0649

Microstructural characterization and validation of a 3D printed phantom for diffusion MRI
Farah N. Mushtaha1, Tristan K. Kuehn 1,2, John Moore 3, Corey A. Baron1,3,4, and Ali R. Khan1,2,3,4,5

1Centre for Functional and Metabolic Mapping, Robarts Research Institute, London, ON, Canada, 2Biomedical Engineering, Western University, London, ON, Canada, 3Imaging Research Laboratories, Robarts Research Institute, London, ON, Canada, 4Medical Biophysics, Schulich School of Medicine and Dentistry, London, ON, Canada, 5The Brain and Mind Institute, London, ON, Canada

Synopsis

Validating diffusion MRI (dMRI) representations and models of brain tissue is challenging because there is no reference ground-truth for in vivo scans. We propose a form of 3D printed phantoms as a flexible paradigm for investigating and validating microstructural indices and crossing fibres that is reproducible with inexpensive materials. As a proof of concept, we obtained multishell dMRI data in samples with varying crossing angles and printing parameters, and investigated the performance of constrained spherical deconvolution to extract diffusion parameters that accurately describe the crossing fibre bundles. We also investigated the effect of printing parameters on the phantoms’ microstructural anisotropy.

Introduction

Diffusion MRI has the potential to quantify histological features of the brain at a micrometre scale 1. However, it is difficult to validate diffusion MRI techniques because there is usually no well-characterized “ground truth” for comparison. Many phantoms with independently verifiable microstructural characteristics have been proposed, but many of them have time-consuming or non-trivial preparation processes, and have limited ability to mimic bundles of crossing fibres 2.We propose a novel phantom produced using fused deposition modeling (FDM) 3D printing with a composite material consisting of rubber-elastomeric polymer and a PVA component (PORO-LAY 3). When immersed in water the PVA dissolves, leaving behind small water filled pores with anisotropy along the direction of motion of the print head (Fig 1). These phantoms can mimic the diffusion characteristics of axons and be produced with complex orientations. Printing parameters like print-head temperature may provide a means to customize microstructural properties of the pores; accordingly, here we investigate how diffusion tensor MRI (DTI) and constrained spherical deconvolution (CSD) parameters vary with print-head temperature. Further, we investigate the ability of CSD to separate crossing fibres for several crossing angles.

Methods

12 phantoms were created by printing 11 mm radius cylinders with 100μm layers of parallel lines with alternating orientations to mimic crossing fibres with crossing angles of 0°, 30°, 60°, and 90° (Fig. 1). For each crossing angle, a phantom was produced at three different print temperatures: 215℃ (low), 225℃ (nominal), and 235℃ (high). The phantoms were immersed in water for 168 hrs and stacked in a test tube with distilled water for imaging. Diffusion MRI was implemented at 9.4T using: 120, 60 and 20 directions at b=2000, 1000, and 0 s/mm2, respectively, TE/TR=37/2500 ms, FOV=200x200 mm2, 0.7 mm isotropic in-plane resolution, 6 axial slices (3 mm, one per phantom), and scan time 8.5 min per each of 2 scans to cover all phantoms. MRtrix4 was used to compute DTI (axial diffusivity, AD; radial diffusivity, RD; fractional anisotropy, FA) and CSD representations. For CSD 5, a response function was estimated from the 215℃ temperature 0° phantom 6. Fibre orientation distributions (FODs) were segmented to identify two fibre populations 7. Dispersion, peak FOD magnitude, and apparent fibre density (AFD) 8 were estimated for each fibre population, and the fibre crossing angle was estimated at each voxel. This analysis was repeated with the number of acquisitions decreased to 50 (50 151000 302000 and 20 (20 6 b1000, 12 b2000), where isotropic spatial sampling of directions was approximately maintained. To investigate phantom stability, an identical scan and analysis was performed 11 days later.

Results

Peak FOD amplitude, AD, RD, and dispersion had a strong observed dependence on print temperature. (Fig. 2).

The estimated FODs for the optimal temperature phantoms are visualized in Figure 3.

The 60° phantoms were estimated to have crossing angles of approximately 70°, while the crossing angles of the 90° phantoms were correctly estimated with good precision (Fig. 4a).

CSD metrics in the two fibre populations converged for crossing angles greater than 30°, which is expected for our print geometry where only the orientation varied between layers (Fig. 4b-d). CSD metrics in the dominant fibre population change with crossing angle, suggesting that partial volume effects remain after CSD.When the number of diffusion acquisitions was decreased from 200 to 20, a decrease in peak FOD amplitude and AFD, and an increase in dispersion were observed. The difference in mean crossing angle between the two scan dates was less than two degrees in all cases, suggesting stability over at least 11 days.

Discussion

The strong dependence of diffusion properties of the phantoms on print-head temperature suggests that customization of diffusion properties may be possible.The inability of CSD to resolve 30° crossings in this case is consistent with simulation findings that CSD cannot resolve crossing angles less than 40° 5. The observed 10° bias of estimated crossing angle for 60° highlight how this type of validation can uncover potential issues and be used to improve robustness of algorithms. Likewise, the introduction of bias in peak FOD amplitude and AFD with fewer acquisitions highlights the potential of these phantoms for validation. Future work should include more precise characterization of the properties of the 3D printed PORO-LAY material. Diffusion MRI models of brain tissue estimate an increasingly diverse set of quantities, so precise characterization of the phantoms will maximize their usefulness. Future work will investigate extending this validation platform to more complex brain-mimetic geometries like fanning or bending.

Acknowledgements

This work was supported by a BrainsCAN Stimulus Grant from the Canada First Research Excellence Fund, Brain Canada, and Discovery Grants from the Natural Sciences and Engineering Research Council (NSERC)

References

  1. Alexander D.C, Dyrby T.B, Nilsson M, and Zhang H. Imaging brain microstructure with diffusion MRI: practicality and applications. NMR in Biomedicine. 2017;e3841. doi:10.1002/nbm.3841.
  2. Fieremans E. and Lee H.-H. Physical and numerical phantoms for the validation of brain microstructural MRI: A cookbook. NeuroImage. 2018;182:39–61.
  3. Abu-Sardanah S.O, Hussain U, Moore J, Baron C.A, Peters T.M., Khan A.R. Design and evaluation of a diffusion MRI fibre phantom using 3D printing. Proc. SPIE. 10573, 2018.
  4. Westin C, Peled S, Gudbjartsson H, Kikinis R, and Jolesz F. ISMRM '97. Vancouver Canada, (1997, April),1742.
  5. Tournier J.-D, Calamante F, and Connelly A. Robust determination of the fibre orientation distribution in diffusion MRI: Non-negativity constrained super-resolved spherical deconvolution. NeuroImage. 2007;35(4):459–1472.
  6. Tournier J.-D, Calamante F, and Connelly A. Determination of the appropriate b value and number of gradient directions for high-angular-resolution diffusion-weighted imaging. NMR in Biomedicine. 2013;26(12):1775–1786.
  7. Smith R. E, Tournier J.-D, Calamante F, and Connelly A. SIFT: Spherical-deconvolution informed filtering of tractograms. NeuroImage. 2013;67:298–312.
  8. Raffelt D, Tournier J.-D, Rose S, et al. Apparent Fibre Density: A novel measure for the analysis of diffusion-weighted magnetic resonance images. NeuroImage. 2012;59(4):3976–3994

Figures

Figure 1.a) 12 phantoms with varying parameters in a test tube. b) SEM image of a phantom created with low print-head temperature. c) SEM image of a phantom with nominal printing temperature. d) SEM image of a phantom with high printing temperature.

Figure 2. DTI-derived and CSD-derived metrics as a function of print temperature, measured from the 0° phantoms. Error bars indicate standard deviation. a) Fractional anisotropy vs printing temperature b) Axial diffusivity vs printing temperature.c) Radial diffusivity vs printing temperature. d) Peak FOD amplitude vs print temperature. e) Mean AFD vs printing temperature. f) Mean dispersion vs printing temperature

Figure 3: FODs from 16 representative voxels for each crossing angle, taken from the nominal temperature phantoms.a) Linear phantom with 0° crossing. b) Phantom with 30° crossing. c) Phantom with 60° crossing. d) Phantom with 90° crossing

Figure 4. Estimated crossing angle (a) and CSD-derived metrics (b-d) as a function of ground-truth crossing angle and total number of dMRI acquisitions. a) Estimated crossing angles for 60° and 90°; results for 0° and 30° are not shown because only one fibre population was able to be accurately segmented. Peak FOD amplitude (b), AFD (c), and dispersion (d) versus known crossing angle for the two fibre bundles segmented from the FODs with different subsamplings of the acquisition. The legend outside the plots applies to (b)-(d).

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
0649