Improving fiducial and prostate capsule visualization for radiotherapy planning using MRI

Abstract Background and purpose Intraprostatic fiducial markers (FM) improve the accuracy of radiotherapy (RT) delivery. Here we assess geometric integrity and contouring consistency using a T2*‐weighted (T2*W) sequence alone, which allows visualization of the FM. Material and methods Ten patients scanned within the Prostate Advances in Comparative Evidence (PACE) trial (NCT01584258) had prostate images acquired with computed tomography (CT) and Magnetic Resonance (MR) Imaging: T2‐weighted (T2W) and T2*W sequences. The prostate was contoured independently on each imaging dataset by three clinicians. Interobserver variability was assessed using comparison indices with Monaco ADMIRE (research version 2.0, Elekta AB) and examined for statistical differences between imaging sets. CT and MR images of two test objects were acquired to assess geometric distortion and accuracy of marker positioning. The first was a linear test object comprising straight tubes in three orthogonal directions, the second was a smaller test object with markers suspended in gel. Results Interobserver variability for prostate contouring was lower for both T2W and T2*W compared to CT, this was statistically significant when comparing CT and T2*W images. All markers are visible in T2*W images with 29/30 correctly identified, only 3/30 are visible in T2W images. Assessment of geometric distortion revealed in‐plane displacements were under 0.375 mm in MRI, and through plane displacements could not be detected. The signal loss in the MR images is symmetric in relation to the true marker position shown in CT images. Conclusion Prostate T2*W images are geometrically accurate, and yield consistent prostate contours. This single sequence can be used to identify FM and for prostate delineation in a mixed MR‐CT workflow.


| INTRODUCTION
Accurate co-registration of magnetic resonance (MR) and computed tomography (CT) images is essential in radiotherapy (RT) planning using both modalities. MR-CT fusion combines the superior soft tissue contrast of MR images and the electron density from CT images, which is currently required for planning. 1 However, CT and MR examinations take place at different times and over different timescales; the acquisition of detailed MR images covering the tumor volume may require a few minutes, while CT is considerably faster. Physiological motion may thus affect MR and CT images differently, and this is detrimental to the accuracy of MR-CT fusion. In addition, inter-and intra-fraction motion may be significant at the time of RT delivery, introducing further errors. 2,3 In order to mitigate this, fiducial markers can be placed into relatively mobile tumors (or their vicinity), enabling more precise image co-registration to be performed for MR-CT fusion during the planning process 4 and position verification prior to each fraction. 5,6 A more accurate MR-CT co-registration will enable better targeting, therefore markers must be visible, both in MR and CT.
Metallic markers appear bright on CT, often surrounded by reconstruction and beam hardening artifacts, 7,8 but do not yield MR signals and are seen as dark "void" areas on MR. Their susceptibility cause variations in the magnetic field in their vicinity, and they are often better visualized in T2*-weighted (T2*W) images where the signal loss around the markers is emphasized. 9 The design of MR protocols for RT planning thus requires not only geometric accuracy but also that the markers are clearly visible and the image contrast provides confidence in target outlining. Uncertainties and variation in target delineation during RT planning adds a further systematic error.
MRI allows a reduction in interobserver variability for prostate contours compared to CT, 10 however, this is dependent on the sequence used. 11 Previously it has not been possible to provide one single sequence that enables both visualization of the markers and target outlining, and this adds a degree of complexity to the RT planning workflow.
This work investigates a sequence suitable for MR-CT fusion for prostate RT using fiducial markers; in our institution, a set of three gold seeds is implanted in each patient. The MR protocol we implemented consists of two sequences; one standard T2-weighted (T2W) sequence used in diagnostic prostate scans, thus optimized for visualization of intra-prostatic structures, and a second T2*W sequence optimized for marker visualization using the combination of several gradient-echoes with different echo-times (TE) which follow each excitation. The second sequence maximizes visualization of the markers for RT planning fusion.
Studies so far for similar sequences have focused on accuracy of fiducial detection. [12][13][14][15][16][17] In this article we examine the T2*W sequence and investigate whether it is possible to use this sequence alone in prostate studies, considering geometric integrity, the ability to locate marker positions and the ability to provide enough contrast for prostate volume outlining.

2.A | Patient population
Patients were scanned at 1.5 T (Siemens Aera, Erlangen, Germany) as part of the Prostate Advances in Comparative Evidence (PACE) trial (NCT01584258). PACE A randomizes patients between prostatectomy and stereotactic body radiotherapy (SBRT) to a dose of

2.B | Planning CT acquisition
At the Royal Marsden Hospital, all patients receiving RT in PACE have a RT planning CT followed, on the same day, by a planning MRI scan. Patients are scanned with bladder filling and rectal preparation as per institutional guidelines and no intravenous contrast is used. Patients receive 2 days of rectal preparation with enemas prior to planning, and an enema just before their planning CT scan. The CT scan incorporates axial slices of 1.5 mm from mid lumbar spine to below the obturator foramen.

2.C | Planning MRI acquisition
Prostate MRI examinations were undertaken with two two-dimensional (2D) sequences, covering the prostate volume in 28 adjacent slices (2.5 mm thickness). The first one is a standard T2W pulse sequence used in diagnostic MRI of the prostate. This sequence is based on fast spin-echoes and allows visualization of internal structure of the prostate (central and peripheral zone and urethra). The second sequence is applied to the same locations, but it is gradientecho-based and maximizes the signal loss surrounding the markers.
For that purpose, we employed a sequence, which combines several gradient-echo signals, with a range of echo-times (TE), into one single image. This strategy maintains the signal-to-noise ratio in T2*W acquisitions and has been used for other clinical applications. 18,19 Both sequences cover the same volume, centered on the prostate and including at least part of the pelvic bones. Both sequences use the same shimming volume to optimize the magnetic field homogeneity and the manufacturer's own distortion correction software (in 2D). Parameters of both sequences are provided in Table 1.

2.D | Geometric integrity
The field inhomogeneity of the main magnet and the non-uniformity of gradient fields are known to progressively affect the MR images as the distance from the magnet isocenter increases.
Although it is unlikely that the local MR-CT co-registration could be affected by geometric distortion at the prostate location, close to the isocenter, we characterized the hardware-related geometric distortion over the imaging volume. For that purpose we acquired CT and MR images of a previously described test object consisting of straight tubes in three orthogonal directions, known as "Linear Test Object." 20 Images were co-registered and evaluated using the three-dimensional (3D) slicer software package (www.slicer.org). 21 Displacements of test object structures between CT and MR images can be easily detected if they reach half of the voxel size a level of accuracy that is sufficient for the purposes of this study.
In addition a second test object was built by suspending the markers in a gel volume comparable with a prostate (porcine gel, Sigma-Aldricht, St. Louis, MI, 100 g/L, approximately 90 cm 3 ) to verify whether the position of the markers is correctly depicted in the MR images with the sequences used. This step is necessary because the markers themselves disturb the field inhomogeneity, and the associated signal loss is not necessarily symmetric in relation to the true marker position. 22 Therefore, in marker-based registration, it is important to verify that systematic errors are not being introduced.
The markers were orientated approximately in the superior/inferior direction, which most closely resembles their orientation in clinical examinations (Fig. 1). However, the object was rotated by 90°for a second MR acquisition, to evaluate how the susceptibility-related signal loss depends on orientation, and also scanned at different orientations. In order to verify whether systematic errors were introduced, two CT-MR registrations were produced. The first gold standard registration employs the outline of the test object volume, visible in MR and CT. The second registration employs only the marker information, and registration coordinates are compared. In addition, a capsule of cod liver oil was placed on top of the test object to provide a standard for displacements associated with chemical shift. The fat-water chemical shift is known to be 3.5 ppm (225 Hz at 1.5 T), and fatwater displacement was measured by using a readout gradient reversal. 23 2.E | Clinical studies 2.E.1 | Patient population

2.E.2 | Visibility of fiducials
Without reference to the CT images, T2W and T2*W images were reviewed to assess the number of fiducial markers visible.  Considering the test object with markers suspended in gel, the markers are always clearly visible in T2*W images; in T2W images the signal loss is much smaller, as expected (Fig. 4). MR and CT images were co-registered and displacements were shown to be smaller than half pixel size. The signal loss in MR images was thus shown to be symmetric in relation to the true marker position shown in CT images. For both sequences the displacement of fat signals in relation to water signals due to chemical shift was confirmed to be less than 1 mm, as expected.    and size but were seen in eight out of the ten patients, an example is seen in Fig. 6(b).

3.B.2 | Contour variability
Image review shows that the prostate has a high contrast appearance in relation to the surrounding tissues in T2*W images, and internal structures are not demonstrated as clearly as in T2W sequences. Summary of the comparison metrics for all ten patients for each imaging modality is seen in Table 2.
There is good agreement between the three observers for all imaging modalities. Distance measurements between contours were greater and overlap indices lower for CT compared to both MR sequences, indicating a poorer interobserver variability for CT imaging compared to MRI. This was statistically significant when comparing CT with T2*W, as indicated in Table 2.

| DISCUSSION
Test object images demonstrated that prostate MR images are not significantly distorted, and that the T2*W sequence produces a signal void that is symmetric in relation to the true marker position.
This indicates that the signal loss is sufficiently large to obscure the volume immediately adjacent to the seeds where significant image distortion could otherwise be detected. 15  There is a high agreement for prostate contouring on all image sets, likely to reflect the high level of experience of all clinicians, from the same institution and familiar with using MRI for contouring.
The higher agreement for contours on MRI compared to CT is consistent with previous studies as a result of the improved soft tissue contrast with MRI. 28,29 Despite the visual appearance of a more defined prostate capsule on the T2*W sequence, there was no significant difference in interobserver variability when compared to T2W imaging, which again may reflect the users' experience with MR sequences. For this group of observers, the T2*W sequence is similar to standard T2W imaging, but with the added benefit of fiducial identification.
The more recent development of MR-guided RT allows the use of continuous MRI during treatment for motion monitoring and gating. 30 Ultimately the aim would be for an MR-only workflow 31  The accuracy of fiducial detection is paramount and can be either manual 12 or automatic. [13][14][15][16][17] However, ultimately, this must be performed automatically, especially if intrafractional imaging is to be used. Different methods have been described for automatic algorithms including feature extraction 13,15 and template matching. 14,16,17 The fiducial detection is dependent on the signal loss, which varies with factors including seed orientation and TE. 22,27 We demonstrated that calcifications in prostate are a common source of signal voids in T2*W images, and they have been shown to mimic fiducial voids. 32 Although Gustafsson et al. 15 proposed to detect fiducials automatically by considering images at different TEs and the progressive increase in signal loss in multiple-echo pulse sequences, it is unclear whether calcifications will be a significant confounding factor. Further investigation is required to determine whether false positive detection as a result of calcifications is a significant issue and whether calcifications can contribute towards MR-CT co-registration. 32 The full potential of artificial intelligence techniques in fiducial detection has not yet been realized. 33 With progressively more targeted treatment delivery, the accuracy of delineation becomes even more essential. 34 For the prostate, this requires adequate tissue contrast of the capsule to improve confidence in contouring and reduce inter-observer variability. With the development of prostate motion monitoring in MR-guided RT, the prostate contour can be used for gated treatment. 35 This requires easy and accurate identification of the target either visually or using automated algorithms. The latter may either rely on registration of images and propagation of contours or de novo auto-delineation of the prostate on new images. [36][37][38] The sequence described here would therefore be an attractive solution for detailing seeds and the prostate capsule. Further work of significance to MR-guided RT, will be assessment of prostate contouring by treatment radiographers 39 and auto-contouring software on the sequences used here.

| CONCLUSION
We have described here a single T2*W MR sequence suitable for fiducial depiction and prostate contouring. These MR images were demonstrated to be geometrically accurate, the MR signal loss surrounding the fiducial was shown to be symmetric in relation to the true marker position shown in CT and all markers are visible. Prostate contours on MR are more consistent than CT-based contours with good agreement between prostate RT clinicians. We expect T A B L E 2 Summary of the median comparison metrics for three observers contouring all ten patients for each imaging type (with interquartile range in brackets). * Denotes a statistically significant difference when compared to T2*W using a significance level of P = 0.0167 (Bonferroni correction).