**8. Voxel Sizes**

Clinical T2w images often incorporate anisotropic voxel sizes with large slice thickness in the z direction. This allows for higher in-plane resolution along the axial plane, which is primarily used for targeting (Figure 3) [145,149]. Voxel sizes will typically range between 0.45 × 0.45 × 2 mm and 1 × 1 × 3 mm. Lower resolution allows for shorter acquisition times of around 5 min, simultaneously limiting the e ffect of artifacts due to subject movement. However, anisotropic voxels su ffer from partial voluming e ffects (PVE), which refer to the blurring of signals across voxels, resulting in averaging di fferent tissue types and reducing specificity [165]. PVE are especially problematic for small structures such as the STN. Volume estimates are commonly used as an index of scan quality, and have shown consistent deviations of more than 50% from ground truths when slice thicknesses were three times the size of the alternate planes [166]. Moreover, anisotropic voxels will decrease the accuracy of resampling to super resolutions, which is an automatically incorporated step of pre-operative planning systems [167].

**Figure 3.** The effects of voxel geometry on the visualization of subcortical structures (adapted from [164]). Figure 3 shows clinical 3 T T2-weighted (T2w) with 0.45 × 0.45 × 2.0 mm voxel sizes, optimized 3 T T2\* with 1.0 × 1.0 × 1.0 mm voxel sizes, and 7 T T2\* maps with 0.64 × 0.64 × 0.7 mm voxel sizes. All images are acquired from a single subject and are shown at approximately the same anatomical level. The subthalamic nucleus (STN) and substantia nigra (SN) are shown at sagittal, coronal, and axial planes, with the red nucleus (RN) also highlighted in the axial plane. The anisotropic nature of the sagittal and coronal planes on the clinical 3 T do not allow for identification of any structure.

As spatial resolution is dependent on voxel size; smaller voxels should allow for more detailed and finer grained visualization of smaller structures. Voxel sizes can be reduced by increasing the acquisition matrix, reducing slice thickness, or decreasing the field of view. However, these factors can each negatively a ffect the SNR. The loss of SNR can be compensated by simply including more repetitions per sequence, which is an issue for PD populations as it necessitates an increase in acquisition time and requires the patient to be still. However this is often not possible for patients with movement disorders [166]. The loss of SNR caused by decreasing voxel sizes at lower fields can be counteracted through the use of UHF-MRI [130].

When targets in clinical MRI are verified with MER, the large slice thickness means that the spatial resolution is penalized along the *z*-axis. Therefore the depth of the electrode cannot be optimally planned and electrophysiological samplings are conducted to identify the ideal electrode placement [32,38,40]. This testing often requires that the patient is awake and endures behavioral assessments, which are stressful and physically demanding, prolong the time of the surgery, and can increase the risk of infection or hemorrhaging [168–170]. If smaller voxels can increase spatial resolution, three-dimensional anatomical accuracy, and tissue specificity, the requirement for intraoperative microelectrode recordings, multiple test electrode implantations, and awake behavioral testing could be eliminated, ultimately increasing patient comfort and reducing operation time.

However, voxels with a sub millimeter isotropic resolution used purely for identification of DBS targets, rather than for instance venous architecture, may not directly improve targeting accuracy. This is because the spatial resolution of stereotaxic coordinate systems is around 1.2 mm and chronically implanted conventional DBS electrodes are larger than 1 mm [171]. In addition, segmented DBS leads with directional steering may offer increased spatial resolution when recording local field potentials compared to traditional omnidirectional contacts [172,173]. Further, the development of microscale DBS contacts via multiresolution electrodes would allow for finer control of the stimulation volume and more precise targeting of smaller regions, matching the order and spatial resolution of submillimeter resolution MRI [174].
