Prostate brachytherapy
Prostate post-implant dosimetry: Interobserver variability in seed localisation, contouring and fusion

https://doi.org/10.1016/j.radonc.2012.06.014Get rights and content

Abstract

Aim

Reliable post-implant evaluation of prostate seed implants requires optimal seed identification and accurate delineation of anatomical structures. In this study the GEC-ESTRO groups BRAPHYQS and PROBATE investigated the interobserver variability in post-implant prostate contouring, seed reconstruction and image fusion and its impact on the dose–volume parameters.

Materials

Post-implant T2-TSE, T1-GE and CT images were acquired for three patients, in order to evaluate four post-plan techniques: (a) CT, (b) T1 + T2, (c) CT + T2, (d) CT + T1int + T2. Three interobserver studies were set up. (1) Contouring: the CTV-prostate was delineated on CT and T2 by eight physicians. Additionally one reference contour was defined on both image modalities for each patient. (2) Seed reconstruction: seven physicists localised the seeds on T1 and CT, manually and with CT seed finder tools. A reference seed geometry was defined on CT and T1. (3) Fusion: six physicists registered the image sets for technique (b)–(d), using seeds (if visible) and anatomical landmarks. A reference fusion was determined for each combined technique.

Results

(1) The SDref for contouring (1 SD with respect to the reference volume) was largest for CT (23%), but also surprisingly large for MRI (17%). This resulted in large SDref values for D90 for all techniques (17–23%). The surprisingly large SDref for MRI was partly due to variations in interpretation of what to include in the prostate contour. (2) The SDref in D90 for seed reconstruction was small (2%) for all techniques, except for T1 + T2 (7%). (3) The SDref in D90 due to image fusion was quite large, especially for direct fusion of CT + T2 (16%) where clearly corresponding landmarks were missing (seeds hardly visible on T2). In general, we observed large differences in D90 depending on the technique used.

Conclusions

The dosimetric parameters for prostate post-implant evaluation showed large technique-dependent interobserver variabilities. Contouring and image fusion are the ‘weak links’ in the procedure. Guidelines and training in contouring together with incorporation of automated fusion software need to be implemented.

Section snippets

Patients

Three patients treated with 125I seed implants at the University Hospital Gasthuisberg (Leuven) were selected. The pre-implant prostate volumes were 38, 21 and 42 cm3, implanted with 76, 62 and 87 seeds respectively. The dose prescription was 145 Gy to the prostate CTV.

Image acquisition

Post-implant CT, MRI and X-ray images (0° and 60°, for seed count) were acquired 28 days after implantation. CT images of 3 mm were acquired on a 4-slice spiral CT (Somatom, Siemens). MRI scans were performed on a 1.5 T scanner

Interobserver variability in contouring

Fig. 1 shows the results of the prostate delineation on CT and T2. Table 2 gives the calculated volumes. In general, the mean observer volumes corresponded quite well with the reference volumes (except for patient 2 on CT). In almost all cases the CT volumes were larger than the T2 volumes (ratio 1.20 ± 0.12). This finding corresponds with external beam literature [14], [15], [16].

Averaged over the patients, the interobserver variability in contoured prostate volume, expressed as SDref (1 SD with

Interobserver variability in contouring

Several studies have reported a large interobserver variability for CT based post-implant contouring [7], [8], [9]. It is not possible to compare SD data however due to the small number of observers in these studies. For MRI, data on post-implant prostate contouring are scarce and even contradictory. In a single-centre study it was found that the MRI prostate volume exceeded the CT volume on average by 9% [17]. Another study found slightly larger volumes for post-implant contouring on CT than

Conclusion

The dosimetric parameters for prostate post-implant evaluation showed large technique-dependent interobserver variabilities. Contouring and image fusion are the ‘weak links’ in the procedure. This can be approached by establishing widespread training across Europe. Image fusion uncertainties could additionally be tackled with automated fusion software.

Acknowledgements

The authors want to thank the BRAPHYQS and PROBATE members (Bashar Al-Qaisieh, Ann Henry, Christian Kirisits, Peter Niehoff, Alfredo Polo, Alex Rijnders, Carl Salembier, Marco Van Vulpen) and Tom Budiharto, Liesbeth De Wever, Gregor Goldner, Kenneth Poels, Marinus Moerland, Raymond Oyen, for their contribution.

References (27)

Cited by (59)

View all citing articles on Scopus
View full text