Image-based ghost correction for general interleaved EPI

Introduction A new image-based ghost correction technique is described for general interleaved EPI. It estimates the phase distortions causing the ghosts as a general function of (x,y), even in the presence of significant overlap of parent image and ghosts. If the phase distortion is assumed to be a function of x (frequency encode direction) only, the new technique is similar to a recently published correction method (1) for general interleaved EPI. However, that publication concluded that the method for general interleaved EPI did not work. It showed that the signal-tonoise ratios (SNR) of the edge pixels were too low to accurately estimate the phase distortion. The new technique utilizes not one, but many pixels in the image, for estimating the phase distortion. The technique requires a user-defined ROI to outline the parent image and identify so-called eligible pixels for estimating the phase distortion. When all eligible pixels are used, the SNR of the phase distortion estimate is comparable to that obtained in single shot EPI. This study provides the first examples of high-quality image-based correction of multiple ghosts in general interleaved EPI. Methods Two new methods, one iterative and one non-iterative, for ghost correction for general interleaved EPI were derived. Detailed derivations can be found in the reference (2). The iterative method is used when the phase distortion is considered to be a general function of x and y. The non-iterative (i.e., closed form) method is a special case of the iterative method, applicable when the phase distortion is considered to be a function of x only. Gradient recalled interleaved EPI was run on a Signa CV/i 1.5 T MR system (GE Medical Systems, Waukesha, WI) with Version 8.2.5 LX software platform. Evennumber interleaved EPI was acquired using an unmodified “epibold” pulse sequence. The odd-number interleaved EPI sequence (3) was implemented from this base pulse sequence code. Both human and phantom imaging was performed. Human imaging was conducted under an approved research protocol, and all subjects gave informed consent. The signal and noise associated with phase distortion estimation was also investigated using computer simulation, as a function of the size of the object relative to the FOV, and the number of interleaves. Results Figure 1 shows the (a) original and (b) corrected reconstruction of the ellipse using the 16-interleaf sequence, with computer simulated data having a relative echo center delay of 1.013 sampling interval. Figures 2-4 shows the (a) original and (b) corrected axial brain images using eight (Figure 2), sixteen (Figure 3), and nine (Figure 4) interleaved axial EPI acquisitions, with phase encoding in the R/L direction. Ghost intensity after correction averaged 7.1±1.8% for even-number interleaved EPI, versus a worse 11.1% for odd-number interleaved EPI. Ghost correction greatly improved image quality both outside and within the parent image. With correction, artifactual image intensity variations, blurred and replicated tissue boundaries, and replicated tissue structures, were substantially reduced (e.g. compare visual cortex in Figures 2 and 3). Computer simulation validated that evennumber interleaved EPI could be corrected just as well as odd-number interleaved EPI.