The J-difference editing technique is widely used in the in vivo proton MR spectroscopy for the detection of gamma-aminobutyric acid (GABA).^1,2 During the lengthy time for the edit-on and edit-off scans, which is typically more than 13 minutes, B₀ drift and subject motion may happen.^3,4 This will not only cause frequency and phase drifts in the spectrum but also alter the linewidth and lineshape of the spectra. When the linewidths/shapes of edit-on and edit-off spectra do not match, the J-difference spectrum will have residue of Cr peaks, overlapping with the edited GABA peaks; even when edit-on/off spectra match each other, the linebroadening and distortion will deteriorate the quality of the difference spectrum. Frequency and phase drift correction has been recently suggested^3,4, in addition to the conventional spectral processing procedures. However, the mismatch of the edit-on and edit-off spectra and the lineshape distortion still remain. In this study, we used the partially suppressed water signals to match and to transform the spectra so as to improve the quality of the J-edited spectrum.

Data acquisition All data were acquired on a 3 T scanner MR750 (General Electric Healthcare, WI) running on DV24 system software and with an 8 channel coil. Phantom experiments were performed on a standard spectroscopic phantom without GABA (GE, MRS HD Sphere). The MRS data were acquired using the commercial J-editing sequence with PRESS localization (MEGA-PRESS).² The scan parameters are as follows: voxel size: 2 x 2 x 2 cm³, TR/TE = 68 ms, spectral width = 2000 Hz, spectral data points = 1024, number of scan excitation = 256 x 2 (for edit-on and edit-off). In vivo experiments were performed on 3 healthy subjects in comply with the Institutional Review Board. No deliberate subject motion was induced.

Data processing The 8 channel MRS data were combined for each of the individual scans, and the resultant signals were frequency and phase corrected. All edit-on or all edit-off scans were combined, respectively, from which the partially suppressed water signals were extracted. The water signals were used to deconvolve and transform the spectra^5,6:

$$S_{on}=S_{on}^0L/W_{on}$$

and

$$S_{off}=S_{off}^0L/W_{off}$$

where S⁰ is the original signal, W is water signal, and L is the lineshape function. L was chosen as a Gaussian function with linewidth equal to that of the water signal. For comparison purposes, L was also chosen to be W_on or W_off for $$$S_{off}$$$ or $$$S_{on}$$$, respectively. We refer this latter case as "match-only". Finally, the J-difference spectrum was obtained. To facilitate comparison, J-difference spectrum was also obtained using conventional approach without spectral deconvolution and lineshape transform.

We used signal to noise ratio of NAA as an indicator of spectral quality. The results show that the improvement of the edited spectra of the proposed approach depends on the degree of match between the original FIDs of edit-on and edit-off scans. When the summed edit-on and edit-off spectra perfectly match each other, only the “deconvolution + lineshape transform” can improve the quality of the difference spectrum. When disagreement exists between them, both the “match-only” and “deconvolution + lineshape transform” procedures can improve the quality of the edited spectra (Table 1. N₁ is the noise in signal free region, and N₂ is noise calculated around 3 ppm. See also Figure 1). Of the three in vivo spectra processed by the proposed method, the ratio of NAA to N₁ increased by 21.85%, 17.32%, and 13.65%, respectively. The increase of peak heights of Glx (glutamate + glutamine) and GABA was also clearly observed (Figure 2).The present results show that using partially suppressed water signals to deconvolve the spectra can eliminate the differences between edit-on and edit-off spectra and the lineshape transform can further improve SNR, thus potentially benefiting the detection of GABA and Glx. The performance of the proposed approach depends on several factors: (1) water residual, which should not be too small; otherwise the estimated water signal is affected by noise; (2) accurate modeling of the water signal, which was achieved by an singular value decomposition based method⁷, and (3) the lineshape and decay rates in lineshape transform, which were chosen as Gaussian lineshape that has lower Cramer-Rao Lower Bounds than Lorentzian⁵ and with the same linewidth as the water signal, respectively. Partially suppressed water signal can be used to deconvolve the J-editing proton MRS signals and thus, together with lineshape transform, can improve the quality of the spectrum. Further study will focus on the quantitative aspects of the proposed method.