Dorsal Root Ganglion Volumetry by MR Gangliography

DISCUSSION

We demonstrate that common methods for MR imaging–based DRG volumetry, which, until now, have been grounded on the assumption of simple ellipsoid geometry, substantially underestimate true DRG volume (observed factor of underestimation: 0.42–0.63). Moreover, we demonstrated how to optimize geometric equations for substantially more accurate DRG volume estimation.

Increasing attention to the DRG within the fields of pain research and pain medicine has been paralleled by methodologic progress in human DRG imaging. In particular, recent progress by structural and functional advances in MR gangliography stand out.^6⇓-8,16 The DRG microstructure consists of a cell body–rich area, which is oriented toward the DRG rim and is embedded in a uniquely dense microvascular network.¹⁷ With more central topography, a nerve fiber–rich area is located within the DRG.¹⁷ Pathophysiologic changes in 1 or both compartments are associated with DRG_vol change. DRG_vol is, therefore, an innovative and novel biomarker promising to become clinically and scientifically useful for a variety of diagnostic and research questions, particularly in the field of pain research, because most pain syndromes originate at the peripheral level of the nervous system.¹¹ In pain and somatosensory afferent signal processing, the DRG organ represents the interface between the peripheral and central nervous systems. Also in the context of degenerative or traumatic processes, the imaging observation of DRG_vol seems valuable because it correlates strongly with the number of DRG sensory neurons, allowing conclusions about local neurodegeneration and loss of neurons.¹⁸ For example, changes in the quantity of neurons could be observed in experimental diabetes models with a loss of large DRG neurons that, on a functional level, was associated with slowed sensory nerve conduction velocity.¹⁹ Another example in which the DRG_vol decrease reflects DRG neurodegeneration is Friedreich ataxia. In this disease, severe neurologic deficits are tightly associated with hypoplasia of the DRG neurons, which translates into decreased DRG_vol on the macrostructural level.²⁰

The imaging methodical key to substantially improve the accuracy of DRG morphometry by MR gangliography is to increase the resolution and contrast of the DRG and of its local vicinity using high-resolution isotropic voxels. The MR imaging signal is heavily T2-weighted with efficient fat suppression as an established contrast for MR gangliography and MR neurography.²¹ In this study, the principal protocol component of MR gangliography was an optimized 3D isotropic fast spin-echo sequence (SPACE). It improved the sharp-edged contrast between the DRG contour and surrounding tissue. It visualized more accurately the distinctive shape and contour of the DRG itself and, most important, differentiated it more accurately than possible before, in particular from the proximally and distally adjacent segments of the posterior nerve root.

In this study, the ground truth of DRG_vol (r) was obtained as a volume average from extensive voxelwise manual segmentation of 96 single DRGs across a total of 1152 slices by 3 independent expert raters. Interrater reliability among the 3 blinded raters was good²² to very good.²³ These ground truth DRG volumes confirm, in vivo and in humans, the previously reported ex vivo studies showing a continuous increase of DRG_vol toward the caudal region (L4–S1) with continuous decrease towards the sacral level.²⁴ The DRGs of levels L5 and S1 were significantly more voluminous compared with other vertebral heights. This finding represents the main reason why these 2 functionally important DRG levels can be proposed as index levels to investigate disease-related DRG_vol alterations. Finally, these findings were further validated in the largest human cohort so far observed for DRG volume estimation (510 DRGs).

To substantially optimize DRG_vol estimation, we tested several equations of ellipsoid body volume against ground truth. In the selection of equations, we applied the following rationale: Model-based approximation of disease target lesions or organ volumes is widely used to facilitate radiologic measurements, eg, for tumor volume estimation. The equation that is typically used for such purposes delivers volume estimation from standard diameter lengths and is typically represented by the term ( ). This term was first described by Kwak et al,¹³ and it has been verified by Lisk et al²⁵ and found wider establishment through the work of Kothari et al¹² as well as Kazui et al.²⁶ It represents a simplification of the complete term of ellipsoid volume, which is because π can be simply approximated to 3. In this pragmatic form, it is widely and easily used as a very simple tool for fast linear measurements that are practically available in virtually every PACS. In contrast, more accurate volume estimation requires fundamental understanding of the target structure shape, is not easily incorporated into PACS, and is considerably more time-consuming. However, more accurate volume estimation can also be relevant clinically as it has been shown by using the example of vestibular schwannoma volume estimation to detect changes earlier.^27⇓-29 Also for intracerebral hemorrhage, it was shown that simple volume approximations may substantially overestimate true hemorrhage volume by as much as 45%.³⁰

In the case of the DRG, so far, volumetry has been mainly derived from measurements of its maximum diameter in only 1 or 2 planes.^24,31 Only a few studies performed more detailed measurements in which maximum diameters^6,7 were determined in 3 orthogonal planes to estimate DRG_vol by the typical simplified ellipsoid equation: DRG_vol = , which corresponded to Equation 2 in this study. Equation 1 of this study represented the mathematically precise equation for a perfect ellipsoidal body ( ). Equations 2, 3, and 4 corresponded to different simplifications of this formula, which have been used in prior research.¹⁴ All of these 4 existing equations to estimate DRG_vol (e1–e4) led to considerable underestimation of DRG_vol in this study compared with ground truth DRG_vol (r) (observed factor of underestimation: 0.42–0.63).

According to a prior study on volumetry of the kidneys,³² we performed linear regression of DRG_vol (r) without and with the intercept and used the resulting regression equations to derive equations DRG_vol (e5) and DRG_vol (e6), which were substantially more accurate. The best approximation was achieved for DRG_vol (e6) = . It unifies both the rapidity of measurement and very high accuracy compared with the ground truth. Equation 6, with its additional intercept, also better accounts for nonellipsoid variation in DRG shape, such as the more tubular shape that regularly applies especially to smaller DRGs that are more closely aligned with the posterior nerve root. The performance of all tested equations relative to ground truth is visualized by Fig 2 using cut volume shells.

In the final part of our study, Equation 6, which provided the best approximation (cut purple versus green volume shell in Fig 3), was applied to the largest human DRG cohort so far investigated volumetrically, to the best of our knowledge. In addition, our study also analyzed nerve root length, measured from the origin of the dural sac to the onset of the ellipsoidal structure of the DRG, as well as their angulation relative to the DRG, showing a significant decrease in the root angles from lumbar to sacral. The length of the nerve root, on the contrary, increased continuously from lumbar to sacral (L4–S2). These results, for the most part, correspond closely to a cadaver study of DRG length and root angulation.³³

To be able to also observe more subtle therapy or disease-specific changes of DRG_vol in larger cohorts in the future, a more accurate assessment and normative reference values of DRG_vol are needed and are both provided by this study. Already in the recent past, MR gangliography could show promising results in the observation of disease-related volume change, for example in chemotoxic polyneuropathy, painful diabetic polyneuropathy, and Fabry disease.^7,9,10

Our study comes with the following limitations: Imaging coverage had to be limited to the lower spine. Our findings may not be transferrable to the cervical or thoracic spine where significant heart motion or lung- and breathing-related artifacts would have to be considered. At these other levels, other intercepts might have to be used to obtain accurate estimates given the slightly different geometric shape of these DRGs. Our observations were made in a large cohort of healthy subjects. Accuracy might be compromised if DRG shape is grossly altered by disease, which, however, seems unlikely, especially during the early disease course, which represents the typical stage of clinical-scientific focus.

Although rare, there remain anatomic variations such as arachnoid or dural cysts that may involve the proximal nerve roots and, in extreme cases, the DRG. These extreme-but-rare variations will continue to complicate the delineation of DRG boundaries in a few cases.

Finally, and obviously, full automation of DRG morphometry and volumetry need to be long-term goals. This seems achievable through the use of machine learning and/or artificial neural networks. Still, the application of equation-based 3D DRG volume estimation methods to larger cohorts as investigated here will certainly be needed as training datasets for this future purpose of automated image processing and analysis.