We aim to demonstrate the feasibility of in vivo cardiac Quantitative Susceptibility Mapping (QSM) for non-invasive quantification of venous oxygen saturation (SvO₂) within the heart chambers, which may play an important role in the detection and management of cardiac shunt and pulmonary disorders.

Eight healthy volunteers were scanned using an ECG gated multi-slice multi-echo 2D-FGRE sequence at 1.5T using an 8-channel cardiac coil with the following scan parameters: 8 echoes, first TE≈3.6ms, $$$\triangle$$$TE≈2.2ms, TR≈23ms, voxel size≈1.25x1.25x5mm³, acquisition resolution=192x192, 8 views per heartbeat, flow compensation in the readout and slice directions, ASSET acceleration factor=1.5, and breath hold≈15 seconds per slice. A stack of 20 short-axis images were acquired for each subject. From the multi-echo complex data, an unwrapped field map was obtained using a magnitude image guided graph cut method¹. Chemical shift from fat was removed via an iterative chemical shift update². Because of the close proximity of myocardium and lung, background field removal was incorporated into the L1 MEDI³ method, such that the susceptibility map was obtained from the total field directly. This method allowed the solution to have susceptibility sources both inside and outside the ROI to account for both local field and background field. A diagonal preconditioner $$$P$$$, which was set to 1 and 40 for inside and outside the ROI, respectively, was introduced to improve the convergence speed:

$$y^*=argmin_y\frac{1}{2}||w(Mf-Md\otimes(P\cdot y))||^2_2+\lambda||M_G\triangledown(P\cdot y)||_1$$

Where $$$w$$$ is a noise weighting term, $$$M$$$ is one inside ROI and zero outside, $$$f$$$ is the total field, $$$d$$$ is the dipole kernel, $$$\otimes$$$ is a convolution operator, $$$\lambda$$$ is the regularization parameter, $$$M_G$$$ is a binary edge Mask, and $$$\triangledown$$$ is the gradient operator. The susceptibility is then $$$\chi=P\cdot y^*$$$. Figure 1 shows a flow chart that demonstrates the steps for processing the phase data. For comparison, we computed the QSM from standard processing scheme (PDF⁴+MEDI) as well.

The ROI for left ventricle (LV) and right ventricle (RV) were obtained through manual segmentation. The susceptibilities in LV and RV were calculated as the average over the respective ROI, and the susceptibility difference between LV and RV was scaled to obtain SvO­₂ value⁵.

The mean susceptibility difference between RV (deoxygenated blood) and LV (almost fully oxygenated blood) from the QSM generated with preconditioned MEDI using the total field as the input was 292ppb±79ppb. Assuming the blood in LV is 100% oxygenated, the mean SvO₂ in these eight healthy volunteers was 78.1%±5.1%. The mean susceptibility difference between RV and LV from the QSM generated with PDF + MEDI was 286ppb±102ppb, and the corresponding mean SvO₂ was 78.6%±7.5%. Figure 2 shows a visual comparison between the QSM map from processing with PDF+MEDI, and preconditioned MEDI with total field as the input. Figure 3 shows the magnitude and the QSM images of three healthy volunteer.

Our preliminary data demonstrated, for the first time, high quality in vivo cardiac QSM. The SvO₂ derived in this study with QSM is about 8% larger than the expected value, which is about 70%. This higher derived SvO₂ could be attributed to the decreasing oxygenation level in the LV during breath hold. According to literature⁶, the oxygenation level in LV will decreased by about 8.6% at the end of a 15 seconds breath hold. The susceptibility in RV should remain the same for a 15 seconds breath hold.

Our proposed cardiac QSM employs graph cut phase analysis that provides robust fat water separation, which is needed to deal with epicardial fat and other fat in the chest and liver. The preconditioned total field inversion eliminates artifacts in the myocardium from standard QSM caused by errors from imperfect background field removal. Although both processing scheme yielded similar SvO₂, the standard processing scheme produces QSM map with obvious artifact, which can hinder its diagnostic value.

We have demonstrated the feasibility of high quality in vivo cardiac QSM with graph cut phase analysis and preconditioned total field inversion. Future work will focus on accelerated 3D acquisition, and applications of cardiac QSM in various cardiac diseases.