Extracting local frequency (phase) shift map (LFS) from data acquired with multi-channel RF coils requires care to remove channel-dependent effects, background field contribution and to eliminate phase wrap. A common approach is to shift the phase data from each channel to a reference phase (e.g. the phase of one echo) in order to eliminate channel dependent phase terms, followed by complex summation. Homodyne filtering,¹ i.e., a combination of phase unwrapping and high-pass filtering, can then be used to generate a LFS map. Recently, the inter-echo variance (IEV) channel combination technique was presented, demonstrating a reduction in errors associated with imperfect channel combination.² The IEV technique processes the individual channel phase data and uses the inverse of the inter-echo variance of unwrapped high-pass-filtered phase as a weighting factor during channel combination. In this work, we investigated the effect pre-channel combination (pre-CC) on the contrast and details preserved in LFS maps and compared the results to post-channel combination (post-CC) phase data processing.

Image acquisition: Two healthy volunteers were scanned with a six-echo, whole-brain GRE sequence at 7T using a 16-channel transmit/receive head coil (TR/TE/echo spacing/flip angle 40/3.7/4.1 ms/13°, voxel size: 0.5 x 0.5 x 1.25 mm, GRAPPA R=2); individual channel data were saved for later processing (phase shown in Fig. 1a).

Post-processing: Two post-processing pipelines were implemented in MATLAB as follows: (i) pre-CC: The channel data were unwrapped³ (Fig. 1 b), then high-pass filtered with a Gaussian filter⁴ (GHPF) for background field removal (Fig. 1 c). The processed channel data were combined using the IEV channel combination technique (Fig. 1d). The GHPF size was varied between 1-mm to 50-mm in this pipeline in order to investigate effect of filtering on the contrast of structures in the phase image. (ii) post-CC: Using complex summation the channel data were combined (Fig. 2a), after the channel-dependent phase term was eliminated by subtracting the first echo from all subsequent images. A homodyne high pass filter (HHPF) was applied to these image data to generate local phase shift maps. Filter size was varied between 10 and 90 percent of the field of view (FOV); phase maps were converted to LFS (Fig. 2b).

Analysis: The phase images resulting from each technique were evaluated for contrast and visibility of structures, including vessels of various sizes in different regions of the brain. The final images obtained from various filter sizes in each pipeline were compared against each other quantitatively using the following protocol. Five vessels were selected from each of the following regions: just above the cerebellum, the level of red nuclei, the level of the lateral ventricles, and just above the lateral ventricles. Line profiles were drawn across the vessels at three points along each selected vessel. The minimum pixel intensity on the line profile, indicating the vessel signal, was subtracted from the mean value of pixels along the tails of the line (defined as beginning two pixels away from the minimum intensity on the line profile on each side) to define signal difference. Contrast was defined as the signal difference divided by the mean of the signals. The optimal filter size for HHPF and GHPF were identified based on the measured contrast. Contrast in the LFS maps generated using the optimized filter parameters was subsequently compared to evaluate performance of each processing pipeline (paired t-test).

For all vessels analyzed, optimal contrast was achieved with a filter size of 7-mm for GHPF and 30 % FOV for HHPF (Fig. 3). Vessel details seen in the raw, individual channel phase images were consistently preserved through the pre-CC pipeline (Fig 1 a, d) but not in the post-CC pipeline (Fig. 2 b). Additional examples, from the different regions and volunteers, comparing the preservation of vessel detail and continuity are seen in Figs. 4 and 5. Quantitative analysis demonstrated significantly higher contrast in the pre-CC pipeline images (1.4 ± 0.5) compared to the LFS maps generated using the post-CC method (1.1 ± 0.2) (p < 0.05).It was shown that LFS maps resulting from pre-CC processing using IEV preserve vessel detail that was seen in the raw channel phase data. Qualitatively, the pre-CC results also demonstrate more robust reconstruction of the phase in the cortical regions (arrowheads in Figs 4 and 5). These findings have important implications for susceptibility imaging and mapping. The ability of pre-CC processing using IEV to preserve susceptibility related structural information in the LFS maps promises to result in more accurate susceptibility maps and images.