Characterizing geometrical accuracy in clinically optimised 7T and 3T magnetic resonance images for high-precision radiation treatment of brain tumours

Abstract

Background and purpose

In neuro-oncology, high spatial accuracy is needed for clinically acceptable high-precision radiation treatment planning (RTP). In this study, the clinical applicability of anatomically optimised 7-Tesla (7T) MR images for reliable RTP is assessed with respect to standard clinical imaging modalities.

Materials and methods

System- and phantom-related geometrical distortion (GD) were quantified on clinically-relevant MR sequences at 7T and 3T, and on CT images using a dedicated anthropomorphic head phantom incorporating a 3D grid-structure, creating 436 points-of-interest. Global GD was assessed by mean absolute deviation (MAD_Global). Local GD relative to the magnetic isocentre was assessed by MAD_Local. Using 3D displacement vectors of individual points-of-interest, GD maps were created. For clinically acceptable radiotherapy, 7T images need to meet the criteria for accurate dose delivery (GD < 1 mm) and present comparable GD as tolerated in clinically standard 3T MR/CT-based RTP.

Results

MAD_Global in 7T and 3T images ranged from 0.3 to 2.2 mm and 0.2–0.8 mm, respectively. MAD_Local increased with increasing distance from the isocentre, showed an anisotropic distribution, and was significantly larger in 7T MR sequences (MAD_Local = 0.2–1.2 mm) than in 3T (MAD_Local = 0.1–0.7 mm) (p < 0.05). Significant differences in GD were detected between 7T images (p < 0.001). However, maximum MAD_Local remained ≤1 mm within 68.7 mm diameter spherical volume. No significant differences in GD were found between 7T and 3T protocols near the isocentre.

Conclusions

System- and phantom-related GD remained ≤1 mm in central brain regions, suggesting that 7T MR images could be implemented in radiotherapy with clinically acceptable spatial accuracy and equally tolerated GD as in 3T MR/CT-based RTP. For peripheral regions, GD should be incorporated in safety margins for treatment uncertainties. Moreover, the effects of sequence-related factors on GD needs further investigation to obtain RTP-specific MR protocols.

1. Introduction

In radiation treatment planning (RTP) of brain tumours, magnetic resonance imaging (MRI) at 1.5 or 3 Tesla (T) is currently being used as the standard anatomical imaging modality owing to its superior soft-tissue contrast compared to computer tomography (CT) [1]. Generally, after co-registration of MR and CT images, the former is used for target volume definition and the latter is used for dose calculation, as tissue electron density information is missing in MR images. However, the current clinical MR techniques are limited in depicting detailed neurologic malformations such as intracerebral tumour spread [2]. With the prospect of clinically-certified ultra-high field (UHF-)MRI systems (≥7T), images with higher signal-to-noise ratio (SNR), higher spatial resolution, and novel contrast mechanisms such as quantitative susceptibility-weighted images will become available, enabling the visualisation of small lesions, basal ganglia, and tumour angiogenesis [3], [4], [5]. The improved ability in detecting microvasculature (diameter ∼100 μm) could play a decisive role in staging primary brain tumours such as glioblastoma (GBM), as tumour angiogenesis is directly associated with tumour grade, and could aid in determining relevant location of dedifferentiated cells for image-guided biopsies [6], [7]. In addition, 7T MRI could aid target volume definition in RTP by visualising microvasculature outside the contrast-enhanced tumour and infiltration of GBM cells along white matter tracts [7], [8], [9]. During follow-up scans, 7T MRI has been shown to reveal radiation-induced microbleeds, indicating potential neurocognitive decline and the need to adapt RTP [10], [11].

However, concerns regarding geometrical distortion (GD) with increasing static magnetic field strength (B₀) have compromised the integration of UHF-MRI into radiation treatment planning. System-related GD are mainly induced by imperfections in hardware (B₀-inhomogeneity, gradient field nonlinearity, and eddy currents), applied gradient pulse scheme and parameter settings (sequence-dependent GD), while phantom-related GD (chemical shift and susceptibility differences) are related to the object’s shape placed in the MR system [12], [13]. To be applicable for high-precision RTP, the spatial accuracy of anatomical images needs to be within 2 mm for non-stereotactic radiotherapy and within 1 mm for stereotactic radiotherapy, as is the case for 3T MRI [13], [14], [15], [16], [17], [18]. In fact, profound GD may have significant dosimetric impact in radiotherapy as the precision of beam targeting and dose calculations could be affected [19], [20]. For high-precision dose delivery techniques that apply steep 3D dose gradients (intracranial stereotactic radiotherapy or radiosurgery), spatial inaccuracies of >1 mm could result in clinically unacceptable dose variations of more than 15% that could lead to underdosage of target volume and/or overdosage of surrounding tissue [20], [21]. Ultimately, inaccuracy of beam targeting and dose calculation could lead to a different clinical outcome and local tumour control [13], [18]. For instance, techniques that apply extremely steep dose gradients, such as stereotactic radiotherapy and radiosurgery, are susceptible to significant dosimetric effects (5–32% underdosage) caused by even small GD of 1.3 mm [21], [22]. The question therefore arises whether our clinical 7T MRI protocol, optimised for anatomical imaging, can produce images with acceptable spatial accuracy needed for reliable high-precision radiation treatment of brain tumours. Previously several investigators have evaluated geometrical distortion on either CT, 3T or 7T MRI, demonstrating a maximum GD of up to several mm in large cylindrical or cubical phantoms [14], [23], [24], [25], [26], [27].

This study aims to investigate the clinical applicability of anatomical 7T MRI in comparison to 3T MRI and CT for reliable high-precision radiotherapy by evaluating the magnitude and location of hardware-related GD. We hypothesise: 1) acceptable submillimetre spatial accuracy could be achieved in our 7T MRI protocol and no significant differences in GD are present between different 7T sequences; 2) the tolerated level of GD in 7T MRI is not significantly different from system-related GD in a predefined 3T MRI protocol, clinically used for RTP. To test these hypotheses on both global level (GD throughout the entire volume-of-interest (VOI)) and on local level (GD relative to the distance from the magnetic isocentre), we performed quantitative image analyses using a dedicated anthropomorphic head phantom in a clinically realistic set-up. Hence, different sequence parameter settings and distortion correction methods were not explored in this study.

2. Materials and methods

2.1. Phantom

A dedicated anthropomorphic head phantom (CIRS Model 603A, Computerized Imaging Reference Systems Inc., USA) was used to evaluate system-related GD in 7T MRI, 3T MRI, and CT images (Fig. S.1). The anatomical shape demanded similar procedures used for clinically imaging patients (FOV setting, shimming, robust positioning inside the scanner and head coil). Furthermore, the centre of the phantom was aligned with the magnetic isocentre.

The phantom was composed of a plastic-based bone substitute and soft-tissue equivalent fluid consisting of a water-based polyacrylamide with added NiCl₂ (Zerdine®, CIRS). Inside the phantom, 3-mm thick nylon rods formed a 3D grid with 15 mm spacing (±0.05 mm manufacturing tolerance), defining a VOI of 15 × 12 × 13.5 cm³ (anterior-posterior (AP) × left–right (LR) × superior-inferior (SI)). At grid-intersections, 437 measurable points-of-interest (POIs) were identified. Nylon shows similar magnetic susceptibility properties to water (χ < 3 ppm) and was assumed not to induce such susceptibility artefacts in spin-echo or gradient-echo sequences [24], [28].

2.2. Image acquisition

MR images were acquired on a 7T MR system (Magnetom 7T, Siemens, DEU) and on a 3T MR system (Achieva 3T, Philips, NL) using 32-channel head coils from Nova Medical inc. (1Tx/32Rx, Wilmington, USA) and Invivo Corp. (Florida, USA), respectively. The 7T MR system was equipped with Siemens’ SC72-gradient system (80 mT/m peak gradient amplitude, 200 T/m/s maximum slew rate). The 3T MR system was fitted with Philips’ Dual Quasar gradient system (80 mT/m peak gradient amplitude, 100 mT/m/s maximum slew rate). Magnetic field homogeneity specification were described as <5 ppm pk-pk and ≤1.1 ppm at a diameter of a spherical volume (DSV) of 45 cm, respectively. All MR sequences were previously optimised to acquire anatomical images of high image quality within an acceptable scan duration, and were not altered for this study [9]. The selected 7T MR sequences included magnetization-prepared rapid gradient-echo (MP2RAGE), T2-sampling perfection with application optimized contrasts using different flip angle evolution (T2-SPACE), T2-SPACE with fluid attenuation inversion recovery (T2-SPACE FLAIR), and multi-echo gradient-echo (T1-GRE) [9]. Sequences with equivalent pulse schemes were selected from a predefined clinical 3T MR protocol used in RTP and included turbo field echo (3D TFE), volume isotropic turbo spin-echo (T2-VISTA) with and without FLAIR, and fast field gradient echo (T1-FFE), respectively (Table S.1). Vendor-installed 3D distortion correction methods were applied to restore gradient non-linearity and 3D volume-based second order shimming procedures were applied to improve B₀-homogeneity [14], [29], [30].

CT images were acquired (SOMATOM Sensation 10, Siemens, Germany) with a slice thickness of 1 mm, 306 slices, 50 × 50 cm² FOV, 140 kV, and 400 mAs.

2.3. Methods of analyses

The known 3D coordinates of all POIs were defined as reference dataset and manually reconstructed in MRI and CT images using Eclipse treatment planning system (Varian Medical Systems, USA). The X-, Y-, and Z-axes defined the LR-, SI-, and AP-directions in both datasets, respectively.

Three methods were used to assess GD at different levels of sophistication using in-house developed MATLAB software (R2014b, MathWorks Inc, USA) (Fig. 1). Firstly, mean absolute deviation (MAD) was computed to assess GD on a global level (MAD_Global) throughout the entire VOI, independently of location within the phantom. Secondly, MAD was calculated to assess GD on a local level (MAD_Local) relative to the distance from the magnetic isocentre. MAD was calculated according to:

$$MAD\left(i\right)=\frac{1}{N\left(i\right)}\sum _{j=1}^{N\left(i\right)}|{\text{D}}_i\left(j\right)-{\delta }_i|$$ (1)

where i = 1…M is an index into set $\mathrm{\Delta }=\left\{{\delta }_1,{\delta }_2,\cdots ,{\delta }_M\right\}$ of M unique distances, $D_i\left(j\right)$ is the measured Euclidean distance for the j-th unique pair of POIs having an Euclidean distance ${\delta }_i$ in the reference dataset, and $N\left(i\right)$ represents the number of unique pairs of POIs. For MAD_Global, $\mathrm{\Delta }$ contained all unique distances that could be observed within the entire phantom. For MAD_Local, $\mathrm{\Delta }$ contained only unique distances that originate from the POI nearest the magnetic isocentre. The mean MAD was averaged over all M unique distances in $\mathrm{\Delta }.$

Fig. 1.

Overview of 3D data acquisition and analyses. XYZ-coordinates are determined on CT and MR images (1a.), and based on product characteristics of the CIRS’ phantom model 603A (1b.). Global MAD (2a.) is based on distances with 2 variable grid-intersection points (*example given for 1 intersection but applies for all points). Local MAD (2b.) is based on distances between magnetic field isocentre and 1 variable grid-intersection points. Displacements vectors (3a.) are determined between the measured and reference coordinates of each individual data-point and indicate the relative 3D geometrical distortion.

The third method of analyses pin-pointed GD by calculating 3D displacement of each individual POI by computing the Euclidean distance directly between the POI-coordinates in the image (X_m, Y_m, Z_m) and the coordinates in the reference datasets (X_ref, Y_ref, Z_ref):

$$D_{\mathit{xyz}}=\sqrt{{\left(X_m-X_{\mathit{ref}}\right)}^2+{\left(Y_m-Y_{\mathit{ref}}\right)}^2+{\left(Z_m+Z_{\mathit{ref}}\right)}^2}$$ (2)

The resulting 3D displacement vector (D_xyz) indicated relative GD of each individual POI, independently of all other POIs.

The three methods of analyses provided complementary information on the geometrical quality of the images. The MAD_Global quantified the level of GD throughout the entire VOI, thereby ignoring information on the absolute and relative locations of the POIs. The MAD_Local retained spatial information by quantifying the GD with respect to the DSV at the magnetic isocentre. To quantify spatial integrity of an entire image, 95% confidence interval (|CI₉₅|) was calculated from MAD and standard deviation (SD) as:

$$\left|{\mathit{CI}}_{95}\right|=MAD+2SD$$ (3)

Based on this interval, a maximum acceptability-level of 1 mm was defined. The level of GD was considered acceptable when MAD_Local ≤ 1 mm (at a specific DSV) and |CI₉₅| ≤1 mm.

2.4. Statistical analyses

To assess differences in sequence-dependent GD, MAD_Global and MAD_Local were analysed between 7T MRI sequences and between 3T MRI sequences using a Kruskal-Wallis test. To assess differences in tolerated GD-levels, MAD_Global and MAD_Local were analysed between equivalent sequence pulse schemes in 7T and 3T MRI using a multiple student t-test with Bonferonni-Sidak correction. For MAD_Local, DSV at which statistical significance in MAD_Local between sequences occurred, was identified by a separate Kruskal-Wallis test for each set of unique distances. All statistical analyses were performed using GraphPad Prism v6.01 (GraphPad, USA). P-values < 0.05 were considered to indicate statistical significance.

3. Results

3.1. Global GD

In total, 95.266 pairs of distances were identified and binned into 100 unique distances, ranging from 15 mm to 164.3 mm. In CT images, MAD_Global ranged from 0.1–0.6 mm with mean MAD_Global of 0.3 ± 0.1 mm. For MR images, MAD_Global ranged from 0.3–2.2 mm and 0.2–0.8 mm in 7T and 3T MRI, respectively (Table 1, Fig. 2). Based on |CI₉₅|, only 3T MR sequences presented GD < 1 mm throughout the entire image. No statistical significance in MAD_Global were found between 7T MR images (P = 0.38) or 3T MR images (P = 0.13). However, significant differences in MAD_Global were detected between all equivalent sequences on 7T and 3T MR system (P < 0.001) (Fig. 3.).

Table 1.

Global geometrical deformation as measured on 3T and 7T MRI (***  = P < 0.001).

		MADGlobal (in mm)					MADGlobal (in mm)
B0	Sequence	Range	Mean ± sd	|CI95|	B0	Sequence	Range	Mean ± sd	|CI95|	P value
7T	MP2RAGE	0.3–1.9	1.0 ± 0.3	1.7	3T	3D TFE	0.22–0.51	0.27 ± 0.23	0.7	***
	SPACE	0.3–2.0	1.0 ± 0.4	1.7		T2-VISTA	0.20–0.70	0.26 ± 0.20	0.7	***
	SPACEFLAIR	0.3–2.2	0.9 ± 0.3	1.6		T2-VISTA FLAIR	0.19–0.62	0.29 ± 0.23	0.7	***
	GRE	0.3–1.9	1.0 ± 0.4	1.8		T1-FFE	0.3–0.7	0.4 ± 0.3	1.0	***

Fig. 2.

Absolute differences in Euclidian distance between the measured and reference dataset (grey dot) relative to the unique distances found in the reference dataset, observed within CT (a), MP2RAGE (b), T2-SPACE (c), T2-SPACE FLAIR (d), T1-GRE (e), 3D TFE (f), T2-VISTA (g), T2-VISTA FLAIR (h), T1-FFE (i). The overall geometric distortion was quantified by MAD_Global (±SD) (blue square). The 95% confidence interval (CI₉₅) is shown as the dotted horizontal line. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3.

MAD_Local values (±SD) relative to the distance from the magnetic isocentre at 7T (a) and 3T (b) MRI, both relative to CT (black square). Presented 7T MR sequence include MP2RAGE (green circle), T2-SPACE (purple diamond), T2-SPACE FLAIR (orange downward triangle), and T1-GRE (blue upward triangle). The same colour- and shape-code was used for the equivalent 3T sequences, 3D TFE, T2-VISTA, T2-VISTA FLAIR, and T1-FFE, respectively. The dotted horizontal line represents the 1 mm-acceptability level required for spatially reliable RTP. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.2. Local GD

In CT images, MAD_Local ranged from 0.1 to 0.3 mm with mean MAD_Local of 0.2 ± 0.04 mm (Fig. 4). All MR images acquired at 7T presented a steady increase in MAD_Local with increasing distance from the magnetic isocentre. MAD_Local ranged from 0.2–1.8 mm for 7T MRI and from 0.1–0.7 mm for 3T MRI. Each 7T MR sequence remained below the clinical tolerance-level of 1 mm up until a defined DSV (Table 2). For T1-GRE, MAD_Local exceeded 1 mm only at DSV of 87.5–88.7 mm.

Fig. 4.

Distortion maps of MP2RAGE on 7T MRI (a) and 3D TFE on 3T MRI (b) measured in the axial plane nearest the magnetic isocentre (Y = 0), in the coronal plane (Z = 0), and in the sagittal plane (X = 0).

Table 2.

Local geometrical deformation as measured on 3T and 7T MRI (*** = P < 0.001, § = DSV at which statistical significance between sequences occurred).

		MADLocal (in mm)							MADLocal (in mm)				P value
B0	Sequence	Range	Mean ± sd	|CI95|	<1 mm at DSV	B0	Sequence	Range	Mean ± sd	|CI95|	<1 mm at DSV	Sign.	P < 0.05 at DSV§
7T	MP2RAGE	0.2–1.8	0.8 ± 0.3	1.4	69	3T	3D TFE	0.1–0.6	0.3 ± 0.2	0.6	all	***	42
	SPACE	0.2–1.18	0.7 ± 0.3	1.2	102		T2-VISTA	0.2–0.4	0.4 ± 0.3	0.9	all	***	69
	SPACEFLAIR	0.3–1.2	0.6 ± 0.3	1.2	93		T2- VISTA FLAIR	0.1–0.4	0.2 ± 0.2	0.5	all	***	47
	GRE	0.2–1.1	0.7 ± 0.3	1.3	87–89		T1-FFE	0.2–0.7	0.3 ± 0.3	0.9	all	***	45–56

For all 3T MR sequences, MAD_Local remained below 1 mm throughout the whole VOI. MAD_Local of MP2RAGE was statistically greater than all other 7T MR sequences (P < 0.001, Fig. S.2). No significant difference in MAD_Local was found between T2-SPACE, T2-SPACE FLAIR (P = 0.064), and T1-GRE (P = 0.093), respectively. Within the 3T dataset, MAD_Local of 3D TFE was significantly different from T2-VISTA (P < 0.001), T2-VISTA FLAIR (P = 0.046), and T1-FFE (P < 0.001). T2-VISTA FLAIR presented significantly different MAD_Local from T1-FFE (P < 0.001). MAD_Local was not significantly different between T1-FFE and T2-VISTA (P = 0.51), and the T2-VISTA and T2-VISTA FLAIR sequences (P = 0.061). However, all sequences met the criteria for GD and seemed to have found an equilibrium with requirements for anatomical imaging.

Statistical analyses of MAD_Local between equivalent 7T and 3T sequences indicated significant differences for all sequences (P < 0.001). However, no significant difference in MAD_Local was found between equivalent sequences when the DSV was considered (Table 2).

3.3. 3D displacement

The magnitude of each displacement vector is presented in a 2D mesh-plot in axial (XZ), coronal (XY), and sagittal (YZ) planes, intersecting the magnetic isocentre. The worst GD was presented in MP2RAGE and 3D TFE (Fig. 4). Additionally, anisotropic distributions of GD were observed along the phase-encoding direction and frequency-encoding direction. Particularly the anterior–superior section was affected by spatial deformation and maximal GD was observed in the right-superior section of the anterior region (frontal lobe). Distortion maps of all sequences are presented in Figs. S.3–5.

4. Discussion

This study reports both system- and phantom-related GD on clinically-relevant MR sequences at 7T MRI for neurological imaging and allows direct comparison to GD on both 3T MRI and CT images. System-related GD was present in all 3T and 7T MRI and increased with increasing distance from the magnetic isocentre, even though vendor-provided correction methods and 2nd order 3D shimming were applied. However, GD of ≤1 mm could be assured within a sequence-dependent DSV near the magnetic isocentre.

While UHF-MRI promises clinical gain in terms of spatial resolution, SNR, contrast, and delineation of pathology, implementation of 7T MRI in RTP requires mm-scale spatial accuracy for reliable, high-precision radiotherapy. This geometrical requirement was met by all 3T MR sequences throughout the entire VOI while the magnitude of GD in 7T MRI depended on the distance from the magnetic isocentre and applied sequence. GD ≤ 1 mm was apparent in 7T MRI within a sequence-specific DSV. Moreover, no significant differences in system-related GD were found between 3T and 7T images near the magnetic isocentre. This implies that these 7T MRI with clinically acceptable/tolerable levels of spatial uncertainty could be applied for high-precision RTP in central brain regions, under the assumption that tissue-related GD is negligible. In addition, it has previously been shown that GD in 7T MRI did not affect targeting of basal ganglia and subthalamic nuclei in deep-brain stimulation [23], [31].

Some sequences are more prone to GD than others as gradient pulse scheme and parameter settings affect B₀-homogeneity and eddy currents [25], [29], [32]. Our MP2RAGE sequence, which acquires 2 gradient-echo images at different inversion times, presented the largest GD at 7T (and smallest DSV where GD < 1 mm) [33]. Nevertheless, even in this worst-case scenario, system-related GD at 7T was below the clinically acceptable limit of ≤1 mm within 68.7 mm DSV. Presumably, GD in MP2RAGE could be related to the combination of different sequence parameter settings (receiver bandwidth (rBW), FOV, and matrix size) [13], [14], [25], [29], [31], [32], [34]. The effect of varying rBW on GD and SNR has extensively been studied by Walker et al. (2014) in 4 different MR systems [31]. For instance, increasing rBW would reduce GD, chemical shift, and susceptibility artefacts but would also reduce SNR. Limiting FOV settings (i.e., fewer acquisition lines in k-space) could subside the impact of time-dependent off-resonance effects, reducing GD [34], [35]. However, this could also influence spatial resolution, SNR, and induce potential fold-over artefacts.

Residual GD in phase-encoding direction is expected to be related to gradient non-uniform and could be adequately reduced by vendor-installed distortion correction methods [14], [18], [25], [26]. The technique behind this correction is similar for all main MR manufacturers and achieve comparable and reproducible results for 3T and 7T MRI [4], [14], [15], [18], [31]. Nonetheless, some residual system-dependent GD was detected in this study despite applying correction methods and 3D shimming. It is therefore important to note that vendor-provided distortion correction is not flawless, uncorrected B₀-inhomogeneity and eddy currents could still induce GD, and gradient coils are designed with a tolerated performance variation [24], [36]. For example, general gradient errors of body gradient coils should be >2% of the gradient strength over a 40 cm DSV[37], [38].

Our findings of GD corresponded well with literature. At 3T, Stanescu et al. reported maximal overall GD of 0.6 mm in 3D TFE sequences and Schmidt et al. have shown global GD of less than 0.5 mm in an IR-sequence (rBW = 890 Hz/pixel, resolution 1 mm³) [14], [39]. However, some studies report GD of several centimetres [25], [26], [27]. These findings, however, occurred at the edge of the scanner’s FOV on a large cubical phantom that was placed in a body coil. In contrast to this, we used a dedicated anthropomorphic head phantom to mimic a clinically realistic scan procedure with standard radiofrequency coils, shimming, distortion correction, and FOV settings (≤256 mm).

Similar to our approach and FOV setting, Dammann et al. investigated hardware-related GD in 7T MRI using a cylindrical phantom (190 POIs) and presented local GD smaller than 1 mm at <80 mm DSV, except in T2-TSE (rBW = 350 Hz/pixel, resolution 0.6 mm²) that showed maximal local GD of 1.6 mm [26]. Presumably, this sequence was affected by intravoxel dephasing, image blurring and inaccurate measurements at the phantom’s edge [26]. This could also could explain our finding of 1.2 mm local GD at 80 mm DSV. Differences in rBW-setting could explain GD differences found in MPRAGE (rBW = 600 Hz/pixel, resolution 1 mm³) and MP2RAGE (rBW = 248 Hz/pixel, 0.7 mm³). Recently, Lau et al. demonstrated differences in local GD between 3T and 7T larger than 1 mm at 80 mm DSV for T1-MPRAGE (rBW = 195 Hz/px, 1 mm³) [4]. In our study, MP2RAGE showed differences in MAD_Local between 3T and 7T to be below 1 mm up for a ≤87.5 mm DSV.

Distortion maps of MP2RAGE indicated an anisotropic distribution of GD along the phase-encoding (AP) and frequency-encoding (SI) direction (Fig. 4). In a comparable 7T phantom study by Cho et al., coronal distortion maps showed 0.4 mm GD along the frequency-encoding direction and 0.8 mm along the phase-encoding direction [22].

We observed GD in CT images with MAD_Global ranging from 0.1 to 0.6 mm and MAD_Local ranging from 0.1 to 0.3 mm. These findings were unanticipated as it is generally recognized that CT images are distortion-free. Other studies have merged MRI with CT, ignoring this innate GD and measured fundamental errors. Artefacts in CT images could originate from incorrectly reported scanner table speed or shearing distortion (incorrectly reported gantry tilt or table bending). In this study, GD in CT images could represent partial volume effects and variations in manual reconstruction leading to slightly different xyz-coordinates measurements of POIs (experimental inaccuracies).

In this study, we limited ourselves to evaluating system-related GD and not tissue-related GD. Relative to system-related GD, tissue-induced GD is rather small in 3T sequences that are currently being used for RTP of brain tumours, but it is nevertheless not negligible [16]. Simulations on 3T MRI suggest that susceptibility-induced GD were on average 0.6 mm at cranial air cavities [40]. This has recently been confirmed by Schmidt et al., who specified susceptibility-related GD of 0.4 mm in the naso-oropharyngeal cavities and around the internal ear canal using a T1-weighted sequence (rBW = 890 Hz/pixel, resolution 1 mm³) [14]. It remains to be determined what the impact of tissue-related GD is for 7T MRI. Susceptibility artefacts are proportional to the magnetic field strength and TE but could be reduced by increasing rBW [12]. Regardless, patient-specific correction methods are required [16], [37], [40]. Recently, Rai et al. (2018) developed an 3D printed human skull with sinuses and mastoid air cells, water-based brain tissue and eyes and surrogate cortical bone [41]. Such realistic anatomical skulls could be used to further assess tissue-induced GD in simulated air cavities, bony anatomy, and soft tissue. Since system-related GD could be further reduced by appropriate setting of MR sequence parameters, research has been performed to systematically assess the impact of e.g., rBW, FOV, matrix size, scanning direction, and various (post-)processing correction algorithms [16], [18], [27], [31], [36]. However, 7T MR protocols need to be optimised specifically for each clinical application and the trade-off between anatomical image quality (spatial resolution, SNR, visibility of pathology) and image integrity (spatial accuracy) needs to be made, while respecting specific absorption rate restriction [29], [32]. For example, high-precision RTP relies heavily on high geometrical accuracy but could compensate on image resolution, while in radiology larger GD could be tolerated.

To apply 7T images in radiotherapy, GD should be ≤1 mm to ensure reliable RTP and dose delivery with local dose variations less than 5–10% [21]. This strict criterion was met by all the clinically relevant sequences we tested on 3T and 7T MRI up until 92.5–101.7 mm DSV around the magnetic isocentre, except for the MP2RAGE sequence. Delineations of brain lesions and organs-at-risk in the frontal or occipital lobe on 7T MRI can be significantly influenced by GD as the spatial uncertainty exceeds 1 mm in these regions. Even though systematic and random treatment uncertainties are foreseen in a planning treatment volume (PTV), the degree and location of GD is rarely taken into account during radiation dose planning. This study has shown that system-related GD should be incorporated into the PTV margin to achieve a high level of dose delivery accuracy for intracranial (stereotactic) radiotherapy. This can be accomplished by incorporating the GD present at the tumour location within a margin recipe for dose delivery as suggested by Van Herk et al. and Seravelli et al. [42], [43]. We therefore recommend applying larger tumour site-specific PTV margins in regions where spatial integrity of 7T MRI could not be warranted within 1 mm. Apart from adjustment of PTV margins, it is advised to find a trade-off between image quality and image distortion, and apply MR imaging protocols that are dedicated for RTP purposes.

In conclusion, GD of ≤1 mm could be assured within a sequence-dependent DSV near the magnetic isocentre, implying that this 7T MRI protocol could be applied with clinically acceptable/tolerable levels of spatial uncertainty for high-precision RTP in central brain regions, under the assumption that tissue-related GD can be ignored. For peripheral regions, 7T MR protocol for RTP should incorporate GD in tumour-site specific PTV margins for treatment uncertainties. Nevertheless, dedicated MR protocols are aspired for application in radiation oncology and further optimisation of sequence parameter settings and GD reduction methods are needed.

Disclosure of interest

The authors report no conflict of interest.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Supplementary figure 1.

Supplementary figure 2.

Supplementary figure 3.

Supplementary figure 4.

Supplementary figure 5.