Magnetic resonance neurography detects diabetic neuropathy early and with Proximal Predominance

Abstract

Objective: The aim of this work was to localize and quantify alterations of nerve microstructure in diabetic polyneuropathy (DPN) by magnetic resonance (MR) neurography with large anatomical coverage.

Methods: Patients (N = 25) with mild-to-moderate (Neuropathy-Symptom-Score [NSS]/Neuropathy Deficit Score [NDS] 3.8 ± 0.3/2.6 ± 0.5) and patients (n = 10) with severe DPN (6.2 ± 0.6/7.4 ± 0.5) were compared to patients (n = 15) with diabetes but no DPN and to age-/sex-matched nondiabetic controls (n = 25). All subjects underwent MR neurography with large spatial coverage and high resolution from spinal nerve to ankle level: four slabs per leg, each with 35 axial slices (T2- and proton-density-weighted two dimensional turbo-spin-echo sequences; voxel size: 0.4 × 0.3 × 3.5 mm(3) ) and a three-dimensional T2-weighted sequence to cover spinal nerves and plexus. Nerve segmentation was performed on a total of 280 slices per subject. Nerve lesion voxels were determined independently from operator input by statistical classification against the nondiabetic cohort. At the site with highest lesion-voxel burden, signal quantification was performed by calculating nerve proton spin density and T2 relaxation time.

Results: Total burden of nerve lesion voxels was significantly increased in DPN (p = 0.003) with strong spatial predominance at thigh level, where average lesion voxel load was significantly higher in severe (57 ± 18.4; p = 0.0022) and in mild-to-moderate DPN (35 ± 4.0; p < 0.001) than in controls (18 ± 3.6). Signal quantification at the site of predominant lesion burden (thigh) revealed a significant increase of nerve proton spin density in severe (360 ± 22.9; p = 0.043) and in mild-to-moderate DPN (365 ± 15.2; p = 0.001) versus controls (288 ± 13.4), but not of T2 relaxation time (p = 0.49). Nerve proton spin density predicted severity of DPN with an odds ratio of 2.9 (95% confidence interval: 2.4-3.5; p < 0.001) per 100 proton spins.

Interpretation: In DPN, the predominant site of microstructural nerve alteration is at the thigh level with a strong proximal-to-distal gradient. Nerve proton spin density at the thigh level is a novel quantitative imaging biomarker of early DPN and increases with neuropathy severity.

Figure 1

Array of representative source images indicating segmented tibial fascicles within the sciatic nerve at the thigh level (upper row) and their continuation as the tibial nerve distally at the ankle level (lower row). Nerve lesions are conspicuous under hyperintense signal alteration, which was found to be an effect of increased proton spin density by quantitative signal analysis. Within symptomatic diabetic polyneuropathy (DPN) groups (mild/moderate or severe), there is a strong proximal focus of lesion predominance located at the thigh level (the white reference line in the bony anatomical scheme on the left indicates the slice position). A strong proximal‐to‐distal lesion gradient with a marked decrease at the distal ankle level (lower white reference line at the ankle level) and increasing fascicular involvement across groups of increasing DPN severity (from left to right) were observed. NDS = Neuropathy Disability Score; NSS = Neuropathy Symptom Score.

Figure 2

Spatial mapping of lesion voxel count per position from the proximal thigh (slice position 1) to ankle levels (slice position 140). The spatial predominance of the lesion voxel burden in symptomatic diabetic polyneuropathy (DPN; mild/moderate or severe) is located proximally (thigh level slice positions 1–70) and is highest at this location for severe DPN. Increasing symptom severity is associated with an overall increase in the total lesion voxel burden, which is most pronounced proximally (at the thigh level, left side of the x‐axis) and less marked distally (at the ankle level, right side of the x‐axis). The differences in the proximal lesion count are significant for severe DPN (*p = 0.013) and mild‐to‐moderate DPN (**p = 0.003) compared to nondiabetic controls.

Figure 3

Ordered logistic regression analysis was significant for the quantitative microstructural marker of proton spin density (ρ/rho). This plot illustrates the cumulative probabilities predicted from the logistic regression model for different clinical stages of diabetic polyneuropathy (DPN) as functions of quantitated proton spin density (x‐axis). An increase of 10² proton spins was significantly associated (p < 0.001) with an odds ratio of 2.9 (95% confidence interval: 2.4–3.5) for predicting higher symptom severity.