Cy7 DiC18 – Eptifibatide

The inﬂuence of the image registration method on the adaptive radiotherapy. A proof of the principle in a selected case of prostate IMRT

A B S T R A C T
Purpose: To analyse the inﬂuence of the image registration method on the adaptive radiotherapy of an IMRT prostate treatment, and to compare the dose accumulation according to 3 diﬀerent image registration methods with the planned dose.Material and methods: The IMRT prostate patient was CT imaged 3 times throughout his treatment. The prostate, PTV, rectum and bladder were segmented on each CT. A Rigid, a deformable (DIR) B-spline and a DIR with landmarks registration algorithms were employed. The diﬀerence between the accumulated doses and planned doses were evaluated by the gamma index. The Dice coeﬃcient and Hausdorﬀ distance was used to evaluate the overlap between volumes, to quantify the quality of the registration. Results: When comparing adaptive vs no adaptive RT, the gamma index calculation showed large diﬀerences depending on the image registration method (as much as 87.6% in the case of DIR B-spline). The quality of the registration was evaluated using an index such as the Dice coeﬃcient. This showed that the best result was obtained with DIR with landmarks compared with the rest and it was always above 0.77, reported as a re- commended minimum value for prostate studies in a multi-centre review.
Conclusions: Apart from showing the importance of the application of an adaptive RT protocol in a particular treatment, this work shows that the election of the registration method is decisive in the result of the adaptive radiotherapy and dose accumulation.

1.Introduction
Adaptive radiotherapy (ART) has been described as a strategy to include patient variations during a course of radiotherapy that are due to anatomical changes or due to tumoral or biological changes. As a result, the actual delivered dose diﬀers from the planned dose [1]. Image registration, which relates the points in one image to the points in another image are used to compensate for daily variations. Several algorithms have been described and used in clinical practice but their performance diﬀers [2]. Image registration, which relates the points in one image to the points in another image are used to compensate for daily variations. Several algorithms have been described and used in clinical practice but their performance diﬀers.The objective of this study was to evaluate the use of an oﬀ-line,adaptive radiotherapy process in a selected IMRT prostate treatment. The aim was to accumulate the dose using 3 diﬀerent registration al- gorithms and compare their respective results with the initial dose plan.This work therefore only tries to show the huge diﬀerences that can be reached in a particular case and, using validation systems of the re- gistration, bring to light the importance of the image registration method employed. In any case, our major interest focused on the de- formable image registration since the rigid registration is inadequate for the extracranial registrations.

2.Material and methods
Fig. 1 shows the overall workﬂow of the study. One prostate cancer intensity modulated radiotherapy (IMRT) treatment, which had 3 CT studies, was selected. Every CT represented a third of the total radio- therapy course. The patient was given enemas prior to simulation and daily during the course of radiotherapy and he was advised to have a comfortably full bladder during treatment. The prostate, rectum and bladder were manually delineated by the same radiation oncologist on each CT. A planning target volume (PTV) was created from a prostateexpansion. The ﬁrst CT was used for planning 76 Gy, administered in 38 fractions, using XIO v.4.62.00.13 (Elekta AB, Stockholm, Sweden). The resulting IMRT plan was saved as a template and translated to other CT studies. The isocentre was placed using the spatial coordinates from the planning CT, the monitor units were kept exactly the same and the dose was recalculated for every CT study. Therefore every CT study had its own CT images, dose and structures, although the plan was the same between studies. Then, all this DICOM-RT information was exported to the 3D-Slicer package v.4.5.0-1 (www.slicer.org) using the SlicerRT module.After carrying out the rigid registration, a large discrepancy was observed in the rectum and bladder position between the studies, which suggested that the ﬁrst step should be the comparison of the DVHs of the diﬀerent dose plans of each repeated CT independently in order to quantify the diﬀerences. This visual discrepancy was also conﬁrmed numerically using the Hausdorﬀ distance (HD) and Dice similarity coeﬃcient (DSC) parameters during the rigid registration.The next step was the registration of CT1 vs the other two CTs. All registrations used the ﬁrst CT (CT1) as the reference volume, and subsequent CTs (CT2 and CT3) were taken as the moving volumes. The registration procedure has been described elsewhere [3,4]. In brief, after a rigid image registration (RIR) using the mutual information (MI) as the cost function, the moving volume was used for the initialization of the deformable image registration (DIR).

Two parametric DIR algo- rithms were used: a B-spline algorithm with the mean squared error (MSE) as the cost function, and a landmark-based deformable algorithm (Landwarp 3D-Slicer module) which used the gaussian function as the Radial Base Function (RBF).The deformation ﬁelds, from CT2 to CT3 toward CT1 that were created, were used to transform and warp the dose map. Warped doseswere summed to the CT1 dose map and PTV, rectum and bladder dose- volume histograms (DVHs) were produced and compared between them and with the original one (CT1 alone). Finally the ﬁnal dosimetry obtained after each registration and dose accumulation was compared with the initial dosimetry using the gamma index. The parameters for calculating the gamma index were: distance-to-agreement criteria was set to 3 mm; dose diﬀerence criteria was 3% deﬁned as percentage of the reference dose value, in this case the global maximum value; and the dose threshold for gamma analysis was 10% of the maximum dose. The choice of parameters was not important because our purpose was the comparison of the results between the registration methods, al- though these parameters represent the clinical standards. The dis- crepancy in volume segmentations after deformation was also quanti- ﬁed by the 95% Hausdorﬀ distance (95%HD), average HD, and the DSC.

3.Results
Fig. 2 and Tables 1 and 2 show the DVHs of the PTV, rectum and bladder for each of the three CTs scanned independently, without any registration. DVH discrepancies are evident, in particular for the rectum and bladder.Fig. 3 and Tables 3 and 4 show the DVHs and DVH parameters of PTV, rectum and bladder, which compare the initial dose plan on CT1 with the dose accumulation plan after the rigid, deformable and de- formable with landmarks registration of the 3 CTs. These comparisons were carried out taking into account only the structures contoured on CT1 while assessing, at the same time, the eﬀect of dose accumulation according to the diﬀerent image registration methods on these struc- tures. Large diﬀerences were expected in these DVHs since importantvariations were observed in the anatomy of the rectum and bladder around the PTV, which are hardly aﬀected in the RIR case. In addition, the dose accumulation results were compared with the initial dose plan using the gamma index (Table 6). The largest diﬀerences appear in the deformable registration, especially in the most deformed areas such as bladder and rectum (Fig. 5). The high gamma index in the rigid regis- tration case is due to the small calculation diﬀerences between CT1 and the other two CTs (CT2 and CT3). Only large diﬀerences on transla- tions, rotations, or dose calculations between the diﬀerent CTs could produce relevant gamma index variations.The quality of the 3 registration methods was evaluated by the DSC, the 95%HD and the average HD. On Table 5, the average and 95% HD as well as the DSC of the registration between CT1 vs CT2 and CT3 for the prostate, rectum and bladder are presented. These results show thatthat each session represented one third of the overall treatment. That obviously does not reﬂect reality since the prostate shows important interfraction diﬀerences. Analysis of the causes of rectum and bladder ﬁlling diﬀerences were out of the scope of the study, in fact the patientin some cases the registration obtains good results in relation to the rigid registration but in other cases the result of the registration is not as good, especially for the B-spline DIR. The initial diﬀerences between CT volumes are huge; for instance the volume diﬀerence for the rectum between CT1 and CT2 is more than double (112 cc vs 245 cc) (Table 7). Graphically, Fig. 4a shows this fact after having carried out the rigid registration based on the bone structure between CT1 and CT2. The deformable registration was very important in the bladder area, how- ever, due to the lack of radiological information, the prostate area was carried away unnaturally (Fig. 4b).

4.Discussion
Our results demonstrate that registrations carried out by diﬀerent algorithms can produce diﬀerent results according to the functions used. We have obtained dose and volume diﬀerences after registration according to the options of the algorithm. Clinical experience has shown us that radiotherapy patients can show signiﬁcant anatomical changes due to physiological factors or secondary eﬀects of the treat- ment. As a consequence, the initial treatment plan cannot reﬂect the actual dose delivered to the patient. Our study was to demonstrate how signiﬁcant dose accumulation diﬀerences can be obtained with respect to an initial dose plan for a particular IMRT prostate treatment. To that aim, we accumulated 3 diﬀerent dose fractions, spread out, assuming was accompanied with an instillation of air, which might be the real cause of the increase.The large diﬀerence in the organs at risk was responsible for the DSC and 95%HD diﬀerences. Nevertheless, volume segmentation variability also inﬂuences these results [8,9]. Therefore, the overall volume diﬀerences were not only due to the registration algorithm but also due to the segmentation variability. These two kinds of un- certainties should be split up to evaluate more accurately the quality of the registration algorithms. In fact, the uncertainty due to the seg- mentation increases in regions without clear borders, like the caudal prostate region on the CT images. However, this eﬀect was smaller in the bladder because of the use of contrast agents during each CT ac- quisition, so the good visual contrast from the surrounding soft tissues reduced the uncertainty of segmentation.RIR is not a useful option for dose accumulation because the humanbody is not a rigid body. RIR in the presence of shape and posture changes oblige the use of multiple regions of interest registration [10], an approach that was not used in the current study.

Several algorithms have been created to carry out DIR [2] and we used the B-spline DIR algorithm and the DIR with landmarks (Landwarp) integrated into the SlicerRT module. In spite of the fact that the 3D Slicer has other ways of carrying out the registration, we considered these the most appropriate algorithms, which were adapted from the Plastimatch package. Ac- cording to the DSC, 95% and average HD results, it seems that the landmark-based algorithm is the most reasonable, although it was more labour-intensive due to the positioning of a large number of the cor- responding landmarks. In this case, the use of landmarks meant placing35 points at every CT study involved in both DIR registration (CT1 vs CT2 and CT1 vs CT3), up to a total of 140 points. This was an arbitrary number of points but we considered it the minimum for attaining a visually acceptable registration. We are aware that the limited number of points does not ensure accuracy of registration overall, rather only at particular points, so accuracy beyond the cloud of points or between each point is not guaranteed. This whole process of ART is very time- consuming and in order to be able to use it in the clinical setting a lot of procedures would need to be optimized throughout, which is discussed in a recent AAPM report [11].It is important to mention that the same DIR algorithm (for instance the B-spline algorithm) implemented by diﬀerent groups can produce diﬀerent results [12]. On the other hand the B-spline algorithm requires maximizing or minimizing a cost function to measure the similarity of the two image data sets. The MSE function cost was used instead of the mutual information (MI), which, at present, is considered the most accurate and reliable, not only for single modality registration but for multimodality too. However 3D Slicer only provided the MSE em- ployment for DIR algorithms. The modest result obtained with B-spline could be partially attributed to this fact. At the same time, one of the drawbacks of B-spline is the unrealistic deformations created when it is used for high resolution registrations.

Accuracy of ROI overlap of dif- ferent DIR algorithms for the prostate ranges from 0.77 to 0.95, which includes the segmentation uncertainty of the prostate [12]. Our DSC for DIR with landmarks (Table 5) was aﬀected by this uncertainty too and was consistent with the reported values from a multi-centre review [19] for the prostate and for the organs at risks.It could be argued that B-spline DIR (without landmarks) canproduce, in some cases, worse DSC and 95%HD. The results presented here are, in fact, a ﬁnal result after several attempts that involved several parameter combinations of the algorithm. Interactive registra- tions were carried out to ensure more accurate and realistic deforma- tions. The AAPM report stresses that, with a section dedicated to the use of registration software and to the importance of optimizing the pro- cedure [11]. We want to note that the marked diﬀerences found on bladder and rectum (Fig. 4) produced unnatural prostate displacements. Nevertheless, it is diﬃcult not to move the prostate in the registration since the most relevant radiological information to drive deformation is borne in the bladder. The rectum could not be optimally registered, either due to the presence of gas pockets in one the CTs. Dose de- formation has not been exempt from criticism and concern exists aboutdose deposition related to bladder and rectum volumes [6,13–16]mainly due to the non-correspondence problem and to the diﬃculties in validating calculated doses against measurements. Deformable regis- tration relies on the assumption that every point of one image corre- sponds to some point in the other image, and the presence of bowel gas or diﬀerent faeces volumes violates this assumption [17]. Methods to deal with this rectal inconsistency and ambiguity between image stu- dies is very limited and require manual intervention: for example,creating “artiﬁcial gas” [18], detecting and painting of bowel gaspockets [19] or using an opposite approach, deﬂating the gas pockets [17], none of them implemented in the software package used.Interestingly, although our DVH metrics are below the QUANTEC recommendations for the prostate [20] in this particular case, the im- pact of the dose accumulation on decisive metrics with respect to the initial dose plan should make us consider the validity of this ﬁnaladministered treatment.

5.Conclusion
We have demonstrated that diﬀerent registration algorithms bring diﬀerent results, so caution is necessary when evaluating the ﬁnal re- sults of registration in the clinical setting. Dose Cy7 DiC18 accumulation proce- dures, which rely on DIR suﬀer the same criticisms.