4614
Detection and Correction of MR EPI Data Corrupted by Spike Noise1Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, 2Department of Diagnostic Radiology, Li Ka Shing Fadiology, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong, China
Synopsis
EPI-based MR images are prone to the spike artifact, which is usually caused by small electrical discharges i.e. sparks that emit radio frequency power within the bandwidth of the scanner receiving system 1 during the MR acquisition. Spike noise causes ripples or stripes covered on the object and can hamper the qualitative or quantitative analysis of the MR images. In this work, we developed a reliable technique that combines Robust Principal Component Analysis 2 (RCPA) with median filtering to robustly detect and correct spike-affected images in human pelvic diffusion weighted imaging (DWI). The overall image quality and the lesion conspicuity were improved after spike removal.
Introduction
Spike noise is common in echo-planar imaging (EPI) -based MRI. It results in stripes or ripples across the images and degrades image quality, as seen in Figure 1(a). Spikes are usually caused by brief bursts of radiofrequency (RF) noise (k-space spikes) during MR data acquisition. Figure 1(b) demonstrates the Fourier transform of the magnitude-squared EPI, which displays the bright and duplicated spike noise signals in the pseudo-k-space (pk-space) 3. Various factors can lead to the high intensity spikes in k-space, including hardware problems of the MR system, such as vibrations during gradient-heavy sequences, loose connections in RF or gradient coils, or improper shielding 2. EPI-based DWI images are particularly prone to spike noise since they are gradient-heavy sequences.
The most commonly adopted method to detect and correct spike noise includes removing k-space signals above the user-defined threshold and interpolating from neighboring voxels that are uncorrupted. The algorithm is simple but not robust due to the signal intensity changes present in central k-space 3.
In this work, we have developed a new technique that applies RPCA 2 and median filtering to robustly detect and correct spike-affected MRI in image post-processing. The algorithm is straightforward and reliable, and can effectively remove spikes from the human pelvic DWI data. The scores on the image quality before and after correction are compared using the Wilcoxon matched-pairs signed rank test.
Methods
Spike Detection: Spike detection is based on the pk-space, the Fourier transform of the magnitude-squared EPI data. It contains the greatly modified spike noise signals 3, where spikes are easier to detect due to the attenuation of center signals that are otherwise highly peaked in k-space. The RPCA 4 algorithm is applied to decompose the pk-space data (M) into a low-rank matrix (L, the spike-free pk-space) and a sparse matrix (S, spike signals), by solving the problem: minL,S ||L||*+λ||S||1 subject to M=L+S. Compared to conventional PCA, RPCA has the advantage of operating directly on the raw data to find a low-rank estimate robust to arbitrarily large outliers 2. M represents the multi-frame data with both corrupted and uncorrupted slices.
Spike Correction: The central region of the pk-space mistakenly classified as sparse is restored using a mask. Signals where the RPCA detects as spikes are then replaced by the k-space signals where a median filtering is applied in its image domain, thus filtering the bright spikes in a straightforward way.
Image Assessment: 13 sets of EPI-DWI data of human pelvis (echo time/ repetition time [TE/TR] = 57ms/ 2840ms, FOV =350 x 290 mm2, matrix = 160 x 129, slice thickness = 5 mm, 30 slices) were scored by two experienced radiologists in consensus to evaluate the image quality before and after spike correction on the level of sharpness, artifacts and lesion conspicuity. This study was approved by the Institutional Review Board and written informed consent was obtained. A 5-point Likert scale was introduced for the image assessment, as shown in table 1. The Wilcoxon matched-pairs signed rank test was performed for the statistical analysis of the image quality before and after spike correction. Statistical significance was set at p<0.05 for the tests.
Results
Figure 2 showed the DWI (b=0) images of human pelvis before and after spike removal on two different subjects. According to the figure, the ripples are greatly reduced after spike correction.
Table 2 showed the statistical result of the image assessment on the 13 sets of EPI-DWI data of human pelvis. The result of the Wilcoxon matched-pairs signed rank test uncovered that our developed method for spike noise correction can improve the overall image quality and the lesion conspicuity of the image. The impact of spike artifacts on the MR images was reduced without affecting the image sharpness.
Discussion
This work allows for the restoration of human pelvic EPI-DWI data that have been corrupted by spike artifacts. We used two steps to correct the spike-affected images: (i) apply RCPA to detect spikes in pk-space and (ii) use median filtering to correct the data in which spikes are detected. The proposed method is robust, direct, and efficient. According to the result of the Wilcoxon matched-pairs signed rank test, the MRI data applying the proposed spike correction algorithm is statistically assessed to achieve enhancement in the image quality and lesion conspicuity.Conclusion
In conclusion, we have developed an effective method to detect and correct spike artifacts caused by RF noise and the improvement in the overall image quality can be beneficial for clinical diagnosis.Acknowledgements
No acknowledgement found.References
1. O Josephs, N Weiskopf, R Deichmann. Detection and correction of spikes in fMRI data. Proc. Intl. Soc. Mag. Reson. Med. 2007;15:3440.
2. Campbell-Washburn A E, Atkinson D, Nagy Z, et al. Using the Robust Principal Component Analysis Algorithm to Remove RF Spike Artifacts from MR Images. MAGNETIC RESONANCE IN MEDICINE, 2016,75(6):2517-2525.
3. Chavez S, Storey P, Graham S J. Robust Correction of Spike Noise: Application to Diffusion Tensor Imaging[J]. MAGNETIC RESONANCE IN MEDICINE, 2009,62(2):510-519.
4. Cande`s EJ, Li X, Ma Y, Wright J. Robust principal component analysis? J ACM 2011;58:1–37.
Figures
Figure 2: Comparison of the image quality before and after spike correction on a representative EPI-DWI (b=0) slice of human pelvis. (a) and (b) represent two different sets of data. As shown in figure 2, the ripples are greatly reduced after spike correction.
