In applications of the Double Echo Steady State (DESS) in the breast, phase errors due to respiration-induced B0 variation result in ghosting artifacts that degrade image quality and limit further clinical investigation of the method [1]. Correction of these ghosting artifacts with a DC navigator has shown promising but intermittent results [Figure 1, ref. 2]. In this work we describe a new method in which a set of possible phase errors due to respiration-induced B0 variation are simulated and an image entropy metric is utilized to assess the efficacy of each set of phase errors to correct the ghosting artifacts.

Time Varying Off-resonance Estimate:

Previous intermittent success using a DC navigator for correction of ghosting in DESS in the breast indicates that certain DESS coil/echo DC navigator data effectively captures phase variation due to respiration induced off-resonance variation (Figure 1). Actual motion of the breast tissue is very limited in breast MRI particularly in the medial regions of the breast due to the prone position of the patient. The relative phase variation of DESS echo 1 navigator data from 4 medial channels of the breast coil were averaged and used to estimate the off-resonance variation per TR. The legitimacy of these estimates was verified by comparison with previous measurements of respiration induced B0 variation in the breast.

Simulation of candidate phase propagations:

This in-vivo time-varying off-resonance estimate can be utilized with known imaging parameters (TR, TE, gradient area) and established tissue parameters in the breast (T1,T2 and ADC) to simulate the expected phase errors in echo 1 and 2 of the DESS acquisition. However, both the amplitude of the off resonance variation as well as B1 can significantly vary across the breasts. To effectively capture these variations in our Bloch simulations of phase error due to the estimated off resonance variation, we simulated data for a range of off-resonance amplitudes (0.5, 1.0, and 2.0 x estimated off-resonance amplitude) and flip angles (0.7, 1.0 and 1.4 x prescribed flip angle). A total of 9 different time-varying phase error estimates were simulated. Relative phase from these estimates was rewound in the acquired k-space data providing 9 image datasets with varying levels of artifact correction.

Image Entropy Metric for Assessment of Artifact Reduction:

Respiration induced B0 variation results in coherent ghosting artifacts across the breasts as well as across the region with no tissue between the breasts in our axial acquisition. Based on these artifact characteristics, an image entropy metric [3] was utilized to quantify the level of artifact for each corrected image. The metric was calculated on a slice-by-slice basis for each echo/coil combination. An uncorrected dataset and a set of images corrected directly from the in vivo DC navigator phase measurements were also assessed with image entropy metric. For each slice/coil/echo combination, the corrected image with minimum image entropy was selected as that with lowest level of artifact and combined to create a final set of corrected images. A flowchart summarizing the proposed method is shown in Figure 2.

The acquired and simulated phase data from a volunteer acquisition for the specified range of off-resonance amplitude variations and flip angles is shown in Figure 3. Correction of the acquired image data with the candidate simulated phase errors did result in varying levels of artifact reduction and the image entropy metric effectively quantified the level of artifact in the images (Figure 4). The proposed DC navigator and image entropy metric based method reduced the presence of ghosting artifacts in both echoes of DESS in the breast in comparison to direct application of phase errors in the acquired DC navigator data (Figure 5).

In this initial investigation, the proposed method is successful at reducing ghosting artifact in both the first and second echo DESS breast images. The various levels of correction achieved with different simulated phase errors supports the use of a set of simulated phase errors and also demonstrates that the chosen amplitude and flip angle ranges effectively capture the in-vivo variation. The most tenuous aspect of this method is the choice of DC navigator data from which to estimate the off-resonance variation. Future work will include a more systematic method of determining “good” navigator data on which to base the estimate. Finally, this method does not provide a voxel-by-voxel off resonance correction but the improvement in image quality may still facilitate further investigation of this technique in the clinic.