Precise brain phantoms are important for evaluating the quality of segmentation tools for brain MRI. Here we suggested the construction of a 3D physical brain phantom as gold standard to validate the performance of those tools. Folding patterns of grey and white matter compartments were replicated using 3D-printed models from a real structural brain scan. T1 and T2 intensities of these brain regions in a 3 Tesla MRI were mimicked by a 0.6% agar mixture containing the appropriate concentrations of the paramagnetic compounds Ferumoxide and Manganese chloride. With its 3D-printed brain-like design, the phantom showed to be a promising alternative to existing methods for MRI segmentation validation.
The study was divided in three stages. Initially, MRI scans of a healthy subject were performed to derive a ground truth for brain T1 and T2 values. T1 relaxation maps were acquired at a clinical 3 Tesla (T) Whole-Body MRI Scanner with a 20-channel head coil using an inversion recovery spin echo sequence (resolution = 1x1x5 mm3, TR = 6000 ms, TE = 12 ms, TI = 30/50/100/200/500/750/1000/2000 ms). T2 relaxation maps were acquired using a multi-echo spin echo sequence (resolution = 1x1x5 mm3, TR = 3000 ms, TE = 13.2 - 422.4 ms, echo spacing 13.2 ms). Average T1 and T2 intensities within regions of interest (ROIs) for WM (2 ROIs of dimensions 3x3 mm2) and GM (4 ROIs of dimensions 3x3 mm2) were computed. The ratio derived from average intensities of WM and the ones of GM was then mimicked by testing different concentrations of the two MRI contrast agents Manganese Chloride (MnCl2; concentrations = 0 to 0.25 mM) and Ferumoxide (concentrations = 0 to 10 mM) in a 0.6% agar gel.
At a second stage, anatomic scans were performed to define a model for the construction of the brain scaffold. A whole-brain isotropic MPRAGE scan (resolution = 1x1x1 mm3, TR/TI/TE = 1570/900/2.48 ms, α = 8, Figure 1) of the same subject was automatically segmented by Freesurfer (Harvard University, Cambridge, Massachusetts, USA), allowing the identification of the pial surface and the GM/WM interface. These surfaces were used as template for the phantom shape and, thus, converted to two stereolithography files. The corresponding mesh density was reduced by 50% of the initial triangle count, the surfaces slightly smoothed, and the GM/WM boundary surface was decreased by 40 mm of its maximal width in the sagittal plane (Meshmixer 2.9, Autodesk, Inc., San Rafael, California, USA). Only the surfaces of the left hemisphere were then 3D-printed using polylactic acid (PLA) thermoplastic (Figure 2). To create flexible molds, easily allowing the detachment of the paramagnetic gels, the 3D-printed surfaces were covered by a brushable platinum-cure silicone rubber with a shore A hardness of 30 (Figure 3).
At the last construction stage, the phantom was assembled as follows. The GM/WM interface mold was filled with the WM solution, the mold was removed, the pial surface mold was placed on top of the WM gel and the GM solution was injected into the gap between WM gel and outer silicone cast. Finally, for the evaluation of the phantom, the same inversion recovery spin echo, multi-echo spin echo, and MPRAGE sequences were used to verify the structural accuracy as well as the T1 and T2 intensities of the two phantom compartments.
1. Despotovic, I., Goossens, B. & Philips, W. MRI Segmentation of the Human Brain: Challenges, Methods, and Applications. Comput. Math. Methods Med. 2015, (2015).
2. Collins, D. L. et al. Digital Brain Phantom. IEEE Trans. Med. Imaging 17, 2005–2007 (2007).
3. IBSR, "The Internet Brain Segmentation Repository," 2013, http://www.nitrc.org/projects/ibsr.
4. Khan, A. F. et al. A novel MRI-compatible brain ventricle phantom for validation of segmentation and volumetry methods. J. Magn. Reson. Imaging 36, 476–82 (2012).
5. Chen, S. J. et al. An anthropomorphic polyvinyl alcohol brain phantom based on Colin27 for use in multimodal imaging. Med. Phys. 554, (2013).