Metal artefacts in MRI-guided brachytherapy of cervical cancer.

The importance of assessing the metal-induced artefacts in magnetic resonance imaging (MRI)-guided brachytherapy is growing along with the increasing interest of integrating MRI into the treatment procedure of cervical cancer. Examples of metal objects in use include intracavitary cervical applicators and interstitial needles. The induced artefacts increase the uncertainties in the clinical workflow and can be a potential obstacle for the accurate delivery of the treatment. Overcoming this problem necessitates a good understanding of its originating sources. Several efforts are recorded in the literature to quantify the extent of such artefacts, in phantoms and in clinical practice. Here, we elaborate on the origin of metal-induced artefacts in the light of brachytherapy applications, while summarizing recent efforts that have been made to assess and overcome the induced distortions.

Purpose

In the last two decades, magnetic resonance imaging (MRI) has become an indispensable tool used to diagnose, treat and monitor gynecological cancers [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]. The increasing access to MRI in radiation oncology departments, and its superior soft tissue contrast have favored it over other imaging modalities that have dominated cancer imaging for many years, such as ultrasound (US) and computed tomography (CT). The significant improvement in target/organ delineation offered by volumetric MR images can lead to improved treatment planning and delivery, better local control, and recently reported, higher survival rates [7, 8, 9, 10, 11, 14, 16, 17]. However, MRI also comes with its own technical and logistical challenges such as increased cost, sophisticated patient care workflow, geometric imperfections, patient-related distortions, and its restrictive compatibility with magnetic materials [1, 4, 14, 18, 19, 20, 21, 22]. The latter concern is particularly important for MRI-guided brachytherapy in which metallic objects, such as cervical applicators and titanium needles are used. Metal objects induce artefacts in the acquired MR images, which can degrade the quality of the image and/or its clinical usability [1, 5, 19, 2324, 25, 26, 27]. Consequently, this might affect the potential improvement in the treatment planning expected from integrating MRI into the workflow. It is therefore important to understand the fundamentals of metal artefacts and their potential effect on brachytherapy treatment in order to maximize the benefits from MRI and enhance the treatment outcome.

The purpose of this review is to summarize the consequences of using metal objects in brachytherapy treatment and explain the rationale behind the expected artefacts, as well as the recent efforts to evaluate and/or overcome them. We will focus on cervical cancer, which constitutes a major application for MRI-guided brachytherapy.

Technical aspects: why do we expect artefacts?

Metal objects do not produce detectable signal in conventional MR images (i.e., they show up as signal voids). Nevertheless, when they are introduced into a static field, they can potentially interact differently than human tissue. The type of interaction can range from minor artefact in the acquired MR image to serious safety hazards as a result of strong translational forces and torques induced in the object [28]. In order to quantitatively describe the response of a material to the magnetic field, the concept of magnetic susceptibility must be introduced:

The volume susceptibility of a material is dimensionless and usually referred to by χ, where χ0 is often used for the susceptibility of water (χ0 = –9.05 ppm) [25]. Materials with negative χ are diamagnetic while paramagnetic materials have positive susceptibility values. The induced forces and torques of materials with ∣χ∣ < ∼10,000 ppm are usually manageable; nevertheless, they must be tested before MRI use [25]. It is important here to notice that the level of compatibility/safety of a material with MRI is determined based on multiple aspects and tests other than straightforward susceptibility values [28, 29]. The susceptibility values of common materials used in brachytherapy are summarized in Table 1.

Table 1

Susceptibilities of water, tissue, and selected materials [25, 44, 49, 68]

Material	Density (g/cm3)	Susceptibility (ppm)
Gold	19.3	–34
PEEK	1.3	–9.33
Water (37°)	0.933	–9.05
Human tissues	∼0.92-1.05	∼(–11.0 to –7.0)
Air (NTP)	1.29 × 10–3	0.36
Aluminum	2.7	20.7-20.9
Tungsten	19.3	77.2-80
Titanium	4.54	182
Stainless steel (nonmagnetic, austenitic) 8.0	3520-6700

PEEK – polyether ether ketone, NTP – normal temperature [20oC] and pressure [101.325 kPa]

When a material is exposed to an external magnetic field it becomes magnetized; its magnetization is proportional to the magnetic susceptibility of the material and the strength of the applied magnetic field at the point where the material is placed [25]. Mathematically this can be expressed by:

M = χH,

where M is the induced magnetization, χ is its susceptibility and H is the applied magnetic field. Intuitively, when a foreign object is inserted in the region of MR magnetic field, it changes the resultant magnetization. For instance, if a needle is inserted during cervical brachytherapy treatment, it will induce a change in the magnetization proportional to the change in the susceptibility such that Δχ = χ – χ. Therefore, from MR perspective, needles with the same susceptibility as the surrounding tissue would be ideal for such applications rather than needles with zero susceptibility. In practice, however, all the metal objects used in brachytherapy applications are paramagnetic and have larger susceptibilities than that of human tissues. Once magnetized, the object produces a magnetic field that opposes the applied field and hence distort the initial magnetization. This distortion also varies with the shape and orientation of the object with respect to the main field, B0. As MR image acquisition relies mainly on predetermined magnetic field variations, these metal-induced distortions in the magnetic field will consequently induce artefacts in the received image, as described next.

Technical aspects: image distortions

In order to understand the significance of metal artefacts in brachytherapy applications, it is important to grasp the MR principles lying behind such distortions.

The objects placed within the static magnetic field exhibit a magnetization along the direction of the main field (B0). Once a radiofrequency pulse is applied, this magnetization vector is tilted towards the transverse plane, and starts to exhibit a special type of rotation around the main field called, precession. Gradient fields are used to spatially encode the magnetization by slightly changing the precession rates as a function of position. One might then conclude that the presence of a foreign object, disturbing the initial magnetization, would consequently result in unexpected precession rates around that object, causing distortion in the decoded image. A simple analogy is to imagine the process of tuning a radio station from transmitted radio-waves. If there is a distortion in either the receiving or the transmission processes, one might either lose the intended station, hear multiple stations at the same time, or hear a wrong station. In terms of MRI artefacts, this is analogous to signal loss, signal pile-up, or geometric distortions, all of which will be collectively defined here as displacements artefacts [23]. For instance, when a 2D slice is selected, the radiofrequency pulse is applied with certain bandwidth around a particular resonant frequency that corresponds to a particular slice position. When distortions in the magnetic field exist, the excited slice will no longer represent the desired image position and signal from various slice positions could either accumulate or not get excited, causing signal pile-up or signal loss. Likewise, during the signal readout each location is expected to be mapped to a resonant frequency and the presence of unexpected magnetic field variations can cause geometric distortions, signal pile-up, or signal loss effects [30].

In addition to displacements artefacts, signal loss also occurs from the rapid dephasing of spins (T2* decay) around the metal objects; i.e., loss of signal coherence within each voxel causing rapid signal decay. Another problem is the failure of some fat suppression techniques that rely on the homogeneity of the magnetic field [31].

Clinical significance: dosimetric impact

Mapping these artefacts to a brachytherapy application, such as using a titanium applicator, one might expect: 1) the location of the tandem on the MR image might not represent its exact physical location inside the patient in relation to the adjacent organs; 2) signal pile-up and signal loss will occur remarkably around the applicator. The extent of these artefacts will depend on the acquisition parameters, the magnetic field strength, as well as the shape and the orientation of the applicator with respect to the main field [25].

Failing to accurately localize the applicator leads to erroneous identification of the source position during treatment delivery. Accuracy of delivered dose depends on the geometrical accuracy of the source position relative to the target and organs at risk (OARs) [19, 24, 27]. It has been long established that target coverage and OAR sparing is strongly correlated with clinical outcomes. Hence, applicator displacements or mislocalization can lead to dosimetric errors, which in turn compromise the care for patients [19, 24, 26, 32, 33, 34]. Errors in localizing the applicator can be also reflected as uncertainties in the reconstruction process [32]. This results in uncertainties in the dose volume histogram (DVH) parameters for tumor and OARs. Reconstruction uncertainties will be added to other sources of uncertainties in MR-guided workflow such as contouring uncertainties, inter- and intra-fraction variabilities, etc. Tanderup et al. have studied possible systematic uncertainties in applicator reconstruction by simulating shifts in the applicator position (in the order of 2 mm) [24, 26]. They concluded that DVH parameters will deviate less than 10% for 90% of the patients if systematic uncertainties were avoided. Schindel et al. [32] have recently performed a study to simulate the dosimetric impact of applicator displacements as well as the uncertainties of reconstruction. They concluded that in order to avoid more than 10% change in the prescribed doses, applicator displacements and its reconstruction uncertainties should not exceed ± 1.5 mm and 3 mm, respectively. The displacements were only simulated in the cranial-caudal direction. In another study, ± 2 mm displacements in the anterior-posterior direction were found to have significant dosimetric changes [34]. The dosimetric consequences of applicator shifts in PDR 192Ir was studied by DeLeeuw et al. [33]. They found that applicator shifts have larger dosimetric impact on OARs compared to the target. This also matches Tanderup et al. findings, which concluded that rectum and bladder are the most sensitive organs to reconstruction offsets [26], with D2cm 3 variations were 5 ± 1%/mm offset in the anterior-posterior direction, where 90% of the patients have changes < 6%/mm [24, 26]. Nevertheless, it is worth noting that dwell position uncertainty from reconstruction and source positioning for one intracavitary brachytherapy fraction was estimated in some studies to be 4% for either OARs (D2cm 3) or the target (D90) [27].

Metal applicators and needles

It is well known that MRI is the preferred tool for target definition in GYN applications, especially following GEC-ESTRO Gynecological workgroup recommendations [18, 19, 35, 36]. However, when metal applicators/needles are used in MRI-only workflow, it is important also to identify the extent and the significance of the artefacts as it pertains to the accuracy of applicator reconstruction and the influence on target delineation processes. This need became more prominent when titanium applicators were introduced for MR-guided intracavitary brachytherapy treatment. Titanium applicators are used because they offer greater strength with smaller diameter (∼3.2 mm) than plastic applicators [1]. On the other hand, titanium has higher susceptibility than human tissue (∣Δχ∣ = ∼190 ppm [25]), and the resultant artefacts can be a concern in clinical practice. In particular, displacements artefacts, as described above, could present an obstacle for accurate applicator reconstruction if the apparent location of the applicator tip on the image was shifted from its actual location or if the entire applicator position was misregistered with respect to the adjacent anatomy. This is particularly a concern at high field strength ≥ 3.0 T. In addition, signal pile-up and signal loss could lead to erroneous localization of needles/applicators tip.

Previous effort was made to characterize the artefacts of titanium applicators on 1.5 T and 3.0 T, as well as to optimize its visualization for accurate reconstruction [37, 38, 39, 40]. Phantoms and in vivo results have demonstrated that it is feasible to reconstruct the titanium applicator at 1.5 T with a mean inter-observer variability of < 1 mm [1, 37, 40] (Figure1). Artefacts could be significantly larger on 3.0 T (up to 7 mm for 2D T2w fast spin-echo [39]), and careful implementation of the MRI-only planning was strongly recommended [1, 27, 39]. If diffusion-weighted imaging (DWI) is to be acquired while titanium applicator is in situ, careful interpretations of the images should be then performed, as signal-pile up/loss can affect the resultant apparent diffusion coefficient (ADC) maps [2]. The feasibility of DWI on 3.0 T with the titanium applicator in place is questionable [1].

Fig. 1

Adopted from Haack et al. [37], with permission. Coronal (top row) and transversal (bottom row) images showing the titanium applicator in the phantom on computed tomography (CT) images (A and E), in phantom on T1-weighted magnetic resonance (MR) images (B and F), patient T1-weighted MR images (C and G), and patient T2-weighted MR images (D and H) – scanned at 1.5 T. The line on (B) shows the position of the applicator tip according to tip position found in the CT image (A). The arrows in (F and G) indicate the artifact used for determining the rotation of the ring

Recently, tungsten-based tandem applicators have been introduced for high-dose-rate (HDR) intracavitary cervical cancer brachytherapy to achieve more conformal dose distribution to non-symmetrical target volumes [41] – so called direction modulated brachytherapy (DMBT) [41, 42, 43]. This new tandem applicator is designed to have 6 peripheral holes instead of a single central channel used in conventional tandems. Tungsten alloy (> 90% tungsten) was chosen for its high physical density (ρ = 18.0 g/cm3), allowing better OAR dose sparing than titanium or plastic. Tungsten is also paramagnetic and thus expected to produce metal-related artefacts. Luckily, its susceptibility value is relatively closer to human tissue than that of titanium (i.e., ∣Δχ∣ = ∼90 ppm [25, 44]). Preliminary results on phantoms have shown that it produces minimal in-plane artefacts on 1.5 T (∼0.5 mm) and 3.0 T (Figure 2) [45, 46, 47, 48]. A potential advantage of this new applicator is that it has a tip cap manufactured from PEEK, which has similar susceptibility to water (∣χΔ∣ = ∼0.3 ppm) [49]. However, this state-of-the-art technology is still developing and commercial TPS reconstruction libraries are not yet available. Further clinical validation is essential to confirm the preliminary findings, particularly in patient studies.

Fig. 2

Figure adopted from Soliman et al. [47], with permission. A) Shows a plastic radial-fiducial with 4 rods – used to visually evaluate the spatial shift – mounted on a prototype of the tungsten-based DMBT tandem applicator. B) shows a sketch of the orientation of the applicator during the scan. The scan was performed in water doped with MnCl2 using four sequences at 3.0 T. C) T2w-FSE: T2-weighted fast spin-echo, D) PDw-FSE: proton density weighted fast spin-echo, E) T1w-FSE: T1-weighted fast spin-echo, and F) T1w-GE: T1 weighted gradient echo

Interstitial titanium needles are also another place where metal artefacts can be seen. In contrast to conventional plastic needles, titanium needles do not bend inside the patient and often result in better implants [19, 37, 39]. Due to the absence of metal artefacts, plastic needles are a better choice for MRI, however their trajectory cannot be easily predicted within the patients when they bend. On the other hand, the induced artefacts around titanium needles explain the reluctance of adopting them in clinical practice in place of plastic ones, particularly for MRI-only planning. Computed tomography is a better choice to reconstruct titanium needle trajectory, as it can provide a better defined path compared to conventional T2w-MRI [19, 37]. Displacements artefacts as well as signal fluctuations (e.g., pile-up) around the titanium needle can render its localization challenging, which adds a potential source of uncertainty in calculating DVH parameters [26, 50]. It might be worth noting that the impact of reconstruction uncertainty for combined interstitial-intracavitary applicator (plastic) was found to be similar to those from tandem-ring applicator except for the sigmoid where interstitial case has slightly larger impact [26].

Nevertheless, signal pile-up from metal needles and applicators could be beneficial in identifying their location in some cases. For instance, Haack et al. utilized the artefacts to guide the reconstruction process of the titanium applicator [37], while Petit et al. employed 3D spoiled gradient-echo (SPGR) in order to distinguish the applicator from the surrounding tissue [40]. Kapur et al. also benefited from the artefacts surrounding the needles to identify them using a 3D fat-suppressed balanced SSFP sequence [51].

An interesting different approach by de Leeuw et al. is to use a dedicated pulse sequence to locate the center of the magnetic field disturbance of the HDR sources [52], and hence allowing a real-time MR-tracking of the sources. The technique was initially implemented to locate prostate permanent seeds implants [53, 54], but was recently modified to locate HDR sources. Obviously this approach will have its own technical and logistical challenges. For instance, it requires an MR-compatible afterloader, which is not yet commercially available [52], in addition to the specific technical requirements (pulse sequence, post-processing, etc.) that are not yet available at all centers/vendors.

Imaging sequences for applicator reconstruction

While GEC-ESTRO guidelines for GYN recommend conventional 2D-T2w fast spin-echo (FSE) for target and OAR delineation with specific parameters [18, 19], the recommendations did not highlight optimal MR parameters in terms of applicator reconstruction. Isotropic 3D sequences, however, were recommended as they would allow free reformatting of the slices in any plane during image interpretation. One option for 3D imaging is the 3D T2w FSE with variable flip angle, which has been successfully applied in diagnostic applications [55, 56, 5758]. A potential aim could be to replace the conventional 2D FSE with one 3D sequence for both target/OAR delineation and applicator reconstruction. For instance, 3D fast-recovery FSE was previously used for this purpose [59]. However, a serious challenge for such approach is the difference in relative tissue contrast in 3D sequences [55], compared to the conventional contrast provided by 2D sequences, regularly used by clinicians.

It is important to notice here that the methodology of assessing the extent of artefacts of titanium applicator and the accuracy of its reconstruction varied remarkably between groups, particularly from MR imaging perspective. For instance, while some studies highlighted the proton-density FSE sequence to provide better visibility of the applicator [2, 38], T1w gradient-echo was selected as a favorable sequence to visualize the applicator in other studies [39]. Petit et al. [40], on the other hand, have used proton-density spoiled gradient-recalled-echo in evaluating the visibility of the titanium Rotterdam applicator. Sequences also varied between 2D and 3D [1, 37, 39, 40, 59]. Additionally, the slice thickness used in these studies varied between 1 and 5 mm [27, 37, 39, 40, 60]; this affects the level of the out-of-plane artefacts, and consequently the uncertainties of the reconstruction in the slice direction, as well as the SNR of the images on which the reconstruction is performed. Studies also differ in whether or not MRI-markers are used to guide the reconstruction process [1, 2, 19, 37, 38, 39, 40, 61].

Advanced pulse sequences to correct for metal artefacts have been developed in the MRI community particularly for diagnostic purposes near metallic implants [62, 63, 64, 65, 66]. Unfortunately, there has been little interest to optimize these sequences for brachytherapy applications. One main reason might be their lengthy acquisition times. Warner et al. [67] have explored various metal-artefact reduction techniques for brachytherapy planning and concluded that SEMAC [64] was more efficient than the other methods.

While there is not yet an optimal standard sequence to visualize the applicator, there is a clear practical consensus that T2w sequence is suboptimal for applicator reconstruction compared to other clinical sequences that have shorter echo-time (TE) such as proton-density FSE or T1w FSE.

Summary, recommendations, and future perspectives

The last decade has seen an increasing interest in understanding technical and clinical challenges accompanying the usage of metal applicators and needles in gynecological brachytherapy. Titanium needles and applicators are preferred over plastic as they offer greater strength at smaller diameters. Tungsten offers an additional advantage, as it allows for non-symmetrical dose distribution, with specific tandem designs, owing to its higher physical density. However, due to the difference in magnetic susceptibilities, the induced metal-artefacts render the accurate localization of metal objects more challenging in MRI-only planning. Phantom and clinical studies demonstrated that conventional 2D T2w FSE is suboptimal in localizing the applicators/needles, particularly when thick slices (≥ 4 mm) are acquired [37]. Sequences with shorter echo-time (TE) such as PDw have provided better visibility necessary for applicators reconstruction [38].

FSE-based sequences have the advantage over gradient-echo (GRE) sequences of overcoming much of the signal loss occurring from the signal dephasing (incoherence) that is seen around the metal object. However, 3D FSE acquisitions are substantially lengthy, compared to 3D GRE-based sequences that can cover the whole imaging volume in ≤ 5 minutes [40]. Nevertheless, 3D isotropic resolution can be achieved using an optimized “3D FSE with variable flip angle sequence” in a reasonable scan time (≤ 10 min) [18, 56, 58, 59]. Higher readout bandwidth and higher spatial resolution can reduce geometric and in-plane distortions induced by metal objects. The use of thin slices also reduces the distortions in the slice direction. However, this will also result in lower SNR, particularly at low field strength (e.g., 1.5 T). Multiple averages can be then acquired to compensate for SNR loss.

Given the existing variety of pulse sequences for metal-artefacts reduction, the absence of a dedicated sequence for brachytherapy applications is a witness of the still-existing gap between MRI and radiation oncology communities [5]. Current metal-artefacts reduction techniques need yet to be explored and adopted for brachytherapy needs, such as applicator reconstruction. Larger cohort studies to explore the dosimetric impact of the metal artefacts with clinical sequences may be also needed to resolve the clinical concerns regarding the regular usage of metallic applicators/needles.