CT- and MRI-based seed localization in postimplant evaluation after prostate brachytherapy

Brachytherapy 12 (2013) 580e588

CT- and MRI-based seed localization in postimplant evaluation after prostate brachytherapy Marisol De Brabandere1,*, Bashar Al-Qaisieh2, Liesbeth De Wever3, Karin Haustermans1, Christian Kirisits4, Marinus A. Moerland5, Raymond Oyen3, Alex Rijnders6, Frank Van den Heuvel1, Frank-Andre Siebert7 1 Department of Radiation Oncology, University Hospital Gasthuisberg, Leuven, Belgium Department of Medical Physics and Engineering, St James’s Institute of Oncology, Leeds, United Kingdom 3 Department of Radiology, University Hospital Gasthuisberg, Leuven, Belgium 4 Department of Radiotherapy, Medical University of Vienna, Comprehensive Cancer Center, Vienna, Austria 5 Department of Radiation Oncology, University Medical Center Utrecht, Utrecht, The Netherlands 6 Department of Radiation Oncology, Europe Hospitals, Brussels, Belgium 7 University Hospital of Schleswig-Holstein, Clinic of Radiotherapy, Kiel, Germany 2

ABSTRACT

PURPOSE: To compare the uncertainties in CT- and MRI-based seed reconstruction in postimplant evaluation after prostate seed brachytherapy in terms of interobserver variability and quantify the impact of seed detection variability on a selection of dosimetric parameters for three postplan techniques: (1) CT, (2) MRI-T1 weighted fused with MRI-T2 weighted, and (3) CT fused with MRI-T2 weighted. METHODS AND MATERIALS: Seven physicists reconstructed the seed positions on postimplant CT and MRI-T1 images of three patients. For each patient and imaging modality, the interobserver variability was calculated with respect to a reference seed set. The effect of this variability on dosimetry was calculated for CT and CT þ MRI-T2 (CT-based seed reconstruction), as well as for MRI-T1 þ MRI-T2 (MRI-T1ebased seed reconstruction), using fixed CT and MRIT2 prostate contours. RESULTS: Averaged over three patients, the interobserver variability in CT-based seed reconstruction was 1.1 mm (1 SDref, i.e., standard deviation with respect to the reference value). The D90 (dose delivered to 90% of the target) variability was 1.5% and 1.3% (1 SDref) for CT and CT þ MRI-T2, respectively. The mean interobserver variability in MRI-based seed reconstruction was 3.0 mm (1 SDref), and the impact of this variability on D90 was 6.6% for MRI-T1 þ MRI-T2. CONCLUSIONS: Seed reconstruction on MRI-T1eweighted images was less accurate than on CT. This difference in uncertainties should be weighted against uncertainties due to contouring and image fusion when comparing the overall reliability of postplan techniques. Ó 2013 American Brachytherapy Society. Published by Elsevier Inc. All rights reserved.

Keywords:

Prostate; Postplanning; Seed reconstruction

Introduction Permanent prostate implantation has become a widely available treatment procedure for low-risk prostate cancer Received 4 January 2013; received in revised form 12 February 2013; accepted 7 June 2013. None of the authors declare a conflict of interest concerning the material presented in this manuscript. * Corresponding author. Department of Radiation Oncology, University Hospital Gasthuisberg, Herestraat 49, B-3000 Leuven, Belgium. Tel.: þ32-16-34-76 45; fax: þ32-16-34-76-10. E-mail address: [email protected] (M. De Brabandere).

patients. According to the latest American Brachytherapy Society (ABS) recommendations, high-risk patients are also eligible to permanent seed brachytherapy, provided they receive supplemental external beam radiotherapy, whereas intermediate-risk patients should be considered on an individual case basis (1). Accurate dosimetric evaluation of completed implants (postplanning) is crucial in developing a proper implant technique and in predicting potential efficacy and is therefore recommended in several publications (1e3). The standard imaging technique for postimplant determination of prostate volume and implanted seed reconstruction is CT. It allows adequate and fast seed

1538-4721/$ - see front matter Ó 2013 American Brachytherapy Society. Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.brachy.2013.06.003

M. De Brabandere et al. / Brachytherapy 12 (2013) 580e588

localization (4e6) but has inferior quality compared with MRI with respect to potential for target delineation (7e10). Therefore, several groups have tried to develop alternative postplan imaging techniques using MRI. One approach is to fuse CT and T2-weighted MR images, a planning technique advocated by several authors (11e14). This technique requires, however, both CT and MRI scanning, followed by a fusion procedure, making it expensive and labour intensive. For this reason, other institutes have investigated approaches focusing on MRI pulse sequences suitable for seed detection. One possibility is to combine different MRI sequences, one optimized for seed visualization, usually using T1- or proton densityeweighted images, and fuse this with a T2-weighted set (15e17). There have also been attempts to develop single MRI sequences suitable for both seed localization and target contouring (18e20), but this approach has not yet convincingly resulted in useful sequences. With the growing interest to use MRI in postimplant evaluation, it is important to investigate and compare the uncertainties associated with seed detection on CT and MRI. Although some phantom studies are available (21, 22), comparative studies on patient images have not been published. An obvious difficulty is that on patient images the exact positions of the seeds are not available to compare the reconstructed seed positions to. To our knowledge, a systematic investigation of MRI-based seed reconstruction accuracy for prostate brachytherapy is not available in the literature. In 2010, the Groupe Europeen de Curietherapie- European SocieTy for Radiotherapy & Oncology working group BRAchytherapy PHYsics Quality assurance System set up a study to evaluate seed reconstruction uncertainties in CTand MRI-based prostate postplanning. Methods and materials General study set-up For three patients, the seed reconstruction accuracy on CT and MRI images was compared by means of an interobserver variability study. The seed reconstruction uncertainty was evaluated on two levels. First, the deviation in reconstructed seed positions of seven observers with respect to a reference seed set was determined. This was done on CT and MRI-T1eweighted images. More details on the seed reconstruction procedure are given below.

581

On a second level, the impact of the interobserver variability in seed reconstruction on a selection of dosimetric postimplant parameters was calculated for three postplan techniques: (1) CT only, (2) MRI-T1 fused with MRI-T2, and (3) CT fused with MRI-T2. A clinical target volume for prostate (CTV-P) was defined on CT images for technique (1) and on MRI-T2 images for techniques (2) and (3), respectively. The observer’s seed geometries were imported in the study containing these CTV contours, and the resulting dosimetric evaluation parameters were compared with those obtained with the reference seed geometry. Patient and image information In the reference center (University Hospital Leuven), three patients treated with I-125 seed implants were selected. The only specific criteria were to include patients with varying preimplant prostate volumes (21, 38, and 42 cm3) and with a different number of implanted seeds (62, 76, and 87 seeds, respectively). The dose prescription was 145 Gy to the prostate gland. All patients were implanted with stranded seeds, model 7000 RAPID Strand consisting of OncoSeed model 6711 seeds spaced at a fixed distance of 10 mm within an absorbable braided carrier (Oncura, Unit of GE Healthcare, Chalfont St. Giles, UK). Postimplant MRI, CT, and X-ray imaging was performed 28 days after implantation. Two series of MR images were acquired on an Intera 1.5T scanner (Philips Healthcare, part of Royal Philips Electronics, Eindhoven, The Netherlands). For seed reconstruction, a T1-weighted gradient echo sequence with 4.0 mm slice thickness was acquired for each patient. This image sequence, further referred to as T1, was previously optimized for proper seed visualization using a dedicated phantom (21). A second MRI sequence of T2weighted turbo spin echo images with 3.0-mm slice thickness was acquired. This sequence, further denoted as T2, offers good soft tissue imaging characteristics and was used for prostate delineation in the dosimetry part of this study. More details on MR imaging parameters are listed in Table 1. CT images of 3.0-mm slice thickness were acquired on a Somatom Sensation 4 CT scanner (Siemens, Erlangen, Germany). The CT images were used for both seed reconstruction and contouring. More details on the imaging parameters can be found in Table 1. The postimplant x-ray images were taken on a simulator (Acuity, Varian Medical Systems, Inc., Palo Alto, CA) at angles 0 and 60 and were used for seed counting.

Table 1 CT and MRI parameters used in the study Imaging modality

Type

Slice thickness (mm)

In-plane resolution (mm)

Pitch

FOV (mm)

TR (msec)

TE (msec)

Echotrain

Scanning time (min)

T1 T2 CT

GE TSE spiral

4.0 3.0 3.0

0.73 0.73 0.97

d d 10/4  2.5

375 375 500

247 2700 d

5 100 d

1 48 d

5.1 5.2 1.9

FOV 5 field of view; GE 5 gradient echo; TSE 5 turbo spin echo; TR 5 repetition time; TE 5 echo time. The pitch is defined as table feed/total detector width of the collimated beam.

582

M. De Brabandere et al. / Brachytherapy 12 (2013) 580e588

Seed reconstruction The postplan CT and T1 images of the three patients were examined by seven physicists, all with clinical experience with prostate postplanning. The physicists all used the same treatment planning system (TPS), Variseed 8.0 (Varian Medical Systems, Inc., Palo Alto, CA), to avoid inter-TPS variations in the results. Seed reconstruction was performed first on T1 and later on CT, with an interval of at least 2 weeks to avoid bias. The physicists had the plan details (number of implanted seeds, seed configuration) and postimplant x-ray images available. On CT, seeds were localized using TPS seed finder tools, and manual adjustments could be made if necessary. On T1 images, the seeds were localized manually as an automatic seed finder algorithm for MR images was not implemented in the software. Seed orientation could not be adjusted and hence matched the longitudinal scan axis direction for all seeds. The reconstructions obtained by the observers were evaluated against a reference seed geometry, determined by consensus by two experienced physicists. During a common reconstruction session, a separate CT- and T1-based seed set was defined for each patient based on axial, sagittal, and coronal views, which was then taken as the ‘‘truth.’’

a CTV-P(rostate), delineated by two uro-radiologists from the University Hospital Leuven. CTV-P is defined as ‘‘the postimplant contour of the prostatic gland defined by the capsule on radiologic examination’’ (3). The prostate D90 (dose delivered to 90% of the prostate), V100, and V150 (volume covered by 100% and 150% of the prescription dose, respectively) were calculated. All dose calculations were performed using the Oncura seed model 6711, based on the line source approximation model (TG-43-parameters as in (23)). The dose grid resolution was 1.000 mm/pixel in all directions. Image registration of technique T1 þ T2 and CT þ T2 was performed by the same two physicists who defined the reference seed geometry. Fusion of T1 þ T2 was basically a zero-match as the two coordinate systems were identical, with only a minor manual adjustment in Patient 1, who had slightly shifted in between the two scans (!2 mm in left-right direction). CT þ T2 registration was performed based on at least four manually defined corresponding matching points. When possible, seeds were used as corresponding landmarks, but because seed visibility was not always clear on T2 images, anatomic structures were used as well.

Results Analysis of reconstructed seed positions The interobserver variability in seed reconstruction was defined as the average magnitude of deviation between the observer’s and the reference seed positions. To calculate this deviation, a Matlab (Mathworks, Inc., Natick, MA) program was written to analyze the reconstructed seed positions in a two-step procedure. First, the observer’s seeds were assigned to the closest reference seeds, with a programmed constraint to be within a predefined distance in any direction from the reference position (i.e., 15 mm). Seeds outside this predefined distance and redundant seeds were denoted as ‘‘not identified’’ and excluded from further analysis. Defining the acceptance constraint required finding a balance between including reconstructed seeds within reasonable distance (small constraint) and including sufficient seeds to obtain a realistic contribution of the reconstructed seeds in the uncertainty calculation (large constraint). In a second step, the absolute deviation was computed for every assigned observer-reference seed pair in the set. The mean deviations were calculated for the x-, y-, and z-direction separately, corresponding with the left-right, anterior-posterior, and cranio-caudal directions, respectively. For each reconstructed seed, the global deviation in three dimensions was calculated as the vector size. Analysis of dosimetric parameters The effect of the interobserver variability in CT- and T1-based seed reconstruction on dosimetry was evaluated for three techniques: (1) CT, (2) T1þT2, and (3) CT þ T2. Dosimetric parameters were calculated for

Interobserver variability in seed reconstruction Table 2 gives an overview of the seeds reconstructed by the reference and by the observers on CT and T1 images. For Patient 1 and 2, the number of seeds reconstructed by the reference was the same as implanted. For Patient 3, the reference plan contained 3 seeds less as implanted, as the two postimplant x-rays revealed 3 seeds that had migrated out of view. On CT, the observers generally localized the correct number of seeds. For Patient 1 and 2, some observers identified less seeds than the reference. In Patient 3, many observers reconstructed 87 seeds instead of 84, which is equal to the number implanted, but not in agreement with the postplan x-ray information. Of these reconstructed seeds, on average, 98% could be assigned to a reference seed for CT, while the mean seed assignment rate for T1 was 93%. For CT, the deviation between the observer’s and the reference seed positions ranged from 0.5 to 2.5 mm over all plans, with a deviation of 1.1  0.5 mm averaged over all observers and patients. For MRI, the deviation ranged from 1.7 to 4.7 mm, with a mean deviation of 3.0  0.9 mm. On both imaging modalities, the mean deviations were largest in the z-direction (longitudinal axis). In all cases, the interobserver variability in reconstructed seed positions was larger on MRI-T1 than CT. For illustrative purposes, Fig. 1 shows the relative distributions of the seed deviations for Patient 2, including all observers. The distributions are narrow for in-plane deviations (x and y) and broader in craniocaudal direction (z). In

M. De Brabandere et al. / Brachytherapy 12 (2013) 580e588

583

Table 2 Results of the seed reconstruction study on CT and T1 images, as performed by seven observers CT

MRI Deviation (mm)

Obs Patient 1 1 2 3 4 5 6 7 Obs mean Patient 2 1 2 3 4 5 6 7 Obs mean Patient 3 1 2 3 4 5 6 7 Obs mean Patient mean

Deviation (mm)

Rec

Ass

Global

x

y

z

Rec

Ass

Global

x

y

z

0 0 1 0 0 0 1

0 1 1 1 0 0 1

0.5 1.3 0.9 1.1 0.6 0.9 0.9 0.9

0.1 0.2 0.1 0.2 0.1 0.1 0.1 0.1

0.1 0.5 0.1 0.3 0.1 0.1 0.1 0.2

0.5 1.2 0.9 1.0 0.6 0.9 0.9 0.9

0 0 1 0 0 0 0

1 0 2 3 1 0 2

2.0 1.9 3.0 2.8 1.7 2.1 3.1 2.4

0.7 0.5 1.1 1.1 0.5 0.5 1.4 0.8

0.5 0.5 1.3 1.2 0.5 0.6 1.5 0.9

1.8 1.8 2.5 2.4 1.5 2.0 2.4 2.0

0 0 1 0 0 0 þ2

0 0 2 3 1 0 0

0.5 0.8 1.0 2.5 0.8 0.9 1.6 1.1

0.1 0.1 0.3 1.2 0.4 0.1 0.7 0.4

0.2 0.2 0.3 1.2 0.4 0.2 0.7 0.4

0.4 0.8 0.9 1.8 0.6 0.8 1.2 0.9

0 0 0 0 0 1 0

2 2 7 7 4 2 8

1.8 2.3 2.6 3.2 2.2 2.6 2.0 2.4

0.5 0.6 0.8 1.7 0.8 0.8 0.7 0.9

0.5 0.8 0.8 1.3 0.7 0.6 0.8 0.8

1.6 2.1 2.4 2.4 2.0 2.4 1.7 2.1

0 þ3 0 þ3 þ3 þ3 0

0 1 1 3 2 1 1

1.0 1.3 1.4 1.1 1.1 1.8 1.4 1.3

0.0 0.2 0.2 0.2 0.4 0.6 0.2 0.2

0.1 0.4 0.4 0.4 0.5 0.7 0.4 0.4

1.0 1.2 1.3 1.1 0.9 1.6 1.3 1.2

0 þ3 2 þ3 þ3 þ3 þ3

6 10 10 13 8 7 7

3.6 4.7 4.0 4.3 4.3 3.6 3.9 4.1

1.7 1.8 1.5 2.1 2.1 1.3 1.8 1.8

2.0 2.5 2.1 2.0 2.2 2.0 1.8 2.1

2.4 3.6 3.1 3.2 3.0 2.7 3.0 3.0

1.1

0.3

0.4

1.0

3.0

1.1

1.2

2.4

Shown are the difference between the number of reconstructed (Rec) and assigned (Ass) seeds compared with the number of reference seeds. The reference seed number was 62, 76, and 84 for patients 1, 2, and 3, respectively. The mean deviation between reconstructed and reference seed positions (mm) is given for each patient per observer and averaged over all observers. The patient mean is calculated for both techniques. This is done for the global deviation (vector size) and for the deviation in three directions separately, with x, y, and z, respectively, the lefteright, anterioreposterior, and craniocaudal directions.

all directions, the T1 distributions are broader and have longer tails than CT distributions. Interobserver variability in dosimetry parameters Table 3 shows the D90, V100, and V150 values based on the reconstructed seed positions, for each patient and for the patient mean. The CT-based seed distributions were used for postplan evaluation of CT and CT þ T2, whereas the T1-based seed distributions were used for calculation of the dose parameters obtained with T1 þ T2 postplanning. The effect of the variability in seed localization is reflected in the standard deviation (SD) (k 5 1) of the investigated parameters. For postplan techniques CT and CT þ T2, the interobserver variability in D90 was less than 2% in all patients. The D90 SDref values (1 SD with respect to the reference value) for T1 þ T2 were distinctively larger, that is, 6.6% averaged over all patients, reflecting the larger seed reconstruction uncertainty on MR images. The difference was statistically significant according to an F-test: p-value 0.0005 between CT and T1 þ T2 and p-value 0.0055

between CT þ T2 and T1 þ T2. The interobserver variability in V100 was small (patient mean SDref ! 1%) for the postplan techniques using CT-based reconstruction. For T1 þ T2, the SDref was about 3%. The V150 showed an interobserver variability of about 3.6% and 3.2% for CT and CT þ T2, respectively, and about 8% for T1 þ T2. Absolute dose values In general, the observer mean D90 values were close to the reference values. For all patients, the evaluated parameters differed substantially depending on the postplan technique used. CT alone systematically resulted in the lowest D90 values, whereas the highest D90 values were found for CT þ T2. Similar figures were observed for the prostate V100 and V150.

Discussion Evaluation of prostate dosimetry after permanent seed brachytherapy is essential to obtain feedback on implant

584

M. De Brabandere et al. / Brachytherapy 12 (2013) 580e588

Fig. 1. Relative distribution of the deviations of the reconstructed seeds with respect to the reference seeds, including all observers, for patient 2. The left and right columns show CT- and T1-based reconstructions, respectively, in lefteright (x), anterioreposterior (y), and cranio-caudal (z) directions. CT bin size was 0.2 mm in the x/y directions and 0.4 mm in the z-direction. The T1 bin size was 0.5 mm in the x/y directions and 1 mm in the z-direction.

quality and evaluate individual patient results. Because CT alone comes with substantial interobserver variation in contouring and subsequent dosimetric interpretation (7, 10, 17), MRI-based solutions have been explored for postplanning. Most straightforward is the fusion of CT with MRI, by combining the excellent seed localization properties of CT with the superior soft tissue contrast resolution on MRI-T2. However, to reduce the cost and workload of registering two imaging modalities, MRI sequences have been explored on which seeds can be identified. Ideally, such sequences provide sufficient seed and soft tissue contrast to use them as a single modality postplan

technique. The group of Dubois et al. (18) developed an MRI sequence visualizing both prostate gland and implanted sources by accentuating the artifact produced by the sources using a proton densityeweighted fast spin echo sequence. The uncertainties in seed localization, especially for extra-prostatic seeds, diminished the advantage of improved volume definition. Tanaka et al. (20) used a contrast enhanced T1-weighted sequence (CE-T1) for MRI-based postimplant dosimetry and compared this with CT/MRI fusionebased dosimetry. The D90, V100, and V150 were significantly overestimated in MRI-based dosimetry compared with CT/MRI, which was attributed to the

Table 3 Postimplant D90, V100, and V150 values calculated for three patients and the patient mean, based on the reconstructed seed positions of seven different observers D90 (Gy) Patient 1 Reference Obs mean SDref (%) Patient 2 Reference Obs mean SDref (%) Patient 3 Reference Obs mean SDref (%) Patient mean Reference Obs mean SDref (%)

V100 (%)

V150 (%)

CT

T1 þ T2

CT þ T2

CT

T1 þ T2

CT þ T2

CT

T1 þ T2

CT þ T2

122 121 1.2

123 126 7.8

128 128 0.6

82.9 82.8 0.4

84.9 85.8 2.8

86.6 86.6 0.3

57.5 57.4 1.1

56.7 61.9 11.3

65.2 65.2 1.1

119 118 1.7

128 128 4.5

147 147 1.9

80.0 79.5 1.3

85.1 85.4 1.9

90.6 90.4 0.9

49.0 48.6 4.6

57.4 55.5 5.3

57.3 57.0 4.7

133 134 1.5

163 167 7.6

165 167 1.4

85.8 86.2 1.0

94.0 95.7 3.6

95.5 95.7 0.4

55.1 57.3 5.0

69.9 66.7 7.7

64.2 66.3 3.8

124 124 1.5

138 140 6.6

147 147 1.3

82.9 82.8 0.9

88.0 89.0 2.8

90.9 90.9 0.5

53.9 54.4 3.6

61.3 61.4 8.1

62.2 62.8 3.2

D90 5 dose delivered to 90% of the prostate; V100 5 volume covered by 100% of the prescription dose; V150 5 volume covered by 150% of the prescription dose. The interobserver variability SDref (%) is defined as 1 standard deviation with respect to the reference value. The parameters were determined for CT, T1 þ T2, and CT þ T2.

M. De Brabandere et al. / Brachytherapy 12 (2013) 580e588

low success rate in locating extra-prostatic seeds. It was concluded, however, that CE-T1 is still better than CT for postplanning because CE-T1 dosimetry was closer to CT þ T2 dosimetry than CT dosimetry. McLaughlin et al. (24) compared the imaging characteristics of several MRI sequences in terms of prostate and seed visibility. It was concluded that T1-weighted images, either with or without fat suppression, are inferior to T2 for target delineation. In addition, they stated to be unable to identify an MRI technique that clearly defined seeds inside and outside the prostate reliably enough to do postimplant dosimetry without CT imaging. It seems that an MRI sequence providing sufficient information for simultaneous identification of seeds and target is not available at present. Image registration remains necessary when using MRI without compromising the ability to discern soft tissues. T2-weighted images should then be combined with either CT or an MRI sequence optimized for seed detection. Therefore, it was decided to exclude MRI as a single modality technique for postplanning. Seed reconstruction accuracy was evaluated for a gradient echo T1-weighted sequence optimized specifically for seed visualization (21) to be used in combination with T2. This was compared with the seed reconstruction accuracy on CT images. The impact on dosimetry was investigated for CT, T1 þ T2, and CT þ T2. Interobserver variability in seed reconstruction Ideally, the accuracy of seed reconstruction is assessed by comparing reconstructed seed positions with exactly known reference positions. This is only possible in a CT and/or MRI compatible phantom with implanted seeds at known positions where the material has the same seed visualization characteristics as prostate tissue. However, although such phantom studies provide accurate assessments on seed reconstruction accuracy, they may lack traceability to actual postplanning because there are no anatomic impurities influencing seed localization. Furthermore, phantom construction would become extremely complex if one wanted to incorporate different organs on which to calculate the effect on the dosimetric parameters. To take into account these aspects, patient images are needed. In the present study, CT-based seed reconstruction variability was on average 1.1  0.5 mm. This uncertainty corresponded well with results from a phantom study, previously performed by this group, where the reconstruction uncertainty on 3 mm CT was determined as 0.9  0.6 mm (21). The results were also in line with a multicentre phantom study performed by Siebert et al. (22) in which reconstruction deviations of 0.2e0.6 mm (1 SD) were found. In general, it was concluded that CT-based seed reconstruction uncertainties are small and that the reconstruction accuracy is well predicted by phantom studies. On MRI, the mean interobserver variability in seed reconstruction was 3.0  0.9 mm in this patient study,

585

which is larger than on CT. The use of a seed finder tool for CT, making seed localization more robust and objective, can explain this to some extent only. Seed distinction on T1 was more complicated because of the local magnetic field distortions caused by the seeds themselves. It was shown that the artifacts produced by these distortions create a complicated pattern around the seeds that depend on the direction and magnitude of the read gradient (25, 26). The appearance of the sources makes seed distinction difficult, especially when the seeds cluster together. Another factor adding to the difference in CT and MRI seed reconstruction accuracy was the presence of anatomic impurities on the patient images that disturb seed visualization on T1 considerably more than on CT. Detailed image analysis revealed that misplaced seeds were found all over the prostate, but most frequently in extra-prostatic areas and where seeds were close together, for example, by defining too many seeds in areas of seed ‘‘chains.’’ In fact, the difference in seed reconstruction accuracy between CT and T1 would be even larger if the nonassigned seeds (outside the predefined acceptance distance) were also included in the calculation of the interobserver variability. Most of the nonassigned seeds were sources that were erroneously reconstructed twice on one seed signal void (on two adjacent slices), whereas another signal void remained undetected. MRI-based interobserver variability in seed reconstruction found in this patient study was compared with the results determined previously in a gel-based (agarose) phantom (21). In this phantom, seed reconstruction deviations on 4 mm T1-weighted images were assessed to be 2.0  1.6 mm and 1.6  1.2 mm on a Siemens and a Philips scanner, respectively. Hence, MRI seed detection was slightly less accurate on patient images than in a phantom. This is plausible, given the presence of anatomic impurities in prostate images. In the present study, a slice thickness of 3.0 mm was chosen for CT as this reflects clinical practice for many centers. As a smaller CT slice thickness increases seed reconstruction accuracy (21, 22), one can expect that it would also decrease interobserver variability. In addition, an indirect effect of slice thickness on interobserver variability exists. The Variseed seed finder algorithm places the seeds at discrete distances in craniocaudal direction, corresponding to the slice positions (Fig. 2). As a subset of reference seeds was manually corrected on sagittal and coronal images, a comparable proportion of observer’s seeds (which were not corrected) showed a small shift in craniocaudal direction with respect to the reference. This deviation decreases with slice thickness. The discrete seed positioning by the seed finder tool explains why the distributions are not typically Gaussian ( p-value calculated with the Shapiro-Wilk normality test were !0.001 for all directions). The high CT peaks are because of the fact that the majority of seeds were localized by the seed finder tool by both reference and observers, which needed no

586

M. De Brabandere et al. / Brachytherapy 12 (2013) 580e588

Fig. 2. Seed positions as localized by the seed finder tool (left) and after corrections for the reference seed set (right).

positional craniocaudal correction (Fig. 1). The deviations, largest in craniocaudal direction, correspond mainly with seeds that formed a seed pair with a corrected reference seed. For T1-based seed reconstruction, 4 mm images were acquired, as they provide a better seed reconstruction accuracy than 3 mm or 5 mm slice thickness according to a previous phantom study (21). This was attributed to the fact that the large MRI seed artifact is visible on multiple slices. Refining the slice thickness creates too many possible seed locations, complicating manual seed reconstruction. Also the T1 seed deviations did not show a typical Gaussian distribution due to the long tails, with ShapiroWilk normality p-values !0.001, both in-plane as in craniocaudal direction. The larger in-plane deviations on MRI than CT were not related to in-plane pixel resolution (which was in fact more accurate for MRI, Table 1), but to the larger seed voids/artifacts, and to the larger reconstruction uncertainty in longitudinal direction. As TPS reconstructs only the seed centers without accounting for rotations, misplacements in longitudinal direction also affect in-plane accuracy. It is expected that our results on CT and MRI seed reconstruction accuracy are independent of seed type. Siebert et al. (22) compared seed reconstruction accuracy of nine different seed models in a phantom study and found no significant influence of seed type. A detailed seed visibility study by Al-Qaisieh et al. showed similar greyscale seed profiles for five different seed types on CT and MR images (except for one seed model that is now not commercially available anymore), suggesting that identification of seed location does not vary with seed type (27). Image timing is not expected to affect seed visibility and reconstruction accuracy. Although early imaging may show more surgical artifacts than postplan imaging at 4 weeks, this will only

have an impact on the postplan dosimetry through target volume contouring. Interobserver variability in dosimetry parameters In a second phase of this study, the effect of the measured seed reconstruction deviations on several dosimetric parameters was calculated. Averaged over the three patients, we found a D90 interobserver variability of 1.5% and 1.3% (1 SDref) for CT and CT þ T2, respectively. These values were similar to the results found by Mangili et al. (28), who performed an interobserver variability study on CT (one patient, seven observers) with a resulting SD of 1.7% (1 SD). Also the V100 values were small and comparable with the results by Mangili et al.: 0.9% and 0.5% (1 SDref) in this patient study for CT and CT þ T2, respectively, vs. 0.4% (1 SD) in the CT patient study. For T1 þ T2 using MRI for seed localization, the D90 interobserver variability was 6.6%, which was notably larger than the uncertainties in the other postplan techniques. The difference is remarkable, given the fact that the MRI sequence was dedicated to seed detection. Unfortunately, we found no other dosimetric interobserver patient studies on MRI seed reconstruction to compare our values to. However, an interesting Monte Carlo study was performed by Lindsay et al. (29), investigating the relationship between reconstruction uncertainties and dosimetric parameters. By simulating random seed displacements of varying magnitude (expressed as 1 SD from a Gaussian distribution), a curve was derived showing the relationship between the seed localization uncertainty and the D90. Visual interpolation of the graph enabled a comparison between these simulated uncertainties and the interobserver variability observed in our patient study. For a localization uncertainty of 1.1 mm, as found in our CT-based interobserver study,

M. De Brabandere et al. / Brachytherapy 12 (2013) 580e588

Lindsay et al. computed a decrease in D90 of about 1.5%. This corresponds well with the dosimetric uncertainty of 1.5% and 1.3% (1 SDref) for CT and CT þ T2, respectively, in our patient study. For a seed localization uncertainty of 3 mm, as found in our T1-based interobserver study, they predicted an average D90 change of about 4.5%, which was lower than the D90 variability of 6.6% for T1 þ T2 in the present study. The fact that on average 7% of the reconstructed seeds were not taken into account in our reconstruction uncertainty calculation (because they were not assigned to a reference seed within acceptable distance) certainly contributes to this difference. Lindsay et al. simulated that if 10% of the seeds were replaced at newly randomized locations, this resulted in a D90 change of 3.8%. Finally, it should be noted that plans with different seed localizations may yield similar dose quantifiers. Dosimetric parameters such as D90, V100, and V150 may mask poor seed localization. Depending on where the seed reconstruction uncertainties occur, this may have varying effect on prostate dosimetry and treatment outcome. Comparison of postplan techniques The obtained D90 values differed substantially for the three techniques (Table 3). The lower D90 values for CT were mainly because of the larger prostate volumes (on average þ17%) compared with T2. The higher D90 on CT þ T2 relative to T1 þ T2 could partially be attributed to a better seed detection on CT and to some extent to a ‘‘reshaping/cutting’’ effect introduced during fusion: when two image sets are registered, the contoured 3D objects are slightly reshaped, usually resulting in smaller objects, and hence larger D90. In a previous study, this effect was shown to be more pronounced in CT þ T2 because of the larger fusion uncertainty (17). It is not possible to determine what the exact postimplant dose values are for these patients. Literature is at present not unequivocal regarding the relationship between dosimetry and clinical outcome, with some groups reporting statistically significant correlations between D90 and PSA (prostate-specific antigen) control rates (30, 31) and others reporting that such correlations could not be validated (32e34). Lee (35) summarized the existing data and concluded that a doseeresponse relationship does exist, with D90 being a better predictor than V100, but that the specifics of the relationship are still problematic as there is no consistency in the reported cutpoints. Most of these studies were performed on CT, and given the substantial uncertainty in CT contouring, the contradictory findings are understandable. To evade this problem, MRI-based solutions have become more prominent in the search for an alternative, more reliable postplan technique. However, these techniques raised other specific uncertainties related to image fusion and seed reconstruction on MRI (17). It is important that these uncertainties are also considered

587

when new postplan techniques are introduced. This is also valid when other approaches are explored involving, for example, ultrasound images (36). It is clear that improving the dosimetric reliability and reproducibility of the postplan technique is a crucial prerequisite to relate dosimetric quantifiers to clinical results. In this study, we have quantified for the first time the uncertainty of MRI-based seed reconstruction on patient images and found it to be slightly larger than the reconstruction uncertainty on CT. Attempts to improve MRI seed visibility by adding protons were never successful because of the small volume of the seeds. Improving seed detection on MRI should therefore focus on further optimization of sequences in combination with automatization of the reconstruction process.

Conclusion In this multicentric interobserver study, we compared the seed reconstruction uncertainty on CT and MRIT1eweighted images by means of an interobserver variability study on patient images. We found that CT seed reconstruction deviations are small, on average 1.1 mm (1 SD) and consistent with the results from previous phantom studies. The effect of the interobserver variations on the D90 in postplan techniques CT and CT þ T2 was less than 2%. MRI-based seed reconstruction was less accurate than CT, with a mean interobserver variation in seed positioning of 3.0 mm (1 SD). This resulted in a non-negligible mean interobserver variation in D90 of about 7% for T1 þ T2. These uncertainties should be taken into account and weighted against other uncertainties such as contouring and image fusion when comparing the overall reliability of postplan techniques.

Acknowledgments The authors are grateful to the members of the GECESTRO (Groupe Europeen de Curietherapie - European SocieTy for Radiotherapy & Oncology) BRAPHYQS (BRAchytherapy PHYsics Quality assurance System) and PROBATE (PROstate BrAchyThErapy) working groups for the useful discussions. References [1] Davis BJ, Horwitz EM, Lee WR, et al. American Brachytherapy Society consensus guidelines for transrectal ultrasound-guided permanent prostate brachytherapy. Brachytherapy 2012;11:6e19. [2] Ash D, Flynn A, Battermann J, et al. ESTRO/EAU/EORTC recommendations on permanent seed implantation for localized prostate cancer. Radiother Oncol 2000;57:315e321. [3] Salembier C, Lavagnini P, Nickers P, et al. Tumour and target volumes in permanent prostate brachytherapy: A supplement to the ESTRO/EAU/EORTC recommendations on prostate brachytherapy. Radiother Oncol 2007;83:3e10.

588

M. De Brabandere et al. / Brachytherapy 12 (2013) 580e588

[4] Brinkmann D, Kline R. Automated seed localization from CT datasets of the prostate. Med Phys 1998;25:1667e1672. [5] Dubois D, Bice W, Prestidge B. CT and MRI derived source localization error in a custom prostate phantom using automated image coregistration. Med Phys 2001;28:2280e2284. [6] Tubic A, Zaccarin L, Beaulieu L, et al. Automated seed detection and three-dimensional reconstruction. II. Reconstruction of permanent prostate implants using simulated annealing. Med Phys 2001;28: 2272e2279. [7] Al-Qaisieh B, Ash D, Bottomley DM, et al. Impact of prostate volume evaluation by different observers on CT-based postimplant dosimetry. Radiother Oncol 2002;62:267e273. [8] Crook J, Milosevic M, Catton P, et al. Interobserver variation in postimplant computed tomography contouring affects quality assessment of prostate brachytherapy. Brachytherapy 2002;1:66e73. [9] Lee WR, Roach M III, Michalski J, et al. Interobserver variability leads to significant differences in quantifiers of prostate implant adequacy. Int J Radiat Oncol Biol Phys 2002;54:457e461. [10] Petrik D, Araujo C, Kim D, et al. Implications of CT imaging for postplan quality assessment in prostate brachytherapy. Brachytherapy 2012;11:435e440. [11] Amdur R, Gladstone D, Leopold K, et al. Prostate seed implant quality assessment using MR and CT image fusion. Int J Radiat Oncol Biol Phys 1999;43:67e72. [12] Crook J, McLean M, Yeung I, et al. MRI-CT fusion to assess postbrachytherapy prostate volume and the effects of prolonged edema on dosimetry following transperineal interstitial permanent prostate brachytherapy. Brachytherapy 2004;3:55e60. [13] McLaughlin P, Narayana V, Kessler M, et al. The use of mutual information in registration of CT and MRI datasets post permanent implant. Brachytherapy 2004;3:61e70. [14] Polo A, Cattani F, Vavassori A, et al. MR and CT image fusion for postimplant analysis in permanent prostate seed implants. Int J Radiat Oncol Biol Phys 2004;60:1572e1579. [15] Katayama N, Takemoto M, Yoshio K, et al. T2*-weighted image/T2weighted image fusion in postimplant dosimetry of prostate brachytherapy. J Radiat Res 2011;52:680e684. [16] Ohashi T, Momma T, Yamashita S, et al. Impact of MRI-based postimplant dosimetric assessment in prostate brachytherapy using contrastenhanced T1-weighted images. Brachytherapy 2012;11:468e475. [17] De Brabandere M, Hoskin P, Haustermans K, et al. Prostate postimplant dosimetry: Interobserver variability in seed localisation, contouring and fusion. Radiother Oncol 2012;104:192e198. [18] Dubois DF, Prestidge BR, Hotchkiss LA, et al. Source localization following permanent transperineal prostate interstitial brachytherapy using magnetic resonance imaging. Int J Radiat Oncol Biol Phys 1997;39:1037e1041. [19] Prete JJ, Prestidge BR, Bice WS, et al. Comparison of MRI- and CTbased post-implant dosimetric analysis of transperineal interstitial permanent prostate brachytherapy. Radiat Oncol Investig 1998;6:90e96. [20] Tanaka O, Hayashi S, Matsuo M, et al. Comparison of MRI-based and CT/MRI fusion-based postimplant dosimetric analysis of

[21]

[22]

[23]

[24]

[25]

[26]

[27]

[28]

[29]

[30]

[31]

[32]

[33]

[34]

[35] [36]

prostate brachytherapy. Int J Radiat Oncol Biol Phys 2006;66: 597e602. De Brabandere M, Kirisits C, Peeters R, et al. Accuracy of seed reconstruction in prostate postplanning studied with a CT- and MRI-compatible phantom. Radiother Oncol 2006;79:190e197. Siebert F, De Brabandere M, Kirisits C, et al. Phantom investigations on CT seed imaging for interstitial brachytherapy. Radiother Oncol 2007;85:316e323. Rivard MJ, Coursey BM, DeWerd LA, et al. Update of AAPM Task Group No. 43 Report: A revised AAPM protocol for brachytherapy dose calculations. Med Phys 2004;31:633e674. McLaughlin PW, Narayana V, Drake DG, et al. Comparison of MRI pulse sequences in defining prostate volume after permanent implantation. Int J Radiat Oncol Biol Phys 2002;54:703e711. Wachowicz K, Thomas SD, Fallone BG. Characterization of the susceptibility artifact around a prostate brachytherapy seed in MRI. Med Phys 2006;33:4459e4467. Lagerburg V, Moerland MA, Seppenwoolde JH, et al. Simulation of the artifact of an iodine seed placed at the needle tip in MRI-guided prostate brachytherapy. Phys Med Biol 2008;53:59e67. Al-Qaisieh B, Smith DW, Brearley E, et al. Comprehensive I-125 multi-seed comparison for prostate brachytherapy: Dosimetry and visibility analysis. Radiother Oncol 2007;84:140e147. Mangili P, Stea L, Cattani F, et al. Comparative study of permanent interstitial prostate brachytherapy post-implant evaluation among seven Italian institutes. Radiother Oncol 2004;71:13e21. Lindsay PE, Van Dyk J, Battista JJ. A systematic study of imaging uncertainties and their impact on I-125 prostate brachytherapy dose evaluation. Med Phys 2003;30:1897e1908. Stock R, Stone N, Tabert A, et al. A dose-response study for I125 prostate implants. Int J Radiat Oncol Biol Phys 1998;41: 101e108. Potters L, Cao Y, Calugaru E, et al. A comprehensive review of CTbased dosimetry parameters and biochemical control in patients treated with permanent prostate brachytherapy. Int J Radiat Oncol Biol Phys 2001;50:605e614. Orio P, Wallner K, Merrick G, et al. Dosimetric parameters as predictive factors for biochemical control in patients with higher risk prostate cancer treated with Pd-103 and supplemental beam radiation. Int J Radiat Oncol Biol Phys 2007;67:342e346. Gastaldi E, Chiono L, Gallo F, et al. Dosimetry doesn’t seem to predict the control of organ-confined prostate cancer after I-125 brachytherapy. Arch Ital Urol Androl 2009;81:215e217. Butler WM, Stewart RR, Merrick GS. Evaluation of radiobiologic biochemical control in a large permanent prostate brachytherapy population from a single institution using AAPM TG-137 parameters. Brachytherapy 2011;10:16e28. Lee WR. Permanent prostate brachytherapy: The significance of postimplant dosimetry. Rev Urol 2004;6:49e56. Haworth A, Ebert M, St Clair S, et al. Impact of selection of postimplant technique on dosimetry parameters for permanent prostate implants. Brachytherapy 2005;4:146e153.