| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Full paper |
Departments of 1Radiotherapy Physics 2YCR Clinical Oncology, Weston Park Hospital, Whitham Road, Sheffield 3Department of Radiology, Royal Hallamshire Hospital, Glossop Road, Sheffield, UK
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
The first step in the treatment process is localization of the tumour, usually using radiographic films and/or CT data. Tepper et al [12] showed that performing a planning CT scan in addition to existing diagnostic information enabled improvements in target localization in 49% of patients. In the authors' oncology centre, intravenous (IV) contrast is used routinely in diagnostic CT scanning of these patients, but not used routinely when acquiring CT scans for treatment planning purposes. The use of enhanced CT scans can offer improved tumour visibility in many cases and may enable improved localization for planning [13, 14]. Whilst seemingly obvious, improvements may be available in marking-up visible gross tumour volumes (GTVs), although the overall impact on the target or planning target volume (PTV) and general plan quality is unknown. The use of contrast-enhanced data has been shown (for other sites, e.g. Zhou et al and Valcenti et al [10, 15]) to provide improved tumour delineation. For brain tumours and head and neck cancer, there have been no previous reports that assess the impact of contrast-enhanced CT on the delineation of the GTV and the effect of such change on the PTV.
When contouring, the oncologist will mark the gross tumour volume (GTV), which is the visible extent of the tumour, with the help of diagnostic MRI images. A margin is added to this to allow for non-visible tumour infiltration, creating a clinical target volume (CTV). An additional margin is added to the CTV to account for patient movement and set-up inaccuracies, generating a final planning target volume (PTV) to be treated [16]. For this study, PTVs are not marked on directly, but are always generated by adding a uniform 2D margin to the initial GTV. Once generated, the initial PTV may be edited by the clinician to achieve a final PTV contour. This editing may be necessary, for example, if the software-generated PTV extends beyond the patients' anatomy. Although a radiologist is the recognized expert in the interpretation of medical images, in many centres (the authors' included) definition of the GTV is performed solely by the oncologist. By requiring the GTV to be defined by a radiologist and the remainder of the marking-up process (definition of the CTV and PTV) to be done by the oncologist, improvements in planning accuracy and outcome may be possible.
This study addresses two main questions:
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
gives a list of sites and associated patient numbers. Patient selection criteria: all patients having radical external beam treatment with CT planning for head and neck or brain sites are considered. Each patient must be physically able to undergo CT scanning (weight and girth can result in some exclusions) and be able to consent. Ethics approval was obtained for non-standard administration of contrast as part of the treatment planning process. They each had pre- and post-contrast CT scans carried out in the treatment position. Our standard CT protocols were used for image acquisition: 5 mm slice thickness with 5 mm interslice gap, acquired in helical mode with pitch of 1, on our Picker PQS CT. After administering contrast, the couch was returned to the original position without any patient movement, enabling both scans (unenhanced and enhanced) to be acquired with identical scanner coordinates.
|
The enhanced studies were intrinsically registered with the unenhanced, such that contours marked on the enhanced image would be automatically transferred and stored with all previous others already marked on the unenhanced study. For data sets with identical scanner coordinates, image registration was performed automatically by the AcQSIMTM software (see Figure 1). The accuracy of this registration is dependent on negligible patient movement during scanning. Diagnostic scan data was not registered to planning CT data, but was generally available as hardcopy films and viewed on a light-box beside the AcQSIM work-station. All contours were stored on the unenhanced data set and analysis performed on this data set.
|
In comparing contour pairs, three indices were used:
Volume changes: contour pairs were analysed to identify any changes in the volume of the GTV or PTV contours. All unenhanced patient scans and contours were imported into the CADPLANTM (Dosetek and Varian Medical Systems) treatment planning system (TPS) and dose–volume histograms (DVHs) were generated to yield values for GTV and PTV volume.
Geographical changes: pairs of GTVs were compared to identify any geographical shift of the re-marked contours from the position of the original GTV. On the AcQSIM virtual simulator, shifts between the geometric centres of the GTV contours were measured in three orthogonal axes, defined as lateral (L), anterior/posterior (A/P) and superior/inferior (S/I) shift. Using the Isocentre Manager, the centre of gravity of each GTV was identified automatically. Shifts between centres of GTVs under comparison were then simply found by subtraction of the coordinates in the orthogonal axes. From these shifts, an overall 3D scalar value was computed for each contour pair. The displacement of the re-marked contours can be assessed in terms of comparison with conventional treatment set-up tolerance. That is to say, we can note when the change in geographical isocentre, due to the use of additional input (contrast-enhanced data or radiologist input) is of the same order as an alteration in patient geographical set-up isocentre, which would conventionally require action by staff to reposition the patient.
Area coverage changes: in the TPS, the original treatment plan was applied to all subsequent sets of contours. Pairs of PTVs (original and re-marked) were then compared in terms of tolerance volumes (TV), i.e. the percentage of the target which is either below 90% prescribed dose (target under-coverage) or above 105% prescribed dose (target over-coverage). If, for example, the enhanced target contour is assumed to be the "true" target, the amount of under- or over-coverage the "true" target will experience can be measured, the original treatment plan (based on the unenhanced target contour) having been applied.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
shows a summary of the data for changes in contour volumes. The unenhanced contours marked up by the oncologist and radiologist show good agreement, with percentage volume changes in the order of 1–6%. The mean percentage changes between pairs of GTVs and PTVs correlate well. The PTVs show smaller standard deviations compared with the GTVs.
|
Isocentre shifts between pairs of contours are shown in Table 3. For the oncologist, all shifts are within our daily setup tolerance of 5 mm for mould-immobilized patients. Our standard protocol requires no corrective action by staff for errors in geographical isocentre set-up of this level. Isocentre shift difference between contours marked by radiologist and oncologist are also higher (though not significantly) than precision data from either operator, with value 5.8 mm, which is outside the limit for daily patient setup tolerance.
|
. There is good agreement with plan coverage, with maximum percentage changes being of the order of 5–6% for oncologist mark-up precision.
|
shows percentage differences in contour volumes for the radiologist mark-up precision and for the radiologist compared with the oncologist. Mark-up precision for the radiologist is good, and in fact slightly higher than that of the oncologist, both in mean volume change and SD on the mean. There is a marked difference in mean percentage GTV volume (unenhanced) between radiologist and oncologist, with the radiologist marking larger volumes. This is not statistically significant for the group as a whole, but when the data are split into head and neck, and brain groups.
For data split into the two groups (Table 6), the precision of radiologist mark-up is lower in the brain cases, although this is not significant. Whilst differences in (unenhanced) GTV mark-up between oncologist and radiologist for the group as a whole are not significant, when split into two, mark-up by the differing clinicians is found to be significant at p = 0.01 (head and neck) and p<0.02 (brain). A similar trend significance is found for PTV mark-up (p<0.05). Head and neck cases, in fact show an improved precision for the radiologist compared with the oncologist, with a lower intraoperator SD. Considering data where the oncologist marks-up a PTV based on a GTV delineated by the radiologist, again, significance change in unenhanced PTV percentage volume change is only seen when the data is split into two groups, with head and neck cases showing a significant difference (p<0.02, Wilcoxon rank sum test).
|
.
|
. Neither of the percentage GTV volume changes (for oncologist or radiologist) was statistically significant. However, for the GTVs, a comparison of the impact of contrast for the radiologist and oncologist, is significant (p<0.01, Wilcoxon signed rank test). Obviously, contrast-enhanced data has a differing impact on tumour delineation depending on which clinician is using it. This result can also be compared with the difference between radiologist-oncologist mark-up of unenhanced scans (see Table 2). The use of contrast seems to have a greater difference in the marking-up of the radiologist than that of the oncologist.
Isocentre shifts, which result from use of enhanced data (Table 3), are similar to those of precision data. Whilst there may be changes in volume, the geographical position of the structure has not changed. In agreement with data on volume changes, differences in coverage for contrast-enhanced data also show little change to precision values (see Table 4).
When comparing the difference between enhanced and non-enhanced PTV mark-up, there is a statistical difference (p<0.01) between these for the contours generated from the radiologist's GTV compared with those generated solely by the oncologist, see Table 5. Percentage volume changes for the radiologist mark-up is significantly larger with contrast than without (p<0.01, Wilcoxon signed rank test), so contrast is obviously a much greater influence on mark-up for the radiologist than for the oncologist.
When investigating the data split into two groups, the mean percentage difference in GTV for head and neck cases is larger, at mean –20.2% compared with 8.5% for the brain group (although not statistically significant). There is little difference in PTV changes between the two groups, and similarly in isocentre shifts and tolerance coverage (data not shown).
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
Imaging
The first step in the process is the assimilation of information from the available radiological images and clinical data. Some authors have noted significant variation amongst clinicians at this first step [4, 19].
When first marking-up for treatment, the oncologist may have additional information, which he/she will not have when re-contouring months later. They may have seen and examined the patient recently or have discussed the case with the surgeon regarding tumour extent and geographic infiltration. Excluding the effect of these factors is obviously difficult.
Clinicians when marking-up will use input from other diagnostic scans which may not have been performed with the patient in the same position as the planning scan. The scans may be pre- or post-operative, and thus be anatomically different. There may also be inconsistencies in the timing and modalities of imaging available for different patients. There is obviously a role here for image registration methods. Image registration will obviously increase the accuracy of combining diagnostic data in the planning process. Furthermore, the application of more advanced methods, such as elastic rather than rigid-body registration methods, will improve accuracy, especially for patients with data acquired on different patient couches and patients in slightly different positions.
It is also difficult, over the course of a long study such as this, to monitor and control the use of notes and files for additional information used in marking-up as they are in constant clinical demand and so may not have been available (this was not recorded).
Contouring
When contouring, the clinician will therefore have to mentally translate visual information on certain planes into contours in a different 3D plane. This complex process leads to increased inaccuracies in the final contour volume and may contribute to the large standard deviations we see in this study. Other authors have noted potential errors introduced in this step in the overall process and report large variations in target volumes [2, 4].
The CT images assist in delineating only the gross tumour volume. Assessing microscopic tumour involvement is difficult, highly subjective, and sometimes controversial, as noted by some reports [23]. Subjectivity will also be introduced in margin growing and editing when going from CTV to PTV [4, 22]. Indeed, Yamamoto et al [2] report on reduced mark up precision for CTVs compared with GTVs highlighting this problem.
Although they mark up a GTV in a similar fashion, different oncologists in an institution may have slightly different philosophies when growing and editing margins. There can even be differences in interpretation of the ICRU [16] criteria for marking between clinicians [22].
PTV contouring precision should be better than GTV precision (differences in marking-up at the upper and lower extent of the target will have a smaller impact on the larger PTV volume), but may not be due to large variations in the size of margin added. The difference could be in the order of 0.5–1.0 cm in 2D, making a large difference to volume. Our results show little difference between intraoperator and interoperator volume precision for oncologists. Volume differences are presented as a percentage change from the original contour. In the case of GTVs with small numbers of slices, adding or removing a small numbers of slices will have a much greater effect on the percentage change than for PTVs which have 2–3 times the number of slices. At the extremes of the tumour volume (most superior and inferior positions) marking-up to include or exclude a slice can often be very subjective and will impact on precision results. This may explain the much improved standard deviations for the precision of PTV contour mark-up compared with GTV mark-up, which is in agreement with other authors [20].
Size and sites
In this work, the data are recorded for a sample size of 25 patients, which will have an influence on the uncertainties and statistical confidence. Although not large in statistical terms, the sample size is certainly larger then several similar studies in the literature [3, 4, 15, 19, 23]. Volumes were generated from full 3D CT data using DVH algorithms on our TPS. Volume data will therefore be accurate and should be an improvement on comparative assessments used by other authors involving 2D assessment of maximal tumour extent on specific CT slices [10], limited calculations from selected slices from a volume set [3] or volumes manually calculated from hardcopy films [22].
The cases in this work consist of small numbers from several different sites as shown in Table 1. Analysis of the data as a single group could obscure any benefit that may exist for a particular site(s). In certain circumstances, therefore, the data have been further split into smaller groups for consideration. With small numbers for individual sites, it was considered statistically prudent to merely split the data into two groups, namely brain and head and neck. The difficulty in delineating the target will depend on the site in question. Some authors report a variation between different groups of clinicians (radiologists and oncologists) when dealing with more or less "difficult" cases [19].
In the case of some resections, the oncologist may simply be marking-up a post-surgical cavity as a GTV, although technically this should be a CTV. In others, they can mark the site of the original tumour. Many of our patients had only cavities remaining. For the brain patients, the variation in marking was already so large that any difference made by the contrast may be too small to be detected. Some individuals seemed to be marking up the tumour cavity only, and some were marking a larger volume, more like a true CTV. Consistency was lacking and it was not always clear if the marked volumes were supposed to include a margin for microscopic disease or not. Thus, we could have been comparing unlike volumes for some patients. For these patients, getting good agreement between oncologist and radiologist mark-up may prove difficult and, indeed, we do see poor agreement between these clinicians for brain cases in particular. Other authors also report significant differences in mark-up between oncologist and radiologist [4, 19]. Yamamoto et al [2] also note a variation in precision between pre-operative and post-operative cases. Without contrast, contouring is far more dependent on the diagnostic scans. This leads to further inaccuracies if tumour volume changes due to surgery or chemotherapy have occurred.
The use of contrast may be very helpful for some sites, and significantly less so for others. Unfortunately, with low numbers and a large mix of sites, any large changes or improvements for a particular site may not be apparent when looking only at results for the entire group and only by running a much larger study may influences of this nature be more apparent. As an example, consider the three cases shown in Figures 2–4. Figure 2 shows pre- and post-contrast images for a tonsil patient. The post-contrast scan offers little additional information about the GTV. The PTV was never likely to show any significant change as it includes nodal groups in the neck, and so is not solely dependent on ascertaining the exact extent of the GTV. The head and neck tumours in general were much easier to outline, however, even using the diagnostic scan while voluming, as vessels and nodes were more easily distinguishable from other soft tissues. The efficiency of contouring is likely to improve with the use of contrast.
|
|
|
shows a set of pre- and post-contrast images for a nasopharynx patient. This is helpful in showing the extension into the cranial contents. This illustrates that direct tumour enhancement is sometimes useful in head and neck cases. Compare this with Figure 4, which shows images of a patient with a glioblastoma. Here, the periphery of the tumour, between tissue and oedema, is well visualized in the post-contrast image on the left, compared with the pre-contrast scan on the right. We observed that contrast was very helpful in identifying the GTV in patients who had had a biopsy or minimal debulking only. For patients who had maximal debulking of their brain tumours, there was little enhancement. There are several characteristics of the tumour itself which can influence the impact of contrast on imaging. Tumours in certain sites, e.g. oral cavity, are more likely to enhance with contrast due to increased vascularity compared with others, e.g. larynx. As mentioned above, surgery not only changes the anatomy, but also the vasculature and oedema can mimic tumour very well. Contrast should make contouring more consistent, and our results show that.
Finally, it must also be accepted, that although an expert in interpretation of medical images, the radiologist is not as expert in radiotherapy treatment planning. In agreement with other reports [4, 19], the input of the radiologist has shown to lead to significant mark-up changes, both in volume and isocentre shift, a stronger influence than that of using contrast alone. In certain circumstances, the radiologist may mark a considerably different contour to the oncologist, but from our data it is not possible to ascertain whether it is more or less accurate.
In the study, all patients were treated using a plan based on the unenhanced scans marked-up by the oncologist. It is not possible, therefore, to compare the outcome of patients treated on plans generated with and without contrast and therefore it is not possible to say clinically whether the use of contrast-enhanced scan data or radiologist input resulted in more accurate, and therefore, improved treatment. In essence, we cannot say which of the clinicians' mark-ups is the "true" one, or most accurate.
Others also note differences between radiologists and oncologists, with radiologists marking consistently smaller volumes [4, 19], which is the opposite of what we see in this study. However, in this study we have only a single radiologist compared with other reports performed with a larger and perhaps more representative group of radiologists.
|
Conclusions |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
It was reassuring to observe good intraclinician and interclinician precision for the GTVs and PTVs, with isocentre shifts within daily setup tolerance and plan coverage changes also acceptably small. Brain cases showed less precision than head and neck cases.
The use of contrast markedly improved the intraoncologist precision. The impact of using contrast caused greater differences for the radiologist than the oncologist. In fact, this difference was greater than the variation between them. This was more marked for the head and neck cases, although it was non-significant.
We found the radiologist marked significantly different volumes for both GTV and generated PTV. Also, the mean isocentre shifts for these contours were outside our daily setup tolerance.
Contrast enhanced planning appears to offer benefit in planning head and neck patients and those brain tumour patients who have not had a maximal debulking surgical procedure and so still have macroscopic tumour remaining.
Current address for Dr M McJury: Department of Medical Physics, The Northern Ireland Cancer Centre, Belfast City Hospital, Lisburn Road, Belfast BT9 7AB, UK.
The authors gratefully acknowledge support from Weston Park Research Fund (MM) and Yorkshire Cancer Research (KD, MHR).
Received for publication December 15, 2005. Accepted for publication February 7, 2006.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionConclusionsReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  S KIM, W RUSSELL, P PRICE, and A SALEEM Suboptimal use of intravenous contrast during radiotherapy planning in the UK Br. J. Radiol., December 1, 2008; 81(972): 963 - 969. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
BJR review of the year - 2006 Br. J. Radiol., March 1, 2007; 80(951): 147 - 151. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| BJR | DMFR | IMAGING | ALL BIR JOURNALS |
