Synopsis

Keywords: Nerves, Nerves

This study developed a multiparametric qMRI method to quantify fat fraction of leg muscles, myelin and axonal pathologies of peripheral nerves using MTR, MTsat, T₁, PD, T₂*, FA, MD, AD, RD, and nerve fascicular volume. The results will be applied in ongoing longitudinal studies to develop monitoring biomarkers in patients with polyneuropathies.

Introduction

Polyneuropathies result in demyelination and axonal degeneration in peripheral nerves, as well as intramuscular fat accumulation in denervated muscles. Given the rapid advancement of therapeutic developments, there is a pressing need to develop monitoring biomarkers in patients with polyneuropathies. Quantitative MRI (qMRI) offers several advantages over existing measures, including its 1) objective, non-invasive nature; 2) ability to measure proximal (e.g., sciatic) nerve which is largely inaccessible by other techniques (e.g., nerve conduction studies); and 3) sensitivity to axons and myelin pathologies in nerves as well as downstream denervation changes in muscles.¹

The goals of this study were to: 1) develop a multiparametric qMRI method that can be used for characterizing nerve pathologies and tracking the progression of the disease; 2) characterize the distal to proximal length-dependent changes of nerve qMRI parameters in healthy subjects; and 3) investigate the test-retest repeatability and inter-rater reliability of the qMRI methods. The qMRI method quantifies the fat fraction (FF) of leg muscles and parameters of the sciatic and tibial nerves that are sensitive to different aspects of myelin and axonal pathologies from magnetization transfer imaging [magnetization transfer ratio (MTR) and magnetization transfer saturation index (MTsat)], relaxation properties [T₁, proton density (PD), and T₂*], diffusion tensor imaging [fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD)], and morphology [the size of nerves represented as nerve fascicular volume (fVol)].

Methods

Data Acquisition: We studies 17 healthy volunteers (age: 36.2 ± 13.8 years old, body mass index (BMI): 23.1 ± 3.1, 52.9% male) using a Siemens VERIO 3T MRI with an 8-channel Tx/Rx knee coil. Ten of them had a second MRI scan (interval: 12.5 ± 19.7 days). The protocol includes: i) a 6-echo 3D gradient echo (GRE) sequence for fat-water decomposition and calculating the FF of muscles; ii) a 3D GRE sequence for measuring the nerve fVol; iii) two multi-echo 3D GRE sequences with and without an MT preparation for calculating the MTR and T₂*; iv) two 3D GRE sequences to calculate the PD, T₁, and MTsat; v) a pair of 3D GRE fast scans for B₁ mapping; and vi) a DTI scan with a b-value of 800 s/mm² and 20 gradient directions for FA, MD, AD, and RD. Fat suppression was used for all scans except for the 6-echo fat-water and the B₁ mapping scans. The imaging protocol was performed in an axial plane at the mid-thigh level to examine the proximal (sciatic) nerve and then repeated at the mid-calf level to examine the distal (tibial) nerve (Figure 1).

Image Processing and Data Analysis: An in-house developed pipeline in MATLAB was used to process all the data with the following steps automatically: i) image co-registration using SPM12²; ii) fat-water decomposition using a whole image optimization algorithm^{3, 4}; iii) B₁ mapping using a quadratic fitting-based algorithm⁵; iv) a simple linear regression for T₂* mapping⁶; v) a variable flip angle method for T₁ and PD mapping with B₁ corrections⁵; vi) MTR⁷ and MTsat⁸ maps calculated as described; vii) diffusion tensor reconstruction using the command line features in the DSI Studio⁹. A deep learning-based model was used for the automatic segmentation of individual muscles which were then combined as muscle compartments⁵. Sciatic nerve and tibial nerve ROIs were delineated on the high-resolution 3D GRE images using a semi-automated region growing tool in the SPIN software as described¹. Two independent raters delineated the nerve ROIs for inter-rater reliability analysis. Finally, an in-house developed MATLAB program automatically calculated and tabulated all the quantitative values for nerves and muscles.

Statistical Analysis: MATLAB with an in-house developed script was used for statistical analysis and visualization. Test-retest repeatability and inter-rater reliability were evaluated using the Bland-Altman, intraclass correlation coefficient (ICC)¹⁰, and coefficient of variation (COV) analyses. The distal-to-proximal changes were assessed using paired t-test. Correlations among parameters were tested using Pearson’s linear correlation coefficient.

Results

As shown in Figures 2 and 3, normative qMRI values of nerves and muscle compartments had good to excellent test-retest agreement, as well as good to excellent inter-rater agreement (all ICC > 0.8). Whole muscle FF was correlated with the subject’s BMI (r = 0.76 for thigh muscle; r = 0.70 for calf muscle) but not with nerve qMRI parameters (Figure 4). Nerve MTR and MTsat were negatively correlated with T₂*, T₁, PD, MD, AD, and RD, but not with FA (Figure 4). There were significantly length-dependent differences between distal and proximal sites in muscle FF, nerve MTsat, T₁, PD, FA, MD, and RD (Figure 5).

Discussion and Conclusions

Our multiparametric qMRI method shows good to excellent repeatability. This sets a solid base for evaluating disease progression in patients with polyneuropathies. The machine learning model used in the analysis was reliable in measuring compartmental and whole muscle fat fractions. The distal-to-proximal changes observed in the study are promising for evaluating pathological progression in patients with polyneuropathies.

Acknowledgements

This study is supported by grants (R61NS119434, R21TR003312) from NIH.