Towards Microcalcifcation Detection Using Susceptibility Weighted Phase Imaging: Simulations and Phantom Experiments

INTRODUCTION The presence of calcium deposits in the breast can be an indicator of malignancy, so the ability to detect such abnormalities by MRI would be significant. Calcium deposits have a different magnetic susceptibility from surrounding tissues that can in principle be observed using susceptibility weighted imaging (SWI) [1]. Here we perform simulations and phantom experiments to determine whether SWI can be used to detect calcifications in practice. METHODS Simulations The MR signal phase variations caused by a spherical calcium-like object immersed in a homogeneous media was simulated in 2D using the method described in (4). The simulations incorporate phase shift as a function of the static field (B0) strength and orientation, echo time (TE), and the geometry and susceptibility characteristics of the perturbing material. Calculation of magnetic susceptibility differences is performed for each voxel in a “slice” to generate phase images. The simulations employed the following MRI parameters: voxel size = (300 μm), sphere centered within the voxel, B0 = 4.7T, and TE = 0.3 ms and the resulting data can be seen in Fig. 1a. This “template” was then added to noise to yield an image with SNR=5 (Fig. 1b); this resulting image then simulates experimental “data”. We then aim to asses the ability of the template to locate a susceptibility induced signature (i.e., a phase pattern) within an experimentally measured data set by using a 2D cross correlation with the predicted pattern. Phantom studies. Phantoms were constructed with 1.0 mm barosilicate spheres immersed in Agarose gel. MRI employed a 4.7T Varian scanner to obtain 3D gradient echo images with TR\TE\α= 10.5ms\5ms\15, with an acquisition matrix of 256×128×128 over a 80×40×40 FOV. Figure 2 shows the MRI magnitude (panel a) and phase image (b) obtained from the phantom. A phase pattern template (Fig. 1a) was then used to identify whether the phase pattern of small irregularities are present within the phantom phase image (Fig. 2b). Analysis The goal is to locate, within an experimental data set, a characteristic phase response pattern induced by a calcium-like sphere. For the simulation, this amounts to simply applying a 2D cross-correlation (the ‘normxcorr2’ function with Matlab (Mathworks; Natick, Massachusetts) between the template and the data image. The resulting cross correlation matrix will have high values where the template and the target pattern in the image are similar. For the phantom studies, before the cross-correlation is performed, we first correct the phase jumps for phase values outside [–π, π] using the technique provided by (5); an example can be seen in Fig. 2c. RESULTS Fig. 1c depicts the cross-correlation matrix obtained for the simulation studies. It shows a high correlation coefficient (~0.8) near the center of the field of view (FOV) where the calcification is present. As the SNR of this data set is quite low (SNR=5), this offers promise for the technique. Fig. 2d depicts the experimental analogue displaying a correlation coefficient of ~0.65 in the center of the FOV. DISCUSSION and CONCLUSION We have introduced a simple technique to locate a susceptibility induced signature within a data set by computing the cross-correlation between a template and the target image. The preliminary efforts indicate that there is some merit to the approach. Ongoing and future efforts include exploring the ability of the method to find phase signatures with different levels of background noise and structure. Additionally, we are constructing a library of more realistic templates (various sizes, shapes, etc. of calcifications) to increase the applicability of the method. REFERENCES 1. Yamada et al. Radiology 1996;198:171-178. 2. Haacke et al. Magn Reson Med 2004;52:612-618. 3. Shmueli. PISMRM 2008;16:641. 4. Haacke et al. New York Wiley-Liss; 1999. p. 741-71. 5. Goldstein et al. Radio Science 1988;23:713-20. ACKNOWLEDGEMENTS 1R01CA129961, NIBIB 1K25 EB005936, NCI 1P50 098131, and NIH P30 CA68485