Variability of perfusion dark rim artifacts due to Gibbs ringing

Background and aims

Gibbs ringing is a well known source of Dark Rim Artifacts (DRA) in myocardial perfusion imaging [1]. We examine the variability of this artifact. Specifically, we show that Gibbs artifacts are highly dependent on the edge position and that sub-pixel shifts can dramatically change the appearance.

Methods

Sub-pixel shifts were introduced in four in-vivo raw-data perfusion studies, where a DRA was visible. The shifts had a step of 1/8^th of a pixel ranging from 0.125 to 0.875 of the in-plane pixel size. The unprocessed raw-data was phase-shifted using MATLAB before reconstructing it on the scanner using the same image processing as the original data.

The original perfusion study was done on a 1.5 T scanner (Avanto; Siemens): hybrid-EPI sequence with an EPI factor of 4; TR/TE of 5.1/1.7 ms; base resolution 128 pixels; pixel size 2.8 × 2.8 × 8 mm; flip angle 30°; bandwidth 1860 Hz pixel^-1; TI (time of inversion) of 90 ms using a non-selective BIR-4 saturation pulse, TSENSE with R = 2. Perfusion was imaged during first pass of Gd-DTPA; stress induced by adenosine.

The original and shifted magnitude images were compared visually using CMRtools which applies sub-pixel interpolation in the image space.

The DRAs shown were carefully selected so as to not coincide with any known real perfusion defect.

Sub-pixel shifts in short-axis images were also simulated numerically in MATLAB. Images were also reconstructed with image based bicubic interpolation and zero filling with a factor of 4, and compared.

Results and discussion

Figure 1 shows 4 consecutive frames during the arrival of the CA into the myocardium of a particular patient, in the basal slice for two different sub-pixel shifts. The top row (a-d) shows DRAs in the anterior and inferior segments of the subendocardium (white arrows) for the shift where the artifacts were most prominent (0.125 shift). The bottom row (e-h) corresponds to the same frames, but with a half-pixel shift in the vertical direction in relation to the top row (0.625 shift). The DRAs in those regions are highly reduced when compared to the top row.

Figure 1

Figure 1

Figure 2a shows the numerically simulated short-axis image with Gibbs artifacts mainly in the septal, inferior, lateral and anterior regions of the subendocardium. After a diagonal shift of half-pixel, Figure 2b shows a reduction of the artifacts in those regions. Figure 2c–d and Figure 2e–f are identical to Figure 2a–b, but with a bicubic image based interpolation and with zero-filling before FFT respectively. Post-FFT image interpolation shows the same dependency on the position of the edge as the non-interpolated data. In contrast, the artifacts appear consistently without regard to the inplane offsets in the zero-filled data, in agreement with Du et al. [2].

Figure 2

Figure 2

Although the shift selected in Figure 1a–d was the worst-case, it should be recognised that this occurs by chance depending on image plane, in-plane offsets, and the patient's respiratory and cardiac motion. This is consistent with the random nature of DRA occurrence in clinical practice.

Conclusion

The visibility of Gibbs DRAs in perfusion studies is very dependent on the position of the subendocardial wall inside the pixel in the absence of zero-filled pre-FFT interpolation. Position variations from frame to frame in a typical gated perfusion study can explain some of the variability often seen in DRAs. Interpolation by zero-filling prior to enlarged FFT regularizes the DRA appearance.