High Isotropic Resolution T2 Mapping of the Lumbosacral Plexus with T2-Prepared 3D Turbo Spin Echo

Magnetic resonance neurography (MRN) has shown to be a powerful diagnostic tool that enables the localization and characterization of nerve pathologies [1]. Especially isotropic high-resolution three-dimensional (3D) MRN has been repeatedly used in recent years to depict even small and highly oblique nerves of the brachial plexus or lumbosacral plexus (LSP) by allowing reformatting in multiple planes. In this context, 3D turbo spin echo (TSE) or 3D fast spin echo (FSE) are most commonly used for achieving isotropic high-resolution T2-weighted nerve imaging in or near a plexus [2‐4]. Furthermore, fat-suppressed 3D TSE imaging has recently been combined with the improved motion sensitized driven equilibrium (iMSDE) preparation to suppress vessel signal and improve plexus nerve delineation [5‐7].

Most previous studies using high-resolution 3D MRN of plexus nerves have focused on qualitative assessments of T2-weighted signal alterations while quantitative investigations are still mostly lacking [2‐6]. Although increasing efforts are being made to further improve image quality, evaluation of nerves solely based on qualitative imaging remains challenging. Instead, quantitative methods, such as T2 mapping, may provide an objective biomarker for monitoring nerve degeneration and regeneration; however, T2 quantification can be considerably affected by B0 and B1 inhomogeneities that are known to frequently occur in the LSP especially at 3 T. Thus, the present study combines 3D TSE imaging with an adiabatic T2 preparation in order to establish B1-insensitive high-resolution isotropic T2 mapping of LSP nerves. We evaluate the feasibility of this approach for extraction of quantitative T2 values from preselected LSP nerves, and hypothesize a high reproducibility and reliability of our newly introduced method.

The MRN was successfully performed three times in each subject (3 × 15 = 45 scans). None of the scans had to be excluded due to artifacts according to visual image quality assessment. Observer 1 evaluated three scans in each subject with respect to preganglionic, ganglionic, and postganglionic T2 values of S1, L5, and L4 nerves of both sides (3 × 15 × 3 × 3 × 2 = 810 T2 measurements). Of these measurements, 63 (7.8%) were excluded due to T2 values >100 ms (71.4% preganglionic, 28.6% ganglionic). Furthermore, observer 2 evaluated the first scan in each subject, again separately for preganglionic, ganglionic, and postganglionic T2 values of S1, L5, and L4 nerves of both sides (1 × 15 × 3 × 3 × 2 = 270 T2 measurements). Also 32 measurements (11.9%) were removed due to T2 values >100 ms (37.5% preganglionic, 62.5% ganglionic). Mean T2 values (right-sided and left-sided T2 measurements collapsed) according to evaluation of the three scans per subject by observer 1 can be found in Table 1, whereas the mean T2 values of observer 2 are displayed in Table 2. Regarding reproducibility, excellent agreement was observed for all measurements, with ICC values ranging between 0.89 (L5 postganglionic) and 0.99 (S1 preganglionic; Table 1). The corresponding CVs ranged between 1.48% (L4 postganglionic) and 2.57% (L5 postganglionic; Table 1). Concerning interobserver reliability, good to excellent agreement was observed for all measurements of S1, whereas fair agreement was observed for L5 and poor to good agreement was present for measurements of L4 (Table 2). The interobserver ICCs ranged between 0.18 (L4 postganglionic) and 0.78 (S1 preganglionic) while the CVs ranged between 3.18% (S1 ganglionic) and 8.89% (L4 preganglionic; Table 2).

When averaging the T2 values derived from three scans measured by observer 1, S1 showed values of 74.5 ± 6.4 ms (preganglionic; range: 64.0–82.5 ms), 86.3 ± 6.2 ms (ganglionic; range: 77.7–97.6 ms), and 65.0 ± 3.0 ms (postganglionic; range: 59.6–71.0 ms), with statistically significant differences between these measurements (preganglionic vs. ganglionic: p = 0.0240; pre- vs. postganglionic: p = 0.0320, ganglionic vs. postganglionic: p < 0.0001; Fig. 3). For L5, T2 values of 75.5 ± 5.1 ms (preganglionic; range: 69.2–87.2 ms), 85.8 ± 5.6 ms (ganglionic; range: 78.0–99.8 ms), and 66.5 ± 3.0 ms (postganglionic; range: 61.0–72.0 ms) were revealed, and statistically significant differences between measurements were observed (preganglionic vs. ganglionic: p = 0.0346; pre- vs. postganglionic: p = 0.0072, ganglionic vs. postganglionic: p < 0.0001; Fig. 3). Moreover, T2 values for L4 were 72.5 ± 4.7 ms (preganglionic; range: 65.0–81.0 ms), 82.0 ± 4.2 ms (ganglionic; range: 76.6–92.8 ms), and 64.7 ± 2.4 ms (postganglionic; range: 60.8–69.6 ms), with statistically significant differences between these T2 values (preganglionic vs. ganglionic: p = 0.0152; pre- vs. postganglionic: p = 0.0191, ganglionic vs. postganglionic: p < 0.0001; Fig. 3).

The LSP area is commonly regarded as challenging in terms of MRN due to the need for high isotropic resolution to depict the complex oblique geometry of the nerves. It has been demonstrated recently that 3D TSE or 3D FSE combined with iMSDE preparation can achieve this high isotropic resolution and even resolve small nerve branches [2‐6]; however, high-resolution 3D MRN of the LSP area can be significantly affected by B0 and B1 inhomogeneities that frequently occur in this region. The sequence used for T2 mapping of LSP nerves in this study is similar to a recently developed sequence that was successfully applied to T2 mapping of thigh muscles in healthy subjects [8]. When comparing the new T2-prepared 3D TSE sequence with standard multi-echo spin echo (MESE) sequences, a dependency of T2 values of MESE sequences on the transmit B1 field was shown, whereas the T2-prepared 3D TSE sequence was clearly less affected by the transmit B1 field [8]. Consequently, the used T2-prepared 3D TSE sequence enables B1-insensitive T2 quantification, thus tackling the issue of B1 sensitivity of T2 mapping, which is known to affect robust quantitative T2 measurements especially at 3 T (supplementary material).

The B1-insensitive T2-prepared 3D TSE sequence, combined with the presented approach of placing manual ROIs in reformatted images, allowed systematic measurements of T2 values of different LSP nerves and their segments. Although a limited amount of values had to be excluded due to measurements in areas with presumably high fractions of CSF or surrounding vessels (T2 >100 ms; observer 1: 7.8% excluded, observer 2: 11.9% excluded), the dataset analyzed remains considerably large, and the potential usefulness of our approach is reflected by excellent reproducibility. In this context, ICCs of at least 0.89 for all analyzed nerves and segments were observed (Table 1). Concerning the evaluation of interobserver reliability, the obtained ICCs were less convincing when comparing a more experienced to a relatively unexperienced observer, but still indicated fair to excellent agreement except for the L4 measurements (Table 2); however, the CVs calculated were considerably low, which was even true for the L4 (Table 2). The discrepancy between reproducibility and interobserver ICCs might be explained by potential difficulties in identification of anatomical structures during ROI placement since we compared a more experienced to a novice investigator. Moreover, manual placement of ROIs principally harbors the risk that two observers do not perform entirely equally, and small differences in placement may already lead to differences between investigators in a complex region such as the LSP.

In addition to evaluations of reproducibility and interobserver reliability, this study provides normative data on T2 values of three nerves of the LSP, separated into different segments (Table 1 and 2; Fig. 3). In this context, the pattern of values observed for these structures is comparable to a previous investigation using in-plane imaging, although the mean T2 measurements seem to be slightly higher in the present study [12]; however, literature on extraction of quantitative measures of LSP nerves is still limited, and application of MRN in this region is predominantly restricted to qualitative imaging without routine extraction of quantitative parameters. Regarding quantitative approaches, diffusion imaging has been used to outline normative values and microstructural organization of LSP nerves [12‐14]. Furthermore, an iMSDE-prepared sequence for 3D imaging has been used to measure sciatic and femoral nerve angles and diameters in healthy subjects, and the obtained values might serve as normative data for comparison to patients with pathologies affecting the LSP [6].

From a clinical perspective, normative data derived from such 3D MRN may be used for identification and quantification of pathologies of LSP nerves. Frequent pathologies for which quantitative measurements could be regarded as helpful during diagnostics and monitoring might be demyelinating polyneuropathies or structural abnormalities leading to compression of nerves, such as disc protrusion or spinal stenosis [6, 15, 16]. Additionally, adding quantitative measurements to mere nerve visualization during high-resolution MRN might further enhance guidance of interventional nerve blocks and perineural injections. So far, promising results have already been achieved for qualitative MRN, but investigations on combined qualitative and quantitative high-resolution 3D MRN for this purpose have not yet been carried out [17].

Although our study adds valuable data to the limited amount of literature on high-resolution 3D MRN of LSP nerves, we have to bear some limitations in mind. First, no combination with diffusion imaging was established, which might have enhanced correct identification of nerves and accuracy, as suggested by investigations on peripheral nerve pathologies [18]. Second, we only enrolled healthy subjects in this study to establish a new quantitative MRN approach and to provide normative T2 values for clinical purposes; therefore, it has to be evaluated in further studies whether patients with lesions of LSP nerves show significantly altered measurements when compared to data derived from healthy subjects. Third, measurements leading to T2 values >100 ms were systematically removed from the dataset, but it is not clear whether definition of a rather arbitrary threshold is justifiable in later clinical use. Although only a limited fraction of measurements was removed due to this threshold, further studies are needed to refine the approach presented in this work; however, all removed values belonged to measurements in preganglionic or ganglionic segments while none of the postganglionic measurements had to be excluded. When our approach is applied among patients in future studies to evaluate differences in T2 mapping between the affected and non-affected side (e.g. nerves affected by compression or inflammation), postganglionic measurements that are performed distant to the spinal cord and, thus, distant to prominent CSF contamination might already be sufficient for clinical diagnosis. Fourth, partial volume effects may still be present in the manual drawing of the ROIs in the axial reformats; however, the present analysis with the reformatting of the 3D isotropic data should reduce partial volume effects compared to previous quantitative LSP imaging works, which have been traditionally based on straight axial two-dimensional imaging. Fifth, alterations of the T2 values within peripheral nerves could also in part be induced by the dependency of the T2-weighted signal intensity on the orientation of the nerve relative to the main magnetic field (so-called magic angle effect) [19]; however, the elevation of T2 values due to this effect should be small when the axis of a respective nerve forms a small angle with the main magnetic field [20]. This should be the case for the investigated nerve segments close to the spinal cord; however, the magic angle effect should be considered in further analysis. Finally, the present T2 preparation did not use flow or diffusion encoding gradients, thus giving rise to potential signal contamination by surrounding CSF and vessels. The T2 preparation modules with flow or diffusion encoding gradients could be used in the future to remove CSF and vessel signal contamination [5, 7, 21].

The combination of 3D TSE imaging and adiabatic T2 preparation (T2-prepared 3D TSE) enables B1-insensitive and isotropic high-resolution T2 mapping of LSP nerves. This approach allowed measurements of the spatial variation of T2 values of different nerves (S1, L5, and L4) within different segments (preganglionic, ganglionic, and postganglionic) in healthy subjects, thus providing normative data with excellent reproducibility. The T2 mapping approach using T2-prepared 3D TSE might reflect a robust method for future diagnostics of LSP pathologies.