Significance of breast boost volume changes during radiotherapy in relation to current clinical interobserver variations

Patient data 

Ten randomly selected patients with stage T1–2 invasive breast cancer treated with breast conservative surgery were included in this study. None of these patients received adjuvant chemotherapy. For each patient a standard radiotherapy planning CT scan was acquired several days before radiotherapy. The radiotherapy consisted of 25 fractions of 2Gy/fraction, 5 days per week to the whole breast, followed by a boost of 16Gy in eight fractions to the boost PTV. CT slice thickness was 3-mm with 3-mm separation, and the patient was positioned supine with both arms placed in an armrest above the head. Before scanning, the radiation technologist placed a copper wire around the palpable breast tissue in order to facilitate breast target volume delineation.

In addition to the planning CT, repeat CT scans were acquired 3, 5 and 7 weeks after acquisition of the planning CT. These repeat CT scans were automatically registered to the planning CT scan using the Pinnacle v7.6 image fusion module (Philips Medical Systems). The registration was visually inspected focussing mainly on the ribs on the ipsilateral side, and adapted if necessary. Registering the ribs was considered to correspond best with the widespread method of patient positioning at the treatment unit using (on-line) epid corrections.

Definition of volumes 

Four radiation oncologists delineated the VBV. Because previous studies have shown that providing guidelines can improve consistency among observers, a set of guidelines for contouring was developed [13]. These guidelines were first applied by all four radiation oncologists on one test patient case, after which the results were discussed with the observers. This was done to detect and correct ambiguities in the guidelines.

To avoid ambiguity, we will use the term visible boost volume (VBV) to describe the volume that was delineated. From this volume, a CTV and PTV can be derived using appropriate VBV to CTV and CTV to PTV margins, which may differ between institutes. By presenting our results in this way, we avoid them to be influenced by our a-priori choice of margins, but rather show results based on the observable features in the CT scans.

In the guidelines, we defined the VBV as the visible lumpectomy cavity including possible seroma, haematoma and all surgical clips. Furthermore, the VBV should be delineated in all CT slices extending over the VBV. Standardized window and level settings were used by all observers and each observer kept at least 1 week between subsequent delineations of CT scans of the same patient. In our institute, the boost CTV is defined as the expansion of the VBV by 1cm minus the tumour free resection margin and excluding the skin (with a margin of 0.5cm), the ribs and the pectoralis muscle. An additional margin of 1cm is taken into account in the expansion from boost CTV to boost PTV, again excluding the skin with a margin of 0.5cm.

Volume changes over time 

The boost volume was calculated for all delineations and absolute and relative volume differences between CT sessions were analysed for each patient separately. Using the Pinnacle software, the common and encompassing VBVs of the four CT datasets per patient were generated. The ratio of the common and encompassing VBV of a patient, referred to as the conformity index (CIovertime), was calculated to analyse the variation of the VBV over time. If no variation is noted this ratio is equal to one. On the other hand, if no volume is common to all four delineated volumes, the ratio equals 0. For each patient, the CIovertime is calculated for each observer separately, and then averaged over the observers.

In order to quantify the relevance of replanning, we represent the VBVs by spheres of the same volume and analyse the difference in diameter of the spheres for the first and second CT scan. This gives a quantitative indication of the change in the conformal field sizes that would be found when replanning the treatment based on a CT taken 3 weeks after the first CT.

Interobserver variations 

To detect possible systematic observer differences independent from the absolute delineated volumes, the volumes delineated by the four observers were ranked 1 (smallest delineation) to 4 (largest delineation) per CT dataset. Thus, 40 values per observer were generated (4 datasets per patient times 10 patients). To detect if observer differences are consistent between scans, these rankings were averaged per observer, for the planning CTs and the three consecutive CT datasets separately. If an observer would always delineate the smallest VBV, this will result in averaged rankings of 1. An observer who always delineates the largest VBV would score averaged rankings of 4.

Significance of volume changes in the presence of observer variations 

To determine whether boost volume changes over time and interobserver variations are independent factors, a repeated measures analysis of variance was performed.

Because of the restricted number of measurements, sphericity of the data could not be guaranteed. Therefore the number of degrees of freedom in F was corrected for violation of sphericity (Greenhouse–Geisser correction), which led to slightly higher p values. Statistical calculations were performed using SPSS v12.0 (SPSS Inc.).

Volume changes over time 

In Fig. 1 the visible boost volume per patient is given as a function of the time from lumpectomy (or re-excision if applicable), averaged over the four observers. One can see that there is, for most patients, a decrease of the volume over time. This decrease is most pronounced between CT1 and CT2, for example for patients 1, 2 and 10, for whom the time between surgery and the first CT scan is shortest.

View Large Image Download to PowerPoint

Fig. 1. Change of VBV as a function of time after surgery. The average volumes drawn by the four observers are shown. The figure is split into two for ease of visualisation (note the different y-axis scaling). The legends refer to the patient numbers.

Patients 1, 4, 5 and 10 had a visible seroma, and it seems that for these patients the volume decrease is in general larger than for patients who did not have a visible seroma. However, the number of patients in this study is too small to perform a valid subgroup analysis on this parameter.

In the approach where the VBVs are assumed to be spherical, the VBV diameter averaged over all patients reduces from scan 1 to scans 2, 3 and 4 (from 43 to 38mm, 37 and 37mm, respectively). This means that replanning at 3, 5 or 7 weeks after the first planning CT will on average result in an estimated field size reduction of 5, 6 or 6mm, respectively.

The variation of the boost volume and boost position over time is illustrated in Fig. 2 for a patient who showed a relatively low CIovertime of 0.1 (patient 5). The low CIovertime indicates a large change in time of either the boost volume, the boost position or both. For this patient the ribs of the four consecutive scans could be matched accurately, with residual differences of the order of a few mm. The larger differences in the contours of the breast over time – as seen in Fig. 2, Fig. 3 – reveal the flexibility of the breast. Large breasts in general show larger contour variations. The breast volume for patient 5 was 2736cm3, while the volume of the other patients ranged between 487 and 1734cm3. The CIovertime was on average 0.25 (range: 0.08–0.36), which can be seen in Fig. 5.

View Large Image Download to PowerPoint

Example of variation of VBV over time in both volume and place (Patient 5). The delineation of one observer (number 3) is shown. The yellow contour is drawn on the underlying CT dataset. The blue contours are overlays from the other CT datasets.

View Large Image Download to PowerPoint

Example of a relatively small interobserver variation (Patient 5). The green contour depicts the common volume (Comoverobservers) while the red volume depicts the encompassing volume (Sumoverobservers). Datasets CT1 to CT4 are shown from left to right.

Interobserver variations 

Apart from the large variation in VBV over time, we also found large interobserver variations. The relative volume difference between observers varied widely between CT scans, from 10% for scan 4 for patient 5 (average volume of 43cm3) to 149% for scan 4 for patient 6 (average volume of 37cm3). Absolute volume differences between observers in a single CT scan ranged from 5 to 74cm3 (scan 2 for patient 4, average volume: 14cm3 and scan 2 for patient 1, average volume 77cm3, respectively).

The above examples clearly show that the interobserver variation depends highly on the CT scan. The disagreement between observers is large for e.g. patient 6, in which a diffuse area of denser tissue was visible extending from the skin to the ribs. For e.g. patient 4, whose CT scan was considered easy to delineate because of the presence of a clearly visible seroma in a region of less dense breast tissue, the disagreement between observers is relatively small. For all patients, the observers agree that the decrease in volume is largest from scans 1 to 2 and smaller for the other pairs of consecutive scans.

Compared to the average size of 40cm3 of the volume delineated in the first CT, the consecutive scans 2, 3 and 4 showed an average decrease of 29% (12cm3), 34% (13cm3) and 37% (15cm3), respectively.

In Fig. 3 the axial contours of the common (Comoverobservers) and encompassing (Sumoverobservers) volumes over all observers are shown for patient 5 and one can see that for this patient, the delineations of the four observers correspond relatively well. This can also be concluded from the relatively high conformity index over all observers (CIoverobservers=0.52). Fig. 4 shows that the differences between observers do not change considerably over time. Observer 2 often delineated the largest volume (26/40 delineations), while observer 4 often delineated the smallest volume (20/40 delineations). Delineations by observers 1 and 3 were most often close to the average over all observers. This can also be seen when ranking the size of the delineations: observer 2 scores an average ranking of 3.6 whereas observer 4 scores an average ranking of 1.7.

View Large Image Download to PowerPoint

Fig. 4. Ranking of delineated volume size per observer for the four CT datasets. The average rank, with ranking from 1 (smallest delineated volume) to 4 (largest delineated volume) per observer over all 10 patients, is shown.

For all scans, the CIoverobservers ranged from 0.11 to 0.52, with an average of 0.31. This is in the same order of the conformity of the variations over time as shown in Fig. 5 (average CIovertime=0.25, range: 0.08–0.36).

View Large Image Download to PowerPoint

Fig. 5. Conformity indices CIoverobservers (black) and CIovertime (white). When CIoverobservers is larger than CIovertime, this means that the variation over time is larger than the variation between observers.

Significance of time trends and observer variations 

Despite the relatively large interobserver variation, there is a significant decrease in volume between consecutive scans (p=0.02). The decrease in volume between consecutive scans was most prominent between the first and the second CT scan. When the first CT dataset was excluded, the volume decrease was no longer significant (p=0.21). On the other hand, when taking this volume decrease over time into account, the interobserver variation remains significant (p=0.003). This also remained significant when the first CT dataset was left out of the analysis (p=0.002).