To understand, characterize and eliminate dark band artifacts in subtraction dynamic contrast enhanced (DCE) breast MRI images, which reduced confidence for detecting small cancers. Well-known chemical shift artifacts include spatial misregistration due to resonance frequency offset of fat in the presence of readout gradient (first kind) and signal amplitude modulations¹ and etching artifacts at boundaries of fat and water in non-subtraction images (second kind). A related phenomenon was previously reported on opposed-phase images only.^2,3 This work generalizes the later phenomenon to all off-resonance angles and describes the relation to dark band artifacts in post-contrast subtraction images.

Voxels containing 0-100% signal fractions from fat were simulated in Matlab assuming equal signal from fat and water. 50% enhancement was assumed for water and 10% for fat. Total voxel signal was calculated as a function of the phase angle between water and fat (simulated from 0-180° in steps of 20°).

Phantom experiments were performed by constructing mixtures of methylene chloride and oil.⁴ The nonpolar doping agent chromium acetylacetonate (CrAcAc) was used to approximate the T1 value of breast parenchyma. A second set of vials had additional CrAcAc to simulate 50% enhancement of breast parenchyma and 10% for fat. The vials were imaged with a spoiled gradient-echo sequence on a 1.5T scanner.

DCE subtraction images of the breast were obtained using a routine clinical fat-suppressed VIBE sequence (TE/TR=1.6ms/4.1ms) along with a modified in-phase VIBE sequence (TE/TR=4.8/7.2). The modified sequence had increased partial Fourier in both phase and partition-encode directions to maintain the temporal resolution. The modified in-phase VIBE sequence bracketed the routine clinical DCE breast MRI study; therefore, there was a longer delay between pre and post-contrast in-phase images as well as increased delay from bolus administration.

Figure 1 shows routine clinical VIBE images. Note multiple obtrusive serpiginous band artifacts in the breast parenchyma on the subtraction image (white arrows). Note also a concentric dark band artifact at the skin-breast interface seen on the subtraction image only (indicating this is not a chemical shift artifact of the second kind).

Figure 2 depicts schematic diagrams of a voxel containing 25% water and 75% lipid signals, both in and opposed-phase. In the opposed-phase case, there is paradoxical signal loss on post-contrast imaging.^2,3

Figure 3 shows simulation results before and after contrast enhancement and subtraction. Note that contrast enhancement is a function of fat signal fraction as well as echo-time (phase angle between fat and water). Figure 3d shows the absolute error and 3e the % error from the in-phase case. The error in enhancement is always negative, increases as a function of fat-water phase angle and depends on lipid signal fraction.

Figure 4 shows the phantom results for the “unenhanced” and “enhanced” vials as well as the % enhancement. The pure oil vial has approximately 4x higher signal than the pure methylene chloride and therefore maximum signal cancellation occurs near an oil fraction of 20%. Note the similar enhancement dependence on echo-time and oil signal to the simulations (4c to 3c) including paradoxical enhancement.

Figure 5 shows clinical DCE breast MRI subtraction images for (a) routine and (b) modified VIBE sequences. The serpiginous artifacts seen in the routine clinical image are removed from the in-phase VIBE image. Linear dark structures radiating from the nipple to chest wall medially may represent Cooper’s ligaments (dark on pre and post images as well as subtraction). Increased background parenchymal enhancement is caused by the delay from bolus administration, since the in-phase images bracketed the routine clinical protocol.

The dark band subtraction artifact is caused by variable enhancement of water and fat when not imaging in-phase (here referred to as “chemical shift artifact of third kind”); the phenomenon has now been characterized for all off-resonance angles. The simulations show there is always an error or reduction in enhancement when not imaging in-phase, which is a strong function of both echo-time and amount of lipid signal. Fat suppression can suffer from both inhomogeneity and well as rapid signal regrowth due to very short T1, leading to both incomplete signal suppression and variability from study to study. Additional variability may be caused by changes in minimum TE from study to study. A temporally and SNR-equivalent sequence can be obtained while imaging in-phase using other acceleration methods such as partial Fourier, thus eliminating both this artifact and enhancement variability.A new subtraction artifact termed chemical shift artifact of the third kind is characterized. Contrast-enhanced gradient-echo imaging can be performed in-phase to eliminate this artifact as well as misleading errors and variability in contrast enhancement.