and after radiotherapy for measuring the radiation-induced damage (27, 28) seems to be very promising in predicting individual reactions and possibly in adapting the treatment, primarily to reduce early and/or late toxicity. Conflict of Interest None. Acknowledgments The study was founded by an AIRC grant (IG10521). References 1. Barker, J. L. Jr., Garden, A. S., Ang, K. K., O Daniel, J. C., Wang, H., Court, L. E., Morrison, W. H., Rosenthal, D. I., Chao, K. S., Tucker, S. L., Mohan, R., Dong, L. Quantification of volumetric and geometric changes occurred during fractionated radiotherapy for headand-neck cancer using an integrated CT/linear accelerator system. Int J Radiat Oncol Biol Phys 59, (2004). DOI: / j.ijrobp Hansen, E. K., Bucci, M. K., Quivey, J. M., Weinberg, V., Xia, P. Repeat CT imaging and replanning during the course of IMRT for head-and-neck cancer. Int J Radiat Oncol Biol Phys 64, (2006). DOI: /j.ijrobp Robar, J. L., Day, A., Clancey, J., Kelly, R., Yewondwossen, M., Hollenhorst, H., Rajaraman, M., Wilke, D. Spatial and dosimetric variability of organs at risk in head-and-neck intensity-modulated radiotherapy. Int J Radiat Oncol Biol Phys 68, (2007). DOI: /j.ijrobp Lee, C., Langen, K. M., Lu, W., Haimerl, J., Schnarr, E., Ruchala, K. J., Olivera, G. H., Meeks, S. L., Kupelian, P. A., Shellenberger, T. D., Mañon, R. R. Evaluation of geometric changes of parotid glands during head and neck cancer radiotherapy using daily MVCT and automatic deformable registration. Radiother Oncol 89, (2008). DOI: /j.radonc Vásquez Osorio, E. M., Hoogeman, M. S., Al-Mamgani, A., Teguh, D. N., Levendag, P. C., Heijmen, B. J. Local anatomic changes in parotid and submandibular glands during radiotherapy for oropharynx cancer and correlation with dose, studied in detail with non rigid registration. Int J Radiat Oncol Biol Phys 70, (2008). DOI: /j.ijrobp Castadot, P., Geets, X., Lee, J. A., Christian, N., Grégoire, V. Assessment by a deformable registration method of the volumetric and positional changes of target volumes and organ at risk in oropharyngo-laryngeal tumors treated with concomitant chemoradiation. Radiother Oncol 95, (2010). DOI: / j.radonc Wu, Q., Chi, Y., Chen, P. Y., Krauss, D. J., Yan, D., Martinez, A. Adaptive replanning strategies accounting for shrinkage in head and neck IMRT. Int J Radiat Oncol Biol Phys 75, (2009). DOI: /j.ijrobp Mohan, R., Zhang, X., Wang, H., Kang, Y., Wang, X., Liu, H., Ang, K. K., Kuban, D., Dong, L. Use of deformed intensity distributions for on-line modification of image-guided IMRT to account for interfractional anatomic changes. Int J Radiat Oncol Biol Phys 61, (2005). DOI: /j.ijrobp Kuo, Y. C., Wu, T. H., Chung, T. S., Huang, K. W., Chao, K. S., Su, W. C., Chiou, J. F. Effect of regression of enlarged lymph nodes on radiation does received by parotid glands during intensity modulated radiotherapy for head and neck cancer. American Journal of Clinical Oncology 29, (2006). DOI: /01. coc b3 10. Schwartz, D. L., Garden, A. S., Shah, S. J., Chronowski, G., Sejpal, S., Rosenthal, D. I., Chen, Y., Zhang, Y., Zhang, L., Wong, P. F., Garcia, J. A., Kian Ang, K., Dong, L. Adaptive radiotherapy for head-andneck cancer. Dosimetric results from prospective clinical trial. Radiother Oncol 106, (2013). DOI: /j.radonc Li, N., Zarepisheh, M., Uribe-Sanchez, A., Moore, K., Tian, Z., Zhen, X., Graves, Y. J., Gautier, Q., Mell, L., Zhou, L., Jia, X., Jiang, S. Automatic treatment plan re-optimization for adaptive radiotherapy guided with the initial plan DVHs. Phys Med Biol 58, (2013). DOI: / /58/24/ Broggi, S., Fiorino, C., Dell Oca, I., Dinapoli, N., Paiusco, M., Muraglia, A., Maggiulli, E., Ricchetti, F., Valentini, V., Sanguineti, G., Cattaneo, G. M., Di Muzio, N., Calandrino, R. A two variable linear model of parotid shrinkage during IMRT for head and neck cancer. Radiother Oncol 94, (2010). DOI: /j.radonc Fiorino, C., Maggiulli, E., Broggi, S., Liberini, S., Cattaneo, G. M., Dell oca I, Faggiano, E., Di Muzio, N., Calandrino, R., Rizzo, G. Introducing the jacobian-volume-histogram of deforming organs: application to parotid shrinkage evaluation. Phys Med Biol 56, (2011). DOI: / /56/11/ Faggiano, E., Fiorino, C., Scalco, E., Broggi, S., Cattaneo, G. N., Maggiulli, E., Dell Oca, I., Di Muzio, N., Calandrino, R., Rizzo, G. An automatic contour propagation method to follow parotid glands deformation during head and neck cancer Tomotherapy. Phys Med Biol 56, (2011). DOI: / /56/3/ Dice, Lee, R. Measures of the amount of ecologic association between species. Ecology 26, (1945). DOI: / Heimann, T., van Ginneken, B., Styner, M. A., Arzhaeva, Y., Aurich, V., Bauer, C., Beck, A., Becker, C., Beichel, R., Bekes, G., Bello, F., Binnig, G., Bischof, H., Bornik, A., Cashman, P. M., Chi, Y., Cordova, A., Dawant, B. M., Fidrich, M., Furst, J. D., Furukawa, D., Grenacher, L., Hornegger, J., Kainmüller, D., Kitney, R. I., Kobatake, H., Lamecker, H., Lange, T., Lee, J., Lennon, B., Li, R., Li, S., Meinzer, H. P., Nemeth, G., Raicu, D. S., Rau, A. M., van Rikxoort, E. M., Rousson, M., Rusko, L., Saddi, K. A., Schmidt, G., Seghers, D., Shimizu, A., Slagmolen, P., Sorantin, E., Soza, G., Susomboon, R., Waite, J. M., Wimmer, A., Wolf, I. Comparison and evaluation of methods for liver segmentation from CT datasets. IEEE Trans Med Imaging. 28, (2009). DOI: /TMI Maurer, C. R., Ransheng, Qi., Raghavan, V. A linear time algorithm for computing exact Euclidean distance transforms of binary images in arbitrary dimensions. IEEE Transactions on Pattern Analysis and Machine Intelligence 25, (2003). DOI: / TMAMI Han, C., Chen, Y. J., Liu, A., Schultheiss, T. E., Wong, J. Y. Actual dose variation of parotid glands and spinal cord for nasopharyngeal cancer patients during radiotherapy. Int J Radiat Oncol Biol Phys 70, (2008). DOI: /j.ijrobp Wang, Z. H., Yan, C., Zhang, Z. Y., Zhang, C. P., Hu, H. S., Kirwan, J., Mendenhall, W. M. Radiation-induced volume changes in parotid and submandibular glands in patients with head and neck cancer receiving postoperative radiotherapy: a longitudinal study. Laryngoscope 119, (2009). DOI: /lary Lee, C., Langen, K. M., Lu, W., Haimerl, J., Schnarr, E., Ruchala, K. J., Olivera, G. H., Meeks, S. L., Kupelian, P. A., Shellenberger, T. D., Mañon, R. R. Assessment of parotid gland dose changes during head and neck cancer radiotherapy using daily megavoltage computed tomography and deformable image registration. Int

9 Parotid Deformation Prediction 691 J Radiat Oncol Biol Phys 71, (2008). DOI: / j.ijrobp Hunter, K. U., Fernandes, L. L., Vineberg, K. A., McShan, D., Antonuk, A. E., Cornwall, C., Feng, M., Schipper, M. J., Balter, J. M., Eisbruch, A. Parotid glands dose-effect relationships based on their actually delivered doses: implication for adaptive replannig in radiation therapy of head-and-neck cancer. Int J Radiat Oncol Biol Phys 87, (2013) DOI: /j.ijrobp Christensen, G. E., Song, J. H., Lu, W., El Naqa, I., Low, D. A. Tracking lung tissue motion and expansion/compression with inverse consistent image registration and spirometry. Med Phys 34, (2007). DOI: / Ding, K., Bayouth, J. E., Buatti, J. M., Christensen, G. E., Reinhardt, J. M. 4DCT-based measurements of changes in pulmonary function following a course of radiation therapy. Med Phys 37, (2010). DOI: / Dijkema, T., Terhaard, C. H., Roesink, J. M., Braam, P. M., van Gils, C. H., Moerland, M. A., Raaijmakers, C. P. Large cohort dose volume response analysis of parotid gland function after radiotherapy: intensity-modulated versus conventional radiotherapy. Int J Radiat Oncol Biol Phys 72, (2008). DOI: / j.ijrobp Bussels, B., Maes, A., Flamen, P., Lambin, P., Erven, K., Hermans, R., Nuyts, S., Weltens, C., Cecere, S., Lesaffre, E., Van den Bogaert, W. Dose response relationships within the parotid gland after radiotherapy for head and neck cancer. Radiother Oncol 73, (2004). DOI: /j.radonc van Luijk, P., Faber, H., Schippers, J. M., Brandenburg, S., Langendijk, J. A., Meertens, H., Coppes, R. P. Bath and shower effects in the rat parotid gland explain increased relative risk of parotid gland dysfunction after intensity-modulated radiotherapy. Int J Radiat Oncol Biol Phys 74, (2009). DOI: /j.ijrobp Jeraj, R., Cao, Y., Ten Haken, R. K., Hahn, C., Marks, L. Imaging for assessment of radiation-induced normal tissue effects. Int J Radiat Oncol Biol Phys 76, S140-S144 (2010). DOI: / j.ijrobp Bayouth, J. E., Casavant, T. L., Graham, M. M., Sonka, M., Muruganandham, M., Buatti, J. M. Image-based biomarkers in clinical practice. Semin Radiation Oncol 21, (2010). DOI: /j.semradonc Received: November 11, 2013; Revised: February 24, 2014; Accepted: March 10, 2014