Background

Numerous studies have identified alterations in iron concentration in the deep grey matter (DGM) of patients with neurological diseases such as multiple sclerosis (MS). Quantitative Susceptibility Mapping¹ (QSM) measures bulk tissue magnetic susceptibility, or χ, which depends linearly on the concentration of paramagnetic tissue iron. The effective transverse relaxation rate² (R₂^*) depends on both the iron concentration and distribution within the imaging voxel.³ The recently introduced⁴ iron microstructure coefficient (IMC), or κ, quantifies the subcellular distribution of iron based on χ and R₂^* reconstructed from different pulse sequences. In the present work, we introduce a fully automated and generalized pipeline for IMC analysis using QSM and R₂^* obtained from the same multi-echo GRE sequence. The program code will be made available at the time of the conference.

Methods

The pipeline was designed for full automation, reproducibility, and widespread use across different study designs and pulse sequences. Pipeline inputs (Figure 1) included the following: susceptibility and R₂^* maps; T1-weighted (T1-w) subject data; T1-w and QSM hybrid templates;⁵ binary regions of interest (ROIs) defined in the template space. The pipeline output was a comma-separated-values (CSV) file containing the κ value per ROI per subject and their uncertainties.

The pipeline was implemented using makefiles and Docker containerization based on the Neurodocker framework, accessed through a VNC-based virtual desktop (Figure 2). Two Neurodocker images were generated for pre-sorting data and pipeline execution. As shown in Figure 3, a single docker command applies the input ROI labels to QSM and R₂^* maps using T1-QSM-based bi-modal warp field computations (Python 3.0; ANTs), followed by a total least-squares fitting to the voxel value distributions of χ and R₂^* to generate κ values. Susceptibility was referenced to the lateral ventricles.

We tested the pipeline in a pilot study with 28 subjects (14 patients with MS, 1:1 age- and sex-matched controls). Inclusion criteria for patients: McDonald criteria; EDSS >4.0; age >40 years. Exclusion criteria included steroids or other immunomodulatory therapy <30 days of MRI; consumption of alcohol <48 hours of MRI; immunosuppressant agents <6 months of MRI; investigational drug or experimental procedure <30 days of MRI; valium/diazepam <24 hours of MRI. Patient (control) age was 60.6±8.2 (58.4±8.0) years and female:male ratio was 6. Data was acquired at 3T (Canon Vantage Titan) using a 32-channel brain coil and a bipolar 3D MGRE sequence (matrix=328x256x70, 0.7x0.9x1.8 mm3, flip=14°; TR/TE1/dTE/nTE=30ms/4ms/2.5ms/10).

QSM maps were reconstructed using best-path unwrapping, LBV and HEIDI. R₂^* was calculated using log-formalism with the Power method.⁶ We created a custom bi-modal magnitude-QSM template with bilateral subcortical atlas labels for the following DGM regions: globus pallidus; thalamus (including subnuclei); caudate; putamen; dentate; nucleus ruber; substantia nigra. We applied the pipeline with all inputs on a scientific workstation (Ubuntu 18.04; 48 cores @ 2.70GHz; 378 GB RAM).

Using the Statistical Package for Social Sciences (SPSS; version 28; IBM, Armonk, NY), the CSV file was imported and paired t-tests were performed per ROI to identify significant differences between the left and right hemispheres. If a significant inter-hemispheric difference was discovered, each hemisphere was separately analyzed; otherwise, the mean value of the left and right hemispheres was used. Univariate ANCOVA was conducted with sex, age, and MTR (for myelin correction) as covariates. The Benjamin-Hochberg procedure with p<.05 was used to minimize the false discovery rate.

Results

The pipeline execution time was 50 minutes. Visual assessment revealed no failed segmentations. Nominal IMC values were between 2 s⋅ppb in the lateral region of the thalamus and 6 s⋅ppb in the dentate and red nucleus (Table 1), similar to those reported in the previous study.⁴ The thalamic IMCs were significantly different between HC and MS (p=0.039). ANCOVA revealed that patients had significantly elevated IMC values in the (right) globus pallidus (effect size d=4.448, p=0.047). Thalamic IMCs significantly increased with age (d=10.39, p=0.04).

Discussion and Conclusions

The introduced automated pipeline enables IMC analysis in a robust and reproducible manner for large cohort studies using multi-echo gradient echo imaging. The thalamic IMC was found to be dependent on age and significantly higher in MS patients compared to HC, deviating from previous findings⁴ of age-dependency in the caudate, globus pallidus, and putamen, as well as significantly lower IMC in the dentate and caudate of MS patients. We attribute differences in outcomes to the older age of the study cohort, the small sample size, and differences in the acquisition scheme. Future studies with larger cohorts will use the proposed pipeline to establish baseline IMC values and investigate effects from disease in longitudinal studies. The pipeline will be optimized to incorporate data on tissue diffusion properties.

Acknowledgements

Research reported in this publication was supported by the following: the Dana Foundation; the National Institute Of Neurological Disorders And Stroke of the National Institutes of Health under Award Number R01NS114227; the National Center for Advancing Translational Sciences of the National Institutes of Health under Award Number UL1TR001412; the Experiential Learning Network at the University at Buffalo; an equipment grant from Canon Medical Systems Corporation and Canon Medical Research USA, Inc. The content is solely the responsibility of the authors and does not necessarily represent the official views of the funding agencies.