Image fusion of CT and MRI data enables improved target volume definition in 3D-brachytherapy treatment planning

Brachytherapy 2 (2003) 164–171

Image fusion of CT and MRI data enables improved target volume definition in 3D-brachytherapy treatment planning Robert C. Krempien1,*, Sascha Daeuber2, Frank W. Hensley1, Michael Wannenmacher1, Wolfgang Harms1 1

Clinic for Radiology, Department of Clinical Radiology, University of Heidelberg, Heidelberg, Germany Institute for Process Control and Robotics, Department of Computer Science, University of Karlsruhe, Karlsruhe, Germany

2

Abstract

Purpose: To integrate MRI into CT-based 3D-brachytherapy treatment planning using a software system for image registration and fusion. Methods and Materials: Sixteen patients with recurrent head-and-neck cancer, vulvar cancer, liposarcoma, and cervical cancer were treated with interstitial (n ⫽ 12) and endocavitary (n ⫽ 4) brachytherapy. CT and MRI scans were performed after implantation and prior to treatment planning. Image registration to integrate the CT and MR information into a single geometric framework was performed using a software algorithm based on mutual information. Conventional 3D-brachytherapy planning based on CT-information alone was compared to brachytherapy planning based on fused CT and MRI data. The accuracy of the image fusion was measured using predefined corresponding landmarks in the CT and MRI data. Results: The presented automated algorithm proved to be robust and reliable (mean registration error 1.8 mm, range 0.8–4.1 mm, SD 0.9 mm). Tumor visualization was difficult using CT alone in all cases. Brachytherapy treatment planning based on fused CT and MRI data enabled better definition of target volume and risk structures as compared to treatment planning based on CT alone. Conclusions: Image registration and fusion is feasible for afterloading brachytherapy treatment planning. Treatment planning based on fused CT and MRI data resulted in improved target volume and risk structure definition. 쑖 2003 American Brachytherapy Society. All rights reserved.

Keywords:

Brachytherapy; Image fusion; Treatment planning

Introduction Brachytherapy using afterloading techniques has proven to be an effective method in the treatment of primary (1, 2) or recurrent head-and-neck cancers (3, 4), breast cancer (5), prostate cancer (6, 7), and gynecologic cancers (8). Only recently has anatomy-based optimization and evaluation by means of dose-volume histograms, methods well established in external beam radiotherapy planning, been introduced into brachytherapy planning (9, 10). The CT-based reconstruction allows the individual adjustment and optimization of the dose distribution to anatomical target volumes. Received 2 February 2003; received in revised form 20 June 2003; accepted 30 June 2003. This work was funded by a grant from the Deutsche Forschungs Gemeinschaft (DFG), Collaborative Research Center (SFB 414) – Information Technology in Medical Science, Computer- and Sensor-assisted Surgery. * Corresponding author. Clinic for Radiology, Department of Clinical Radiology, University of Heidelberg, INF 400, 69120 Heidelberg, Germany. Tel.: ⫹49-6221-568201; Fax: ⫹49-6221-565353 E-mail address: [email protected] (R. Krempien).

Due to the limited soft tissue contrast, CT still has restrictions in defining target volume and risk structures. Magnetic resonance imaging (MRI) has introduced several additional imaging benefits that may confer an advantage over the use of CT in brachytherapy planning such as improved soft tissue definition (11). However, MRI has not yet seriously challenged CT for brachytherapy treatment planning. The reasons for this include: (1) the presence of intrinsic MR image distortions; (2) the absence of MR signals arising from conventional brachytherapy applicators or guides; and (3) the lack of reliable dummy source position markers for MR imaging (11, 12). To use both CT and MRI information it is necessary to include the information from the different sources into a single geometric framework (13), a task generally referred to as image registration. Image fusion then enables the transfer of information from one image study to another thus potentially adding the advantages of both imaging modalities for therapy planning (11, 14). The purpose of this work was to integrate MRI into 3D brachytherapy treatment planning using a custom made software system (15) for image registration and fusion.

1538-4721/03/$ – see front matter 쑖 2003 American Brachytherapy Society. All rights reserved. doi:10.1016/S1538-4721(03)00133-8

Liposarcoma of the thigh

Vulvar SCC

13–14

15–16

Microscopic positive tumor margin to the sciatic nerve Recurrent disease previously irradiated (50 Gy, 74 Gy), multiple operations

SCC of the cervix uteri 9–12

SCC: Squamous cell carcinoma; PDR: pulsed-dose-rate; HDR: high-dose-rate. In all patients the tumor visualization and target volume definition was difficult or not possible in CT. Evaluation of the tumor volume and extent and the target volume was possible in all patients based on MRI information. CT-based brachytherapy-planning alone would have resulted in suboptimal target volume determination.

15 Gy (PDR) 1 6, 8 Interstitial

15 Gy (PDR) 1 4 Interstitial

22.5 Gy (HDR) 3 2 Cervix Applicator

1 7–14 SCC of the floor of the mouth/tongue 1–8

Table 1 Data of the first 15 Patients

Image fusion Image fusion was performed using custom-made software based on mutual information. Briefly, the presented automatic registration algorithm uses a two-step registration. First, a voxel based method using a probabilistic approach for the detection of the initial transformation and second, an approach based on voxel similarity measures using fast feature space calculations with an evaluation function based on mutual entropy (15, 19).

Interstitial

Tumor Patient No.

Comments

CT-imaging

MRI-imaging

Image acquisition was performed immediately after implantation and prior to treatment planning. Both image modalities were acquired within 45 min. All CT-studies were conducted with a helical CT scanner (PQ-CT, Marconi Medical Systems, Cleveland, OH). Thickness and index were 3 mm in the head and neck area and 5 mm in the other locations. All MR studies were conducted with a 0.23 Tesla open low field MR system (MRI Outlook, Marconi Medical Systems, Finland). MR-Protocol: T1-weighted spin echoes, imaging parallel to the CT-scans with and without Gadolinium (Magnevist, Schering, Berlin, Germany 0.25ml/kg) (TE 27, TR 440, imaging matrix 256 ⫻ 256, thickness equal to the CT); and T2-weighted spin echoes (TE 100, TR 2200, imaging matrix 256 ⫻ 256, thickness equal to CT). The geometric distortions of the MRI-facilities were investigated by a special phantom using spin-echo sequences (16). Within 10 cm around the coil axis distortions were below 1 mm. Furthermore, the phantom studies offered the possibility to correct for distortions if necessary (17). In patients with cervical cancer, differences in bladder and rectal filling can change the position of the cervix (18). To minimize differences in bladder and rectal filling both imaging modalities were acquired within 45 min. To ensure constant bladder filling a bladder catheter was used. No significant changes in rectal filling were seen in the observed patients.

Excellent tumor visualization Target structure (sciatic nerve) high visibility Good tumor visualization, excellent delineation from scar tissue and edema

Image acquisition

Good tumor visualization

Method

No. of guides

No. of sessions

From 1999–2002 sixteen patients (Table 1) were treated with either PDR or HDR brachytherapy. Twelve patients (recurrent head-and-neck cancer n ⫽ 8, liposarcoma n ⫽ 2, and vulvar cancer n ⫽ 2) were treated with interstitial implants (PDR) while 4 patients with cervical cancer were treated with endocavitary brachytherapy (HDR). Conventional 3D-brachytherapy planning using CT-information alone was compared to brachytherapy planning based on fused CT-MRI data. Gross tumor volume (GTV) and clinical target volume (CTV) were determined. In all cases a geometrical dose optimization was used to generate the treatment plan. Brachytherapy planning was performed with the Plato (Nucletron, Veenendaal, The Netherlands) planning system.

Tumor not visible due to tooth metal artifacts and tissue changes (edema, scar) Tumor extent difficult to determine Target structure (sciatic nerve) not visible Tumor difficult to determine due to tissue changes (edema, scars)

Patients

165

Recurrent disease, previously operated and irradiated (mean dose 59.5 Gy, range 50–66 Gy) FIGO IIb–IIIb

Total dose brachytherapy

Methods and Materials

15 Gy Range 13–17 (PDR)

Krempien et al. / Brachytherapy 2 (2003) 164–171

166

Krempien et al. / Brachytherapy 2 (2003) 164–171

Registration The information regarding the geometry of the two modalities was read out of the DICOM headers to calculate a metrically correct three-dimensional cubic datamodel. For registration of the two modalities a matrix was calculated describing the necessary translations, rotations, and scale transformations to combine the images in the same coordinate framework. The automatic registration took about 1–3 min for two data sets of about 60 slices. The images were then superimposed and visualized within the same image slice. This allowed for the interaction of comprehensive anatomical mapping with functional distribution of the tissues within a single image set. Fusion and validation Image fusion enabled the transfer of information from one image study to another. Within the superimposed images, interactive segmentation of tumor volume and risk structures was performed. The fusion function was also used for validation. Since there was no predefined correct object to compare against, the overall fitness of the obtained matching was visualized transparently and the fused images were used to validate the matching accuracy comparing different anatomical structures, i.e., bone and skin contour (Fig. 1). For quality control of the fusion process five corresponding landmarks were predefined in the CT and MRI data before registration. After registration the average root mean square distance between corresponding landmarks was measured.

The predefined corresponding landmarks used for quality control comparison were bony structures. In head and neck patients landmarks used were condylar process of the mandible, tip of the mastoid process and the mental spine of the mandible body. Typical landmarks in patients with cervical or vulvic carcinoma were minor trochanter, pubic symphysis, inferior pubic ramus, anterior portion of the iliac tuberosity. In the sarcoma patients, femoral bone was used as landmark. Since definition of landmarks in this region is difficult, the whole femoral bone was segmented in both imaging modalities and used for control of registration accuracy. Surgical clips were not used as landmarks since these clips are not clearly visualized on MRI and are subject to susceptibility artifacts with spatial distortion that extends up to several mm (20).

Results Brachytherapy planning based on fused CT and MRI data was feasible in all patients. Tumor visualization was difficult using CT alone in all cases. Due to the superior soft tissue contrast, target volume and most risk structures were better defined by MR imaging (Table 1). Image registration and fusion enabled visualization of CT and MRI information in a single geometric framework (Fig. 1). The source position dummy markers were segmented in the CT, since conventional dummy markers are difficult to localize in MRI (Fig. 2). Structures of interest were segmented interactively

Fig. 1. Image fusion of CT and MRI images. As seen in the above image bony structures, skin contour, muscular and fatty tissues fit very well visualizing the good registration process.

Krempien et al. / Brachytherapy 2 (2003) 164–171

167

Fig. 2. Patient 13 with sarcoma in the thigh. (A) CT accurately localizes source positions and surgical clip marker of the tumor bed (arrows) but due to the limited soft tissue contrast and the postoperative changes the nerve, as the primary target structure was hardly detectable. (B) The nerve is clearly visible in the MRI data (arrow).

using both imaging modalities (Fig. 3). Compared to brachytherapy treatment planning based on CT alone, treatment planning based on fused CT and MRI data enabled improved dose distribution to target volumes and structures at risk in all 16 cases (Table 2). With a mean registration error of 1.8 mm (range 0.8–4.1 mm, SD 0.9 mm) the used

image fusion algorithm proved to be robust and reliable. Head-and-neck cancer patients registered more efficiently and accurately than pelvis and sarcoma patients. Accuracy in head and neck patients was in the mean 1.3 cm (range 0.8– 3.4 cm) compared with 1.7 cm (range 1.1–3.7 cm) in pelvis patients and 3.9 cm (range 3.7–4.1 cm) in sarcoma patients.

168

Krempien et al. / Brachytherapy 2 (2003) 164–171

Fig. 3. (a) Patient 13 with sarcoma in the thigh. 3D reconstruction of interactive segmentation of volumes of interest using the fused CT-MRI data. (b) Brachytherapy treatment planning based on fused data. Comparison of sciatic nerve localization based on CT (gray arrow) or MRI (white arrow). Dose prescription to the 100 cGy isodose/pulse/hour (white line).

Oropharyngeal cancer Planning target volume (PTV) definition was defined based on 3D MRI-CT information and included the visible tumor with a safety margin of 10 mm (21). In 8 cases with recurrent oropharyngeal cancer, tooth metal artifacts, scar tissue, and tissue edema due to previous operations and irradiation hampered the tumor definition in the CT. Precise target volume definition based on the CT information alone was not possible. Only the combination with the MRI information resulted in correct target volume definition and therefore enabled an improved coverage of the tumor volume. All patients with oropharyngeal cancer have previously been irradiated with a mean dose of 59.5 Gy (range 50–66 Gy). Thus, the applied dose to risk organs like the vascular structures or the mandibular bone was very important. For the dose distribution to the main vascular structures CT and MRI information proved to be nearly equal. For the definition of the dose to the mandibular bone, CT information proved to be essential due to the better visualization of bony structures. In one case, planning based on MRI alone would have resulted in a dose prescription nearly up to the 200%-isodose on parts of the mandibular bone. Only the combination of both imaging modalities led to optimal 3D planning results (Table 2). Sarcoma Both patients with sarcoma of the thigh had a positive margin after surgical resection near or adjacent to the sciatic nerve. Complete resection would have resulted in a resection

of the nerve. Therefore, brachytherapy-guides and surgical clips were implanted in the tumor bed for postoperative boost irradiation. Source dummy marker and surgical clips marking the tumor bed were segmented in the postoperative planning CT but the nerve was not visible due to the postoperative changes (Fig. 2). In the MRI the nerve could be located easily and precise segmentation was possible (Figs. 2 and 3). With the help of the fused images the nerve and its relationship to the source dwell positions and the surgical clips referring to the tumor bed could be determined in one image set (Fig. 3A). Based on the synthetic images a new dose prescription was applied (Fig. 3B). The usual brachytherapy boost dose in cases with histologically positive tumor margins ranges between 12 and 20 Gy (22). It is known from intraoperative radiation therapy that an electron boost exceeding 15 Gy results in an increased risk of radiation damage to nervous structures (22, 23). The nerve received a dose of about 40 Gy (patient 13) and 25 Gy (patient 14), respectively, based on CT-imaging alone. The additional information from the MRI enabled an optimized dose prescription of 18 Gy and 14 Gy, respectively, to the sciatic nerve (Table 2, Fig. 3). Gynecological cancers In both cases with recurrent vulvic carcinoma, CT was hampered by scar tissue and tissue edema due to previous operations and irradiation. In all cases with cervical cancer, MRI was superior in tumor visualization as compared with CT. The use of fused images enabled an improved 3D-dose

Krempien et al. / Brachytherapy 2 (2003) 164–171

169

Fig. 3. Continued

distribution in the target volume and resulted in a reduced dose to the intestine, the bladder, and the anterior rectal wall (Table 2). Discussion This study was conducted to assess the benefits of integration of superior soft tissue imaging provided by MRI in afterloading brachytherapy treatment planning using image fusion software. Image fusion was performed by an automatic image fusion algorithm based on mutual information. The advent of cross-sectional image-based planning in brachytherapy using afterloading techniques has introduced

the possibility of individually adjusting and optimizing the dose description to anatomical structures (9, 10). CT is normally bound to axial scanning and offers limited soft tissue contrast, therefore, it is sometimes difficult to determine the exact tumor location and spread (11). This may lead to early tumor relapse or severe side effects (11, 24). The imaging quality and versatility of MRI is an ideal additional tool for assimilating the necessary information for brachytherapy planning. One method to use imaging benefits of CT and MRI in radiation therapy is to integrate data from both imaging modalities into a single geometric framework (11), a task generally referred to as image registration. Manual methods of registering different image sets can be inefficient and error prone (25). Automated approaches

170

Krempien et al. / Brachytherapy 2 (2003) 164–171

Table 2 Dose to risk organs based on different imaging modalities

Patient No.

Target region

Planed dose in target volume

1–8

Floor of the mouth/tongue

15 Gy, range 13–17 Gy (PDR)

9–12

Cervix uteri

22.5 Gy (HDR)

13–14 15–16

Thigh Vulvic region

15 Gy (PDR) 15 Gy (PDR)

Risk structures

Anticipated dose max. in risk organs

CT

MRI

Fused images

Mandibular bone

15 Gy

12.3 Gy (9–14)

14.6 Gy (10–22)

13.6 Gy (10–16)

Neurovascular structures Intestine Bladder Rectum Sciatic Nerve Uretral orifice Anterior rectal wall

15 Gy

11.3 Gy (7–14)

12.6 Gy (9–15)

12.4 Gy (8–15)

15 15 15 15 10 10

17.5 Gy (14–22) 17 Gy (15–20) 19.3 Gy (16–23) 32.5 Gy (25–40) 14.5 Gy (13–16) 16 Gy (12–20)

13,3 14.5 15.5 15.5 10.5 10.5

13.8 Gy (12–15) 15.3 Gy (14–16) 15.5 Gy (12–18) 16 Gy (14–18) 10 Gy 10 Gy

Gy Gy Gy Gy Gy Gy

Max. dose at risk structures based on

Gy Gy Gy Gy Gy Gy

(11–15) (13–16) (12–18) (13–18) (11–10) (8–11)

Dose distribution to risk organs based on CT, MRI, and fused images. While MRI had advantages in defining soft tissue risk organs, CT was superior in imaging of bony structures, i.e., the mandibular bone. Only the combination of both imaging modalities led to an optimized dose distribution to all risk organs in image based 3D-brachytherapy planning.

will reduce inter-operator variability and allow more accurate registration between multimodality systems (26). Although images from different modalities exhibit complementary information there is usually a high degree of shared information between images of the same structures. Most methods identify anatomical homologies between the two image studies by using relationships between voxel intensity values and attempt to find an interscan coordinate transformation that matches the homologous structures (27). The presented registration algorithm uses a two-step registration, first a voxel-based method using a probabilistic approach for the detection of the initial transformation and second, an approach based on voxel similarity measures using fast feature space calculations with an evaluation function based on mutual entropy (14). Accuracy of image registration is still challenging and need to be clinically validated. Image fusion even by mutual information remains a challenge since problems due to patient re-positioning, slice thickness, soft tissue mobility, and cropping to internal anatomy versus external anatomy may compromise the registration process (28). One of the main problems with automated registration techniques is that it is diffcult to quantify errors. Mutual information is a measure of how one image explains the other. It makes no assumption of the functional form or relationship between image intensities in the two images. Therefore, a quality control process is required to estimate the accuracy of the registration process. For quality control of the fusion process, five corresponding landmarks were predefined in the CT and MRI data before registration. After registration the average root mean square distance between corresponding landmarks was measured. Over the years many studies have been conducted to integrate the superior soft tissue contrast of MRI into radiation therapy (11). Although image registration and fusion has become nearly a standard technique in many aspects of image-guided surgery or stereotactic irradiation (13, 14), there are few reports of using image fusion in brachytherapy.

Image fusion of CT and MRI data has been used in assessing the quality of seed implants in prostate cancer (29, 30) and brain tumors (31). In brachytherapy applications, MRI has proven to be beneficial in image-guided needle application (32, 33) especially due to better target volume definition. To our knowledge there are no reports using CT and MRI data image fusion to improve afterloading 3D-brachytherapy treatment planning. In this study brachytherapy treatment planning based on fused CT and MRI data enabled in all presented cases improved target definition as compared with treatment planning based on CT alone. With a mean registration error of 1.8 mm (range 0.8–4.1 mm, SD 0.9 mm) the presented automated algorithm based on mutual information proved to be robust and reliable. These data of the first 16 patients show that image registration and fusion is feasible for afterloading brachytherapy treatment planning. Treatment planning based on fused CT and MRI data resulted in improved target volume and risk structure definition.

References [1] Marsiglia H, Haie-Meder C, Sasso G, et al. Brachytherapy for T1-T2 floor-of-the-mouth cancers: The Gustave-Roussy Institute experience. Int J Radiat Oncol Biol Phys 2002;52:1257–1263. [2] Leborgne F, Leborgne JH, Zubizarreta E, et al. Cesium-137 needle brachytherapy boosts after external beam irradiation for locally advanced carcinoma of the tongue and floor of the mouth. Brachytherapy 2002;1:126–130. [3] Geiger M, Strnad V, Lotter M, et al. Pulsed-dose rate brachytherapy with concomitant chemotherapy and interstitial hyperthermia in patients with recurrent head and neck cancer. Brachytherapy 2002; 1:149–153. [4] Shasha D, Harrison LB, Chiu-Tsao ST. The role of brachytherapy in head-and-neck cancer. Semin Radiat Oncol 1998;8:270–281. [5] Harms W, Krempien R, Hensley FW, et al. Five-year results of pulseddose-rate Brachytherapy applied as a boost after breast-conserving therapy in patients at high risk for local recurrence from breast cancer. Strahlenther Onkol 2002;178:607–614. [6] Zietman AL. Localized prostate cancer: Brachytherapy. Curr Treat Options Oncol 2002;3:429–436.

Krempien et al. / Brachytherapy 2 (2003) 164–171 [7] Kestin LL, Martinez AA, Stromberg JS, et al. Matched-pair analysis of conformal high–dose-rate brachytherapy boost versus externalbeam radiation therapy alone for locally advanced prostate cancer. J Clin Oncol 2000;18:2869–2880. [8] Hsu IC, Speight J, Hai J, et al. A comparison between tandem and ovoids and interstitial gynecologic template brachytherapy dosimetry using a hypothetical computer model. Int J Radiat Oncol Biol Phys 2002;52:538–543. [9] Kolotas C, Baltas D, Zamboglou N. CT-based interstitial HDR brachytherapy. Strahlenther Onkol 1999;175:419–427. [10] Willins J, Wallner K. CT-based dosimetry for transperineal I-125 prostate brachytherapy. Int J Radiat Oncol Biol Phys 1997;39:347– 353. [11] Khoo VS, Dearnaley DP, Finnigan DJ, et al. Magnetic resonance imaging (MRI): Considerations and applications in radiotherapy treatment planning. Radiother Oncol 1997;42:1–15. [12] Wesby G, Adamis MK, Edelmann RR. Artifacts in MRI: description, causes, and solutions. In: Edelmann RR, Hesselink JK, Zlatkin MB, editors. Clinical magnetic resonance imaging. Philadelphia: Saunders; 1996. p. 88–144. [13] Hill DLG, Hawkes DJ, Gleeson MJ, et al. Accurate frameless registration of MR and CT images of the head: Applications in planning surgery and radiation therapy. Radiology 1994;191:447–454. [14] Pelizzari CA. Image processing in stereotactic planning: Volume visualization and image registration. Medical Dosimetry 1998;23:137– 145. [15] Pokrandt P. Fast non-supervised matching: A probabilistic approach. In: Lemke HU, Vannier MV, Inamura K, Farman AG, editors. Proceedings of the Tenth International Symposium and Exhibition on Computer Assisted Radiology and Surgery (CAR ’96); 1996. [16] Schubert K, Wenz F, Krempien R, et al. Possibilities of an open magnetic resonance scanner integration in therapy simulation and three-dimensional radiotherapy planning. Strahlenther Onkol 1999; 175:225–231. [17] Krempien RC, Schubert K, Zierhut D, et al. Open low field magnetic resonance imaging (MRI) in radiation therapy treatment planning. Int J Radiat Oncol Biol Phys 2002;53:1350–1360. [18] Buchali A, Koswig S, Dinges S, et al. Impact of the filling status of the bladder and rectum on their integral dose distribution and the movement of the uterus in the treatment planning of gynecological cancer. Radiother Oncol 1999;52:29–34. [19] Krempien R, Grabowski HA, Harms W, et al. Image fusion in radiotherapy. In: W. Schlegel, T. Bortfeld, editors. The use of computers in radiation therapy. Heidelberg: Springer-Verlag; 2000. p. 93–95.

171

[20] Weishaupt D, Quick HH, Nanz D, et al. Ligating clips for threedimensional MR angiography at 1.5T: In vitro evaluation. Radiology 2000;214:902–907. [21] Nag S, Cano ER, Demanes DJ, et al. The American Brachytherapy Society recommendations for high–dose-rate brachytherapy for headand-neck carcinoma. Int J Radiat Oncol Biol Phys 2001;50:1190– 1198. [22] Shaw EG, Gunderson LL, Martin JK, et al. Peripheral nerve and ureteral tolerance to intraoperative radiation therapy: Clinical and dose– response analysis. Radiother Oncol 1990;18:247–255. [23] Johnstone PAS, DeLuca AM, Bacher JD, et al. Clinical toxicity of peripheral nerve to intraoperative radiotherapy in a canine model. Int J Radiat Oncol Biol Phys 1995;32:1031–1034. [24] Terahara A, Nakano T, Ishikawa A, et al. Dose-volume histogram analysis of high–dose-rate intracavitary brachytherapy for uterine cervix cancer. Int J Radiat Oncol Biol Phys 1996;35:549–554. [25] De Salles AWT, Abe M, Kjellberg RN. Transposition of target information from magnetic resonance and computed tomography scan images to conventional stereotactic space. Appl Neurophysiol 1987;50:23–32. [26] van den Elsen PA, Pol EJ, Viergever MA. Medical image matching—A review with classification. IEEE Trans Med Biomed Eng 1993; 12:26–39. [27] Little JA, Hawkes DJ. The registration of multiple images acquired from a single subject: Why, how, what next? Statistical Methods in Medical Research 1997;6:239–265. [28] Hawkes DJ. Algorithms for radiological image registration and their clinical application. J Anat 1998;193:347–361. [29] Amdur RJ, Gladstone D, Leopold KA, et al. Prostate seed implant quality assessment using MR and CT image fusion. Int J Radiat Oncol Biol Phys 1999;43:67–72. [30] Dubois DF, Bice WS Jr, Prestige BR. CT and MRI derived source localization error in a custom prostate phantom using automated image coregistration. Med Phys 2001;28:2280–2284. [31] Julow J, Major T, Emri M, et al. The application of image fusion in stereotactic brachytherapy of brain tumours. Acta Neurochir 2000; 142:1253–1258. [32] D’Amico AV, Cormack R, Tempany CM, et al. Real-time magnetic resonance image-guided interstitial brachytherapy in the treatment of select patients with clinically localized prostate cancer. Int J Radiat Oncol Biol Phys 1998;42:507–515. [33] Popowski Y, Hiltbrand E, Joliat D, et al. Open magnetic resonance imaging using titanium-zirconium needles: Improved accuracy for interstitial brachytherapy implants? Int J Radiat Oncol Biol Phys 2000;47:759–765.