Quality requirements for MRI simulation in cranial stereotactic radiotherapy: a guideline from the German Taskforce “Imaging in Stereotactic Radiotherapy”

Magnetic Resonance Imaging (MRI) has a long-standing role in radiotherapy treatment planning for brain tumors and is essential today in nearly all cranial treatment indications [1]. Stereotactic Radiosurgery (SRS) and Fractionated Stereotactic Radiotherapy (FSRT), collectively referred to as Stereotactic Radiotherapy (SRT), demand exceptional precision to achieve doses of high biological effectiveness in tumors while simultaneously preserving the adjacent normal tissues and organs-at-risk (OARs) [2, 3]. This paradigm of cranial SRT has been shown to achieve high local cure rates with limited to minimal toxicities [4‐6]. The stringent precision requirements of cranial SRT necessitates not only highly precise treatment planning and delivery, but also critically depends on optimal MRI for target volume definition. The previous German guideline on technological quality requirements for stereotactic radiotherapy explicitly pointed out that imaging for SRT treatment planning needs exceptional attention where “the target volume and all organs-at-risk are defined using organ-specific imaging modalities” and “secondary imaging requires accurate registration with the thin-slice planning computed tomography (CT)” [3]. This is especially important for MRI as an integral part of cranial SRT which most often forms the fundamental basis for all further treatment steps. Errors and uncertainties in MRI can often not be compensated at later stages in the treatment planning process and are propagated throughout the treatment planning chain, possibly leading to suboptimal treatment with inferior outcome. Multiple groups clearly demonstrated that inadequate MRI would affect the accuracy of gross target volume (GTV) delineation, which can easily diminish the clinical outcome [7‐9]. Simply expanding the margins to compensate for uncertainties and imaging errors will increase the planning target volume (PTV), consecutively decrease the therapeutic selectivity, and will ultimately diminish the therapeutic principles of precision radiotherapy. Cranial stereotactic radiotherapy was outside the scope of recently published guidelines on MRI simulation in radiotherapy, like the AAPM task group 284 report [10, 11] and a comprehensive guideline on MRI simulation for cranial SRT was missing. Optimal MRI simulation and target volume definition frequently require interdisciplinary input and close collaboration, integrating radiooncologic, diagnostic, neurosurgical and physics expertise. In this guideline, therefore, an interdisciplinary panel including experts from the working group for radiosurgery and stereotactic radiotherapy of the German Society for Radiation Oncology (DEGRO), the working group for physics and technology in stereotactic radiotherapy of the German Society for Medical Physics (DGMP), the German Society of Neurosurgery (DGNC), the German Society of Neuroradiology (DGNR) and the German Chapter of the International Society for Magnetic Resonance in Medicine (DS-ISMRM) have defined minimum quality requirements as well as advanced simulation options for MRI in cranial SRT to increase its quality in clinical practice and ultimately improve treatment outcomes.

To formulate this guideline paper, a cross-disciplinary expert task force was constituted. The task force contained experts from radiation oncology, medical physics, neurosurgery, neuroradiology, radiology and MRI physics. The task force conducted biweekly virtual meetings from December 2022 to July 2023. First a systematic literature review on the areas of geometric accuracy, sequence selection and optimization of sequence protocol parameters, contrast agent-related parameters, time interval between MRI simulation and treatment delivery as well as image registration and imaging in SRT position was performed. This literature review was revisited, evaluated, and refined through iterative discussions in the regular meetings. Subsequently, requirements and recommendations were derived from the literature review with consensus being optimized in a two-stage process. Following an initial round of voting, the proposed requirements and recommendations were re-evaluated and fine-tuned in interdisciplinary discussions before proceeding to a final voting round. Each requirement and recommendation was subjected to a vote on agreement (with possible responses: “yes,” “no,” or “abstention”) and, in the case of agreement, the category (“minimum requirement”, “additional recommendation”, or “optional”). Guideline statements that were classified as “minimum requirement” were considered mandatory for MRI simulation in SRT. “Additional recommendations” were defined as recommendations that should be applied for optimal MRI simulation but are not considered mandatory. Finally, “optional” statements provide advanced options that can be implemented by experienced centers. Consensus was quantified as the percentage of agreement, excluding abstentions. For each requirement and recommendation, the rates of consensus, abstention, and the votes on the statement category are provided.

Patients referred to stereotactic radiotherapy generally have their diagnosis established through diagnostic imaging beforehand. The primary objective of MRI for radiotherapy planning therefore is to accurately depict the tumor location and shape in three-dimensional space so that the gross tumor volume (GTV) can be precisely delineated. Generally, high-resolution isotropic 3D sequences are optimal for this task since they permit accurate multiplanar reconstruction and minimize over- or underestimation of the GTV because of partial-volume effects ([52, 53]; Fig. 2). Additionally, 3D sequences provide continuous imaging of the brain and targets without gaps and demonstrate lower susceptibility to distortions arising from B₀ inhomogeneities compared to 2D sequences [12, 13, 17] (see above for a detailed discussion on optimizing sequence parameters for geometric accuracy). As a rule of thumb, the volumetric error will exceed 10%, if the target is visualized on less than 5 slices, which is particularly relevant for small targets such as brain metastases [53]. Partial volume effects because of large slice thickness will mostly result in an overestimation of the GTV volume. Moreover, thick slices and image gaps can also lead to an underestimation of tumor growth perpendicular to the imaging plane or miss small tumor parts [16, 53].

The most used 3D-T1w MR sequences for brain tumors have been T1w inversion-recovery gradient echo (IR-GE) sequences such as the T1-MPRAGE [54, 55]. For cranial radiotherapy target volume delineation, however, multiple studies have recently suggested that 3D-T1w fast or turbo spin echo (FSE/TSE) sequences may in fact be superior to 3D-T1w IR-GE sequences (Fig. 3; [55‐59]). Since 3D-T1w FSE/TSE sequences show less contrast between gray and white matter [55], they frequently improve the contrast ratio of contrast-enhancing lesions and low-contrasting brain parenchyma as background for targets. An equally beneficial effect for radiotherapy planning purposes is the suppression of vessels in 3D-T1w FSE/TSE sequences [58], which can facilitate discrimination of small tumors from transverse vessel cross-sections. For T1w IR-GE sequences, on the other hand, an additional well-described caveat is that they suffer from reduced visibility of enhancement when low contrast agent uptake is present, which could lead to an underestimation of the lesion boundaries [55, 60].

A further advantage of 3D-T1w FSE/TSE sequences over gradient echo sequences is that metal artefacts are significantly reduced, which is valuable when imaging patients with brain tumors who have shunts or surgical clips in place [23, 61]. However, 3D-T1w FSE/TSE sequences are more susceptible to motion compared to gradient echo sequences and may introduce artifacts due to the reliance on high turbo factors [62].

Despite, diagnostic imaging generally has been performed before MRI simulation, the SRT planning MRI may also need to address diagnostic uncertainty, as targets and patient anatomy may undergo important changes and new complications, like intratumoral bleeding, may arise [8, 63]. In some cases, diagnostic uncertainty may even be the leading factor affecting treatment precision. Neuroradiologic and radiologic sequence protocol recommendations addressing individual patient factors are therefore important to incorporate in addition to the following general guidance.

Additional critical parameters that can affect the delineation of lesions in T1-based MR sequences include the dose of gadolinium-based contrast agent (GBCA), as well as the time interval between contrast application and measurement [52, 71‐73]. In a study by Yuh et al., early imaging (10 min) and late imaging (20 min) following standard dose gadoteridol were compared [71]. After 10 min, 40.6% of metastases < 5 mm were visible, while 75.0% were visible after 20 min. Only after an additional bolus of double-dose gadoteridol, the remaining lesions were evident, demonstrating the importance of increasing the doses of GBCA [71]. Likewise, Baleriaux et al. found an increase in the number of metastases and a better visual conspicuity of the lesions with increasing cumulative doses of gadobenate dimeglumine over several sequential injections [72]. Similarly, Kushnisky et al. found more brain metastases at 15 min after GBCA administration compared to 5 min. They also notably found an increase in metastasis volume after 10 min compared to 5 min, as well as 15 min compared to 10 min postcontrast [73].

The interval between MRI and treatment delivery is one of the most crucial parameters for treatment precision ([7, 8]; Fig. 5). The importance of simulation MRI datasets being up-to-date is especially high for rapidly growing tumors such as brain metastases [8, 63] that are also often associated with fluctuating amounts of surrounding edema [85, 86].

This was confirmed by Seymour et al., who found a large detriment in local control following radiosurgery in brain metastases, if the interval between MRI and radiosurgery was greater than or equal to 14 days (local control 56% vs. 95% at 6 months post-SRS) [7]. Subsequently, Salkeld et al. even reported relevant differences with imaging intervals ≤ 7 days before SRS. In this study, 41% of patients with an interval ≤ 7 days required a change in radiooncologic management, whereas even 78% of patients required changes if the delay exceeded 7 days. Replanning was performed most frequently due to an increase in the tumor or resection cavity volume [8, 87]. In a retrospective review of 101 patients and 531 brain metastases, Kutuk et al. confirmed that changes in tumor size and spatial position occurred as a function of time, with the risk of requiring a margin beyond 1 mm increasing every day by 5% [88]. Therefore, the interval between imaging and treatment delivery should be as short as possible.

The simulation MRI for SRT is most frequently acquired in a diagnostic head coil and subsequently rigidly coregistered to the planning CT obtained in the treatment position with mask immobilization. Therefore, it is an essential prerequisite that the registration of the simulation MRI to the geometry of the planning CT in treatment position is as accurate as possible. In general, rigid registration that involves translations and rotations only is the optimal type of registration for cranial SRT. In case of multiple target volumes, multiple individual registrations in a reduced volume of interest covering each lesion separately should be considered [50]. The software used for image registration must be properly commissioned and validated followed by a quality assurance program for image registration e.g., according to AAPM Task Group report 132 [50]. This involves quantitative validation of registration accuracy with phantoms [50]. For cranial SRT, the employed rigid registration algorithm should have a registration error within 1 mm in registration phantoms [50, 92, 93]. Moreover, in daily clinical practice every MRI-CT registration for treatment planning must be verified qualitatively by a clinical expert with board certification. Software manufacturers developing treatment planning systems for cranial stereotactic radiotherapy are encouraged to offer a quantitative assessment of the registration method’s accuracy using an independent approach from the registration algorithm (e.g., employing an automatic landmark placement technique when the registration is carried out by optimizing the normalized mutual information). For the brain, in individual publications the accuracy of normalized-mutual-information based MRI-CT coregistration has been reported to be ≤ 0.5 mm in plane and ≤ 1 mm along the Z‑axis for CT slices with a thickness between 2 and 3 mm and a 2 mm 3D-T1w IR-GE sequence. However, the registration uncertainty increased by a factor of 2–3 when a 2D-SE sequence with a slice thickness of 5 mm was registered to a 5-mm planning CT [92]. More generally, registration accuracies of ~ 2 mm are reported [94, 95]. This finding highlights the significance of using high-resolution pairs of MRI and CT datasets to ensure adequate registration accuracy. MRI datasets used for registration should be 3D sequences, with equidistant slice planes and without gaps. MRI datasets must also be free from distortions, motion, folding, and metal artifacts and errors in planning CT datasets have to be minimized for optimal registration accuracy [50]. Due to the long scanning times with MRI, it is also important to exclude motion between individual sequences that are not addressed by registration. Many treatment planning systems automatically propagate the transformation from the registration of one MRI sequence to all other sequences in the study, which may lead to poor registration accuracy in case of intermediate motion. It is therefore important to individually verify the registration accuracy of every MRI series used for treatment planning.

Acquiring planning MRI datasets in the treatment position may improve the accuracy of rigid registration, as planning MRI and CT images are acquired in a similar anatomical configuration [30, 93, 96]. Multiple groups have reported imaging setups to acquire simulation MRIs in the treatment position with mask immobilization [24, 96, 97]. Flexible coil setups have been developed most frequently for this purpose, as most stereotactic mask systems will not fit into routine radiological head coils [30, 97‐99]. Masitho et al. evaluated the registration accuracy of three treatment planning systems for coregistering planning MRIs acquired in a diagnostic head coil and in radiotherapy treatment position with mask immobilization to a common planning CT. Generally, the software used for registration had the most significant impact on the accuracy of the registration. However, for less optimal registration methods, registration accuracy was significantly improved if simulation MRIs had been performed in the treatment position [93]. For targets near the foramen magnum, performing the planning MRI in treatment position could be more beneficial due to slight deformation occurring with different extension of the occipito-atlanto-axial joint complex [61]. In addition to slightly improving the accuracy of the registration, imaging in the treatment position with mask immobilization significantly decreases motion artifacts, which may help with the precise definition of the target volume [97, 99]. However, usually some decrease in image quality is expected from imaging with a flexible coil setup in treatment position as compared to using dedicated diagnostic head coils, which may need to be compensated with a longer measurement time [99]. Since flexible coils are more prone to mechanical damage, monthly quality assurance is recommended as proposed in the AAPM TG 284 report [10]. The procedure can be combined with validating geometric accuracy, gain, and spatial resolution which are part of quality assurance recommendations of the ACR [32].

Acquiring planning MRI studies in the treatment position with mask immobilization becomes a necessity when using an MR-only workflow [100]. Synthetic CT, involving the computation of synthetic CT images from one or multiple MR sequences, eliminates the requirement for an additional planning CT, thereby theoretically removing any registration uncertainties [101]. Current methods provide reasonable dosimetric results in most standard situations and dosimetric differences < 1% have been reported for deep learning methods recently [100, 102]. It has also been reported that sCTs provide adequate means for positioning the patient at the linear accelerator, i.e., as reference for registering the CBCT [103, 104] as well as 2D/2D based positioning systems [93]. However, when synthetic CT is employed in an MR-only workflow, care must be taken that the treatment position is exactly reproduced on the MR scanner as the synthetic CT will serve both as a foundation for dose calculation as well as a reference for image guidance at the radiotherapy treatment unit [100]. It is equally important to ensure that no changes in patient position occur between acquisition of the dedicated synthetic CT sequences and the MR sequences used for target delineation that are not addressed by registration. It is therefore recommended that prior to clinical implementation of an MR-only workflow, experience with MR simulation has been established, with optimized sequence selection and protocol parameters as well as by optimal quality assurance using the advanced recommendations as described above.

Accurate MRI simulation is a critical basis for precise treatment in cranial stereotactic radiotherapy and an integral part of stereotactic radiotherapy treatment planning. This report recommends measures and procedures to optimize MRI sequence protocols, to verify spatial precision and to optimize the clinical workflow to secure the fidelity and spatial precision of MR images in SRT. Broadly implementable minimum requirements provide major improvements in accuracy, while additional recommendations and options allow advanced centers to further optimize MR-based treatment planning. This guideline and especially the topics of repeated imaging during fractionated treatment, as well as the role of MR-only treatment planning, will require an update when more evidence becomes available.