**5. SAR Limitations**

SAR refers to the amount of energy deposited into the body due to the radio frequency (RF) pulses applied with MRI sequences. RF pulses are emitted via electrical currents through coils, being used to generate the B1 field [74]. RF deposition can result in tissue heating, and to ensure that the absorbed energy does not induce local thermal damage, there are SAR limitations based on the region of interest, with the amount of SAR depending on tissue type [94,95]. However, field inhomogeneities increase with field strength, as the RF wavelength scales according to the size of the object being imaged, which then reduces its ability to penetrate the brain with a uniform power [96,97]. In the case of UHF-MRI, stronger gradients are required to magnetize tissues in the middle of the brain and to create a homogenous field, which results in higher SAR. Therefore, the safety limits are reached sooner at UHF than with lower field. Moreover, SAR can vary person to person due to individual differences in anatomy. This means that scan acquisition can require real-time parameter adaptation. Maintaining a low SAR can be achieved by increasing the repetition time (TR), reducing the flip angle (FA), or by reducing the number of acquired slices. Unfortunately, introducing these parameter changes to MR sequences can negatively affect the quality of the scan [98,99]. This invites an ethical debate as to whether future FDA-approved sequences and image pre-processing methods at UHF would allow for such real-time deviations in a clinical protocol where SAR limitations are reached and sequence amendments are required.

Further, there are more absolute and relative contraindications at UHF including pacemakers, surgical implants and prosthesis, and foreign bodies, even if they are not metallic or comprised of diamagnetic materials due to potential local heating and subsequent torque and increased SAR. Moreover, in our experience, many DBS candidates may not be scanned due to site-specific criteria. For instance, while a non-metallic or non-paramagnetic dental bridge is not listed as a contraindication, the guidelines for the 7 T site at some locations required such patients to be excluded. Even more contraindications exist at 7 T, including circulatory and clotting disorders, which makes UHF-MRI less compatible with a larger portion of the elderly population, including the majority of PD DBS patients [100]. Therefore, optimizing 3 T remains a viable option where UHF-MRI cannot be applied. However, while 3 T may theoretically be optimized to allow for increased visualization of subcortical nuclei, it is essential to remember that acquisition times will be much longer than that of an analogous 7 T sequence [24,101–103]; this concept will be discussed throughout the paper.

### **6. Shimming and Magnetic Field Corrections**

Shimming refers to the process of homogenizing either the main magnetic field (B0) or the radiofrequency field (B1). Inhomogeneity of the B0 field occurs when materials with different magnetic properties and susceptibility enter the bore, resulting in image distortion and signal loss. For example, the interface between brain tissue and air arising from the sinuses can cause artifacts within the frontal and temporal areas. These brain–air interface-induced artifacts can result in large shifts in the observed

anatomical locations of nearby brain structures and cortical surfaces [104]. While post-processing techniques exist to correct some of these erroneous signals, they cannot control for complete signal loss and dropout. Therefore, the field needs to be shimmed prior to the acquisition of the main MRI scan.

Shimming the B0 field can occur passively by strategically placing ferromagnetic sheets within the bore itself to form the distribution of the magnetic field toward a more uniform state [105] or by using patient-related inserts such as an intra-oral pyrolytic carbon plate [106]. This process is useful for removing field imperfections related to hardware, although is not generally utilized in clinical practice as it is laborious, inflexible, and temperature-dependent. More commonly, the field can be actively shimmed, which uses currents within the MRI system to generate corrective magnetic fields in areas showing inhomogeneous signals [105].

Active shimming is limited by the ability to model and reproduce the distortions that occur within the field. Shimming is generally based on the principles of spherical harmonics (SH), which use orthonormal equations to index changes in signal waveforms representative of field inhomogeneity. The mapping and the correction of the inhomogeneity is achieved by superimposing the magnetic field with an opposing corrective field equal to and a reversal of the polarity within a spatial distribution deemed erroneous by the SH coe fficients [107,108].

The order of SH is dependent on the number of dedicated current-driven coils. Traditional clinical and low-field MR systems will employ lower-order shimming methods mainly due to cost and space restraints [57]. Low-order shims primarily utilize linear terms including addition, scaling, and rotation of the SH coe fficients to model the magnetic field. Linear SH coe fficients function to resemble and compensate large-scale, shallow magnetic field components that can be corrected with a current o ffset applied with a standard gradient coil. This is typically achieved automatically with the use of a pre-scan B0 map. More local changes can be compensated for with dynamic shimming. However, this is most commonly used for multi-slice MR, which is prone to additional eddy current distortions and requires dedicated amplifier hardware. Further, the optimal shim method will depend on the desired contrast [109]. Ideally, each sequence should require an additional shim.

As field inhomogeneities increase with field strength, higher order harmonics are therefore required for UHF. Higher order SH allows for correcting more complex-shaped inhomogeneities by incorporating an additional non-linear quadratic field variation that allows for modelling the bending of curves in space. This requires supplementary dedicated shim coils, which can counter-intuitively induce additional distortions in the middle of the brain. Despite e fforts to harmonize parameters, shimming is often site- and field-dependent, and manual iterative shimming is not always possible due to time constraints and/or limited expertise.

Additional B1 mapping is essential for accurate quantitative measures of signal intensities within the correct geometric space. Inhomogeneous B1 fields can result in distorted flip angles (FAs). FAs index the amount of net magnetization rotation experienced during the application of an RF pulse. If FAs are incorrectly calculated, geometric distortions occur, which reduces the accuracy in T1 and T2 values. B1 mapping allows for the correction of FA values prior to acquiring a structural scan. Primary B1+ mapping methods can be incorporated into sequence acquisition. This is most commonly achieved with the double angle method (DAM), which estimates local FAs from the ratio of two images obtained with di fferent FA values. An additional 3D multi-shot method can be incorporated, which uses non-selective excitation to minimize inhomogeneous spin excitation across slices. Alternatively, spoiled gradient echo (GRE) sequences with variable FAs (VFA) and actual FA imaging (AFI) are commonly employed, which sample multiple T1 values to simulate signal di fferences across tissues [110–113].

Pre-processing of gradient non linearities (GNL) and intensity non-uniformity with retrospective image-based interpolation is also possible. Corrections for GNL are rarely accomplished in clinical settings but are commonplace for research-based applications. The magnitude of GNL increases with distance from the isocenter and can cause the visualization of structures to shift by up to 5 mm, which is detrimental for preoperative planning [114]. Correcting for GNL can be achieved by incorporating a low-pass filter to remove smooth spatially varying functions. Other GNL correction schemes include

surface fitting and feature matching that rely on intensity-based methods. Intensity-based methods assume that di fferent tissue intensities do not vary significantly unless they are subject to an erroneous bias field, where variations within one area can be corrected from the field of another spatial location within the image. Alternatively, histogram-based methods use a priori knowledge and manual input of known intensity and gradient probability distributions to correct images. B1 corrections can be achieved offline via image pre-processing steps with the FMRIB Software Library (FSL), Statistical Parametric Mapping (SPM), or Advanced Normalization Tools (ANTs) [115–118]. However, these methods must be considered experimental and their use in image correction for MRI in pre-operative planning is not currently FDA-approved.

### **7. Sequence Types and Contrasts**
