Magnetic Resonance Neurography Visualizes Abnormalities in Sciatic and Tibial Nerves in Patients With Type 1 Diabetes and Neuropathy

Diabetic peripheral neuropathy (DPN) is a common complication that often remains undiagnosed until later stages. DPN causes irreversible damage to the peripheral nerves. Thus, early diagnosis is important in order to prevent progression of DPN, emphasizing the need for more sensitive diagnostic techniques.

A DPN diagnosis is established based on a neurological examination, nerve conduction studies (NCSs), and quantitative sensory testing. DPN is associated with structural changes of the peripheral nerves, including endoneurial microangiopathy (1), abnormal Schwann cells (2), axonal degeneration (3), and paranodal demyelination, all conditions leading to loss of myelinated and unmyelinated fibers (4).

Peripheral nerve lesions can be visualized by ultrasonography (5–10) and magnetic resonance neurography (MRN) (11–15). MRN enables microstructural imaging of peripheral nerves at the anatomical level of the nerve fascicles. High-field clinical scanners (3-T) and proton spin density (PD) or T2-weighted imaging sequences with fat suppression have shown an increased magnetic resonance (MR) signal in a variety of focal and nonfocal neuropathies and polyneuropathies (11–13,16).

Inclusion of diffusion tensor imaging (DTI) may improve the diagnostic accuracy of MRN in the ulnar and median nerves (17,18). Furthermore, in a previous study of patients with severe DPN (sDPN), we found that DTI is a highly reproducible method that can detect sDPN in patients with type 1 diabetes (19). In the same pilot study, we found that the quantitative target measures of DTI-MRN more accurately separated groups than did quantitatively evaluating changes based on T2 or PD contrast. This suggests that DTI may be the most sensitive noninvasive imaging method to detect microstructural alterations of peripheral nerves in DPN.

The aim of the current study was to evaluate whether MRN of the sciatic and tibial nerves can be used to detect and stage DPN in patients with type 1 diabetes. Furthermore, applying receiver operating characteristic (ROC) analyses, we aimed to determine the sensitivity and specificity of the MR methodology.

This study was approved by the local ethics committee (no. 37251) and registered at www.clinicaltrials.gov (identifier NCT01847937). All study participants gave informed consent.

Through public announcements we recruited 49 patients with type 1 diabetes from the Department of Endocrinology and Internal Medicine (Aarhus University Hospital) and 30 healthy control subjects (HCs). All participants were examined between September 2013 and May 2016. Findings from 21 of these patients and 10 HCs were reported in a previous feasibility study evaluating whether DTI enabled detection of DPN (19). That study included only patients with sDPN or with no neuropathy (nDPN). We now extend these findings by evaluating a larger cohort including patients with milder DPN. All subjects included were aged 60–71 years.

The minimal criteria (20,21) determined the presence of diabetic neuropathy. Presence of neuropathy is determined by at least two abnormal findings in the following four categories (one of these categories had to be abnormal vibratory perception threshold [VPT] or nerve conduction study [NCS]): 1) abnormal VPT (≥98th percentile) at the index finger and great toe; 2) abnormal nerve recordings (more than two) from NCSs; 3) a neuropathy symptom score (NSS) ≥1; and 4) a neurological impairment score (NIS) ≥7. Patients with diabetes were subsequently divided into three groups: nDPN, mild neuropathy (mDPN) (NIS <24), and sDPN (NIS ≥24).

Exclusion criteria were severe cardiac or lung disease, acute or chronic musculoskeletal disorders, acute metabolic dysregulation, other neurological or endocrine disorders, any previous or current asymmetric, proximal lower-limb weakness, and contraindications to MRI. HCs completed the Michigan Neuropathy Screening Instrument; this allowed us to exclude subjects with any symptoms of neuropathy and/or diabetes (22).

Blood samples were collected to measure HbA_1c (millimoles per mole) using standard laboratory methods. Plasma blood glucose was measured using a handheld glucose meter (FreeStyle Lite; Abbot Diabetes Care, Copenhagen, Denmark).

Clinical examinations were performed by a trained neurologist (H.A.) using the NIS (23) and the NSS. The NIS is a combined score obtained from a neurological examination of muscle strength, tendon reflex activity, and sensation at the great toe and index finger. The NSS evaluates motor, sensory, and autonomic symptoms of neuropathy.

VPT were measured at the distal part of the hallux and index finger using the 4-2-1 stepping algorithm (CASE IV; WR Medical Electronics Co., Stillwater, MN) (24,25).

NCSs were performed with conventional surface electrode techniques using electromyography equipment (Dantec Keypoint Focus EMG version 2.11; Natus Medical, San Carlos, CA); skin temperature was ≥32°C during these studies. Results were compared with those from laboratory controls. Motor NCSs were performed in the median, peroneal, and tibial nerves, and nerve conduction velocities (NCVs) and compound muscle action potential (CMAP) amplitudes were determined. Sensory NCV and sensory nerve action potential amplitudes were determined in the median and sural nerves. The peroneal, tibial, and sural nerves were examined bilaterally. The motor and sensory ulnar nerves were studied in persons with carpal tunnel syndrome.

To define the severity of neuropathy in each patient, we calculated a neuropathy rank sum score (NRSS) based on the individual rank scores from the NIS, NSS, VPT measurements, and NCSs.

MR examinations were performed using a 3-T MR scanner (Skyra; Siemens AG, Erlangen, Germany) with a 15-channel transmit/receive knee coil (Siemens AG). MRIs were acquired at predetermined locations along the left leg, including the distal thigh (10% of the distance from the upper part of the patella to the trochanter major) and the midcalf (50% of the distance from the lateral malleolus to the lower point of the patella). At the proximal level (i.e., distal thigh), the entire sciatic nerve cross section was included as a region of interest. This location is referred to throughout this article as “sciatic,” whereas the level at the calf where the region of interest comprised only the fascicles of the tibial nerve is referred to as “tibial.”

The MR protocol was performed unilaterally and consisted of spin-echo (SE) images with 10 different echo times and diffusion-weighted images to calculate diffusion parameters (mean apparent diffusion coefficient [ADC], fractional anisotropy [FA], and trace). ADC and the FA images were analyzed with nerve segmentation from the trace images (Fig. 1). MRIs were acquired from the following pulse sequences:

• 1. Axial, multi-SE, 2-dimensional spectral adiabatic inversion recovery sequence with a strong fat suppression pulse: repetition time (TR) = 3,280 ms; echo time (TE)_1–10 = 13, 25, 38, 51, 63, 76, 89, 101, 114, and 127 ms; field of view 160 × 160 mm²; matrix size 512 × 512; slice thickness 3 mm; voxel size 0.3 × 0.3 × 3 mm³; no interslice gap, 2 averages, and 16 slices; scan time 22 min 10 s

• 2. Axial diffusion-weighted, SE, 2-dimensional echo-planar imaging sequence with a strong fat suppression pulse: TR = 4,200 ms; TE = 112 ms; b = 0 and 800 s/mm²; directions = 12; field of view 175 × 175 mm²; matrix size 128 × 128; slice thickness 3 mm; voxel size 1.36 × 1.36 × 3 mm³; no interslice gap, 4 averages, and 16 slices; scan time 3 min 30 s

The net imaging time was 26 min 10 s at each location, with the inclusion of an anatomical localizer scan (30 s) at each location, and a coil reposition requiring an additional 5 min.

Structural fiber loss is conceived as the dominant histological alteration of DPN and is typically most severe distally (26); it has frequently been evaluated in sural nerve biopsies obtained at the ankle (27). To test a possible longitudinal gradient of fiber loss, we evaluated the ratio of the sciatic (proximal) level versus the tibial (distal) level.

The cross-sectional area (CSA) of the segmented nerves in the multi-SE images was used to evaluate the nerve caliber of the sciatic and tibial nerves. The multi-SE images had a higher resolution and signal-to-noise ratio than images from DTI, providing improved structural visualization.

FSL was used to process and analyze images (FMRIB Software Library, Oxford, U.K.) (28). Nerve lesions were determined based on quantitative analyses of signal intensities of the sciatic and tibial nerves. Nerves in the MRIs were segmented using the median TE (63 ms), providing a segmentation mask for the remaining multi-SE images. T2 relaxation-time (T2) and PD were calculated using a monoexponential curve-fitting algorithm applied to the nerve signal intensities of the multi-SE images (Eq. 1):

where k is the signal gain produced from the scanner and s₀ is the true PD. However, because k is difficult to separate from the true PD, k and s₀ are combined and used as PD in this study. Monoexponential parameters were calculated from the 10 different TE values acquired from the multi-SE sequence using MATLAB 2014a (MathWorks Inc.).

The Student t test was applied for pairwise comparison between single groups, and one-way ANOVA was applied to determine statistical differences between groups. Statistical significance was defined as a two-tailed P value <0.05. Linear regression analyses were performed to evaluate an association between the degree of neuropathy and the DTI parameters. The goodness of fit of the linear approximation was determined using the coefficient of determination, described as the R² value. ROC analyses and area under the curve (AUC) of the FA and ADC values were calculated and then categorized according to the following predefined AUC thresholds: 1.0–0.90, excellent; 0.90–0.80, good; 0.80–0.70, fair; 0.70–0.60, poor; and 0.60–0.50, fail (29). Statistical analyses were performed using STATA 13.1 (StataCorp LP, College Station, TX).

The clinical and demographic results are presented in Table 1.

Pairwise comparisons were calculated for DTI parameters of the sciatic and tibial nerves (Fig. 2). Significant differences were found between groups, indicating that FA values decrease and ADC values increase according to severity of neuropathy in the groups (P < 0.01). No difference was observed when comparing patients with no neuropathy with HCs.

To determine the AUC, sensitivity, and specificity of DTI in order to separate diabetic groups, ROC curves were calculated in three analyses:

• 1. Patients with mDPN compared with patients with nDPN

• 2. Patients with sDPN compared with patients with mDPN

• 3. Patients with sDPN compared with patients with nDPN

ROC curves were calculated for FA and ADC values at the sciatic and tibial nerves (Fig. 3). On the basis of the FA values, there was a good separation between groups for the sciatic nerve (AUC 0.60–0.95) and the tibial nerve (AUC 0.69–0.90). Corresponding to this, for ADC values there was also a good separation between the groups for the sciatic nerve (AUC 0.63–0.70) and the tibial nerve (AUC 0.59–0.78).

Close correlations could be established between DTI parameters and severity of neuropathy (NRSS) in the tibial and sciatic nerves (FA: R² = 0.32 and 0.49, respectively; ADC: R² = 0.15 and 0.19, respectively) (Fig. 4).

DTI parameters were related to the amplitude of the CMAP of the sciatic and tibial nerves (FA: R² = 0.17 and 0.24, respectively; ADC: R² = 0.27 and 0.04, respectively) and the NCV of the sciatic and tibial nerves (FA: R² = 0.18 and 0.37, respectively; ADC: R² = 0.31 and 0.22, respectively) (Table 2). Furthermore, we evaluated the relations between DTI parameters and the NIS (FA: R² = 0.33 and 0.28; ADC: R² = 0.14 and 0.07) (Table 2).

The ratio between DTI findings from the distal (tibial) nerve and the proximal (sciatic) nerve was similar between the groups, as indicated by the FA ratio (HCs: 0.87 ± 0.12; nDPN group: 0.87 ± 0.14; mDPN group: 0.85 ± 0.15; sDPN group: 0.81 ± 0.16) (P = 0.50) and the ADC ratio (HCs: 1.04 ± 0.15; nDPN group: 1.04 ± 0.12; mDPN group: 1.08 ± 0.24; sDPN group: 1.11 ± 0.14) (P = 0.67).

However, a statistically significant difference was found in the distal-to-proximal gradient of the FA values between the three groups; this was found only for the ADC values in the sDPN group (P = 0.05) (Table 3).

T2 and PD of the sciatic and tibial nerves showed no differences between groups (Table 4). In pairwise analyses of the PD, we found a difference between the nDPN and sDPN groups for the sciatic nerve (P = 0.03); a similar difference was found between these same groups for the tibial nerve (P = 0.03).

CSAs in the sciatic and tibial nerves were different between groups (Table 4). Paired sample t tests revealed that the HCs had smaller sciatic nerve CSAs compared with all groups of patients with diabetes (P = 0.01). For the tibial nerve, the CSA was different when comparing the nDPN and mDPN groups (P = 0.01) and for the HCs compared with the mDPN group (P = 0.04).

We established that MRN is able to detect in patients with type 1 diabetes nerve abnormalities that are closely related to the severity of neuropathy, suggesting that MRN can be used to detect structural signs of neuropathy. In this study we extended previous studies investigating MRN in DPN (11,12) by exclusively evaluating DPN in patients with type 1 diabetes and, importantly, by incorporating advanced MRN with DTI, to our knowledge for the first time. Our main finding is the superior diagnostic accuracy of quantitative DTI over PD or T2 values. In line with previous studies, we found that the structural nerve differences as visualized by imaging appear most marked at the proximal rather than the distal level.

In the Toronto criteria, NCS remains the gold standard for diagnosing and grading DPN in clinical research, and new diagnostic methods should be compared with and related to findings of NCSs (30). We established that MRN and NCSs were closely associated, suggesting that MRN reflects the pathophysiological process of DPN in type 1 diabetes. In a previous study we established that the DTI-MRN techniques applied are highly reproducible and reliable (19).

In this study, MRI included DTI (FA and ADC) and multi-SE imaging. FA and ADC describe the restriction of water molecule diffusion in three dimensions. FA reflects restrictions of spatial movement and ADC reflects the diffusion speed of the water molecules. In peripheral neuropathies, loss of axons and myelin leads to less constriction of endoneurial flow along the nerves (31); this could explain the changes in FA and ADC. Experimental studies have shown that FA correlates to nerve fiber density (32–36). ADC changes in relation to membrane, myelin sheath, cell wall, macromolecule, and viscosity alterations in the fluid containing low levels of protein flowing along the nerve fibers (31). In DPN, axonal loss is considered the most prevalent pathological finding (3), suggesting that FA is a good measure of DPN.

The presence and severity of neuropathy were determined from NCSs, quantitative sensory examinations, and clinical examinations. Pairwise comparisons between the groups with diabetes and the HCs showed that FA and ADC values of the sciatic and tibial nerves had the highest discriminatory power between nDPN, subtle DPN, and sDPN. The FA values differed significantly between groups, and even mDPN could be separated from nDPN. Interestingly, this difference was more pronounced at the proximal/sciatic level than at the distal level, which is also reflected by the AUC of the ROC analyses. The proximal dominance of structural alteration in this study is consistent with previous studies showing predominantly proximal structural nerve injury not only in DPN but also in other polyneuropathies with similar distal symmetric symptoms (12,13). Patients with diabetes without neuropathy had ADC and FA values similar to those in HCs, indicating that the abnormal FA and ADC values reflect neuropathic abnormalities rather than an effect of diabetes per se. On the basis of the AUC from the FA and ADC values obtained at the proximal/sciatic level (Fig. 3), FA had the best discriminatory performance compared with ADC.

Linear regression analyses demonstrated close associations between the severity of neuropathy and the FA values of the sciatic and tibial nerves. The association was less pronounced in relation to the ADC values.

Furthermore, FA and ADC of the sciatic and tibial nerves showed good associations with NCV and CMAP, with the closest correlations for the FA values of the sciatic nerve. Evaluating the relation between DTI and NIS, FA was more closely related than ADC.

DTI parameters have been used to evaluate axonal and myelin sheath integrity in the median nerve in healthy subjects (37). These results were validated in part by indicating that FA and ADC correlated with the integrity of the myelin sheath identifiers (NCV) and axial diffusivity with fiber density/axonal integrity (CMAP) (18,38). In another study of patients with both axonal and demyelinating neuropathies, the FA of the sciatic nerve was closely related to CMAP (39). Thus, in this and in previous studies, FA enabled more sensitive and accurate detection of peripheral nerve abnormalities than ADC (40), irrespective of neuropathic severity.

Distal-to-proximal loss of axons was evaluated between and within groups based on the DTI parameters. We found a within-group difference in the proximal and distal FA values, in line with earlier findings (37). This underlines the importance of using predefined anatomical scanning locations.

In our study, the CSA of the sciatic nerve was lower in HCs compared with patients with diabetes. For the tibial nerve, patients with mDPN had a larger CSA than patients with nDPN; however, patients with sDPN did not have a larger CSA. This is in contrast to findings using ultrasonography, where enlarged nerves occurred distally in advanced stages of neuropathy (5). The resolution used for segmentation of the CSA was 0.3 × 0.3 mm²/pixel. Because the average CSA of the tibial nerve is 7 mm², >75 pixels are used for segmentation (225 pixels in the sciatic nerve). This suggests that the different findings using ultrasound and MRI cannot be explained by a lower resolution of MRI. Our finding is consistent with previous studies showing that an increase in CSA occurs predominantly at the proximal level, in addition to DTI and signal alterations (12,13). Ultrasonography is excellent for high-resolution imaging of superficial peripheral nerves; however, it does not enable visualization of deeply situated nerves, nerves surrounded by fat (sciatic nerve), or nerves beneath bones (because of acoustic artifacts) (41,42).

Previous studies have shown that MRN is a more reliable measure than ultrasonography for use to visualize lesions (43). Nevertheless, because MRN is more expensive and time consuming, ultrasonography may be a more feasible method.

To detect DPN, ultrasonography is commonly used to determine CSA; however, hypoechogenicity and maximum thickness of nerve fascicles have also been evaluated (6,8). These studies indicated a correlation between NCSs and ultrasound findings. However, other studies have not been able to find a correlation between CSA and findings from NCSs (7,44). Ultrasonography may enable early detection of subclinical nerve degeneration, but because of the higher resolution, MRN may also be relevant.

In line with previous studies, the T2 value had low discriminatory power. Interestingly, PD was inferior to DTI for group discrimination. The relation between abnormalities observed on DTI-MRN and the pathological process of neuropathy remains unclear. Hyperglycemia leads to cellular nerve damage through mitochondrial overload, polyol pathway–induced oxidative stress, and inflammatory injury (45,46). T2-weighted images are sensitive to edema and fat. We applied a strong fat saturation pulse to remove the epineural fat signal adjacent to the nerve fascicles, causing fat and connective tissue to appear dark in the MRIs. In patients with diabetes, severely damaged nerve tissue is replaced by connective tissue (47) and therefore appears dark in SE images. This could explain the absence of T2 changes in patients with sDPN.

In a previous study using a mixed group of participants with type 1 or type 2 diabetes, the PD signal was higher in patients with both mDPN and sDPN (12). In that study, T2-weighted and PD-weighted imaging were applied, but with a normal fat saturation pulse over a larger area. This approach, when compared with our study, which includes only patients with type 1 diabetes, might explain the difference in findings of PD. Therefore, a study evaluating MRN in type 2 diabetes is necessary in order to evaluate possible differences between the two types of diabetes.

Axonal loss would not result in altered signal intensity of the T2-weighted images (12,48). FA is closely associated with fiber density (32–34), which may explain why FA values are more closely related to the severity of neuropathy compared with T2 and PD.

Our study has several limitations. First, this is a cross-sectional study, and thus how DTI findings develop over time remains unknown. Second, axial and radial diffusivities were not calculated in our study, which in HCs provided additional information about axonal and myelin sheath integrity (37). Third, MRN coverage of the sciatic and tibial nerves consisted of 16 slices (2 × 4.80 cm), evaluating only small parts of the nerve. MRN covering the entire nerve would enable detection of multifocal lesions in DPN; however, this would considerably increase examination time. Also, we did not include the upper-limb nerve to serve as a control for the lower limb, which could have further substantiated our findings; this would, however, also increase the examination time. Finally, the study did not include an assessment of peripheral limb vascular status to evaluate any influence of multifocal ischemic neuropathy on the DTI-MRN findings.

In conclusion, we found close associations between DTI-MRN findings and the presence and severity of neuropathy in proximal and distal nerve segments of patients with type 1 diabetes. DTI-MRN is a noninvasive, quantitative method that may be used to detect and monitor neuropathic processes in DPN.

Acknowledgments. The authors thank Søren Gregersen, Department of Endocrinology and Internal Medicine, Aarhus University Hospital, for helping with patient recruitment.

Funding. This work is funded by the UNIK partnership foundation, Siemens A/G Copenhagen, Aarhus University, the BEVICA Foundation, and the Danish Diabetes Academy, which is supported by the Novo Nordisk Foundation. M.P. (SFB 1158, Project A3) and S.H. (SFB 1118 Project, B05) were supported by the Deutsche Forschungsgemeinschaft.

Duality of Interest. No conflicts of interest relevant to this article were reported.

Author Contributions. M.V., M.P., N.E., S.H., and H.A. designed the study. M.V. and P.L.P. recruited the patients. M.V. and H.A. examined the patients. M.V., S.R., H.T., and H.A. performed the research. M.V. analyzed the data and wrote the manuscript. M.V. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.