Diffusion weighted imaging (DWI) with spin-echo echo planar imaging (SE-EPI) is increasingly used for screening, diagnosis, and treatment monitoring in breast cancer because malignancies exhibit lower apparent diffusion coefficients (ADC) than benign lesions¹. SE-EPI in breast imaging produces lower quality DWI than in the brain due to larger B0 inhomogeneity and abundant adipose tissue, which produce geometric distortion and chemical shift artifacts from imperfect fat suppression. Additionally, inconsistencies between k-space lines of opposite readout polarity (RO+/RO-) due to eddy currents and other factors cause Nyquist, or N/2, ghosts. Even small residual ghosts in source images can produce bias in the resulting ADC maps. These phase errors are commonly corrected by the three-line navigator method², which acquires three readout lines without phase encoding with each scan to measure the phase difference between positive and negative RO lines. This method performs well in brain imaging but fails frequently in breast DWI. The purpose of this work is to compare the performance of this standard technique with alternative EPI phase correction methods to determine which method gives the best artifact suppression in breast DWI.

Six healthy female volunteers and one patient were recruited under IRB-approved protocols. The subjects were scanned prone on a 16-channel Sentinelle breast coil using a Siemens 3T PRISMA fit system. A standard 2D axial multislice SE-EPI bipolar diffusion sequence with four b-values (0, 100, 600, 800 s/mm², 2 averages and 3 directions each) derived from ACRIN 6698³ served as the base protocol (TR/TE=8000/74 ms, 32x32 cm FOV, nominal in-plane resolution 1.7 x 1.7 mm, 4 mm slices, 6/8 partial Fourier, ETL=52ms, right-left phase encoding, GRAPPA R=2, single shot SE-EPI ACS lines).

Five distinct phase correction strategies were used to correct this base DWI acquisition, summarized in Table 1. The 3-line navigator method² (A) used the integrated navigator acquisition to calculate a linear phase correction independently for each b-value, average, coil, and slice. The phase mapping method (B) used additional reference scans with double FOV to produce separate, unaliased, coil-combined RO+ and RO- images. The phase difference between these RO+ and RO- images was fit with a 1D linear model as described by ref [4] for each slice and applied to all repetitions in the base scan. The entropy method (C) found the 1D linear phase correction for each coil and slice that minimized the b=0 s/mm² image entropy using a multi-resolution discrete search^5,6 and applied this correction to all repetitions. The RO reversal method (D) repeated the base scan with the readout polarity reversed and performed a complex addition of the two datasets to cancel phase ghosts, similar to PLACE⁷. Finally, the full phase navigator method (E) used a reference scan acquired without phase encoding to perform a linear 1D phase correction that varied over the RO train⁸.

The intensity of residual Nyquist ghosts was measured by manually masking out the object and measuring the SNR in the background region. This background-to-noise (BNR) was calculated separately for each slice and b-value.

Figure 1 shows the BNR across methods for each subject, b-value, and slice. Residual ghosts were present in all cases at variable levels. Both the phase mapping (B) and entropy (C) methods gave lower average ghost levels and less variation in ghost signal than the standard 3-line navigator correction. Figure 2 shows an example comparing the ADC maps and residual ghost intensities between the standard (A) and best performing (B) methods. The standard 3-line navigator method, which corrects each b-value, average, and coil independently (A) fails frequently in breast DWI. Both phase mapping and entropy minimization performed significantly better even though a single correction from b=0 s/mm² data was applied equally over all repetitions. This suggests that the diffusion weighting gradients and physiologic motion do not have a substantial impact on the correction needed. Monopolar diffusion gradients may have a more noticeable effect. Additionally, the highest performing method (B) was implemented on coil-combined data in order to have enough SNR and coverage to obtain a complete phase map, suggesting that a single correction over all coils may be sufficient. Both phase mapping and entropy minimization can be extended for higher order or 2D ghost corrections.EPI phase correction using either an entropy minimization or a 1D phase mapping technique produces better Nyquist ghost suppression than the standard 3-line navigator in breast DWI. These methods may be applicable for DWI in other regions of the body and can be adapted for use with segmented or simultaneous multislice acquisitions.