In 1H-MRSI, spin echo sequences with relatively long TEs (in the range of $$$50$$$-$$$200$$$ ms) are often used because of the need for suppression of water, lipid, and baseline signals. In these schemes, trading off between SNR, acquisition time, and resolution is highly constrained.$$$^1$$$ This work proposes a novel FID-based acquisition method. Using FID excitations enables short acquisition delays ($$$\leq 4$$$ ms), which can improve the detection of fast relaxing and J-coupled metabolites such as glutamate, glutamine, and GABA. FID acquisitions also allow for the use of short TRs without penalizing SNR efficiency$$$^1$$$, enabling rapid data collection. One main challenge of using this acquisition method for 1H-MRSI is the presence of large nuisance (water and lipid) and baseline signals. To address this, we tailor our acquisition scheme to take advantage of the recently proposed subspace based nuisance removal and reconstruction methods.$$$^2$$$ Experimental results demonstrate in vivo 3D 1H-MRSI of the brain at isotropic $$$3$$$ mm resolution with a total acquisition time of about $$$15$$$ minutes.

The proposed pulse sequence is shown in Figure 1. A slab selective $$$\alpha$$$ pulse is used for signal excitation. In order to rapidly collect MRSI data, we use EPSI (echo planar spectroscopic imaging)$$$^4$$$ with bipolar acquisition to increase sampling efficiency. No nuisance signal suppression pulses were applied, which allows for the use of short TR. With a short TR (on the order of $$$200$$$ ms), we can better trade off between resolution, SNR, and acquisition time (e.g. by changing the number of averages). The flip angle is determined by reported in vivo T1 values$$$^3$$$ of the metabolites of interest and the desired TR that is set by resolution and acquisition time.

Unsuppressed nuisance signals are typically $$$3$$$-$$$4$$$ orders of magnitude larger than metabolite signals. We use a subspace approach$$$^2$$$ to remove these signals in post-processing. The data is represented as a sum of water, lipid, metabolite, and baseline signals, each of which resides in a low-dimensional subspace. We take an auxiliary low-resolution scan to obtain water and lipid bases that span the water and lipid subspaces. The nuisance signals in the high-resolution scan are then fitted using these bases, incorporating spatial support (e.g. lipid signals are spatially separate from the brain region) and $$$B_0$$$ field inhomogeneity. The nuisance-removed data is obtained by subtracting the estimated nuisance signals from the data.

We use the recently proposed SPICE (SPectroscopic Imaging by Spatiotemporal corrElation)$$$^5$$$ method to obtain a high-resolution reconstruction of the metabolic signals. An additional low-resolution auxiliary scan is used to obtain the metabolite and baseline bases. With the subspace determined, SPICE allows sparse sampling of the high-resolution MRSI data to further accelerate data acquisition. The baseline signal is separated from the metabolite signal by using the fast signal decay of the baseline and fitting the baseline spectra with splines.$$$^6$$$ Finally, the reconstruction is done by fitting the metabolite and baseline bases to the nuisance-removed data.

MRSI data was collected on a 3T Siemens Trio scanner from a healthy volunteer with IRB approval. The low resolution auxiliary datasets took $$$5$$$ minutes to collect. The scans were done using $$$16\times16\times12$$$ spatial encodings over a $$$240\times240\times96$$$ mm$$$^3$$$ field of view with $$$300$$$ temporal echoes giving a spectral bandwidth of $$$1515$$$ Hz. The high-resolution MRSI scan took $$$10$$$ minutes to collect with a TR/TE of $$$240/4$$$ ms, a flip angle of $$$35$$$ degrees, and an echospacing of $$$1740$$$ $$$\mu$$$s. The data had $$$80\times80\times32$$$ spatial encodings over the same field of view with $$$100$$$ echoes. There was a field map collected in between the scans to allow for $$$B_0$$$ inhomogeneity correction.

In Figure 2 we show the original and nuisance-removed high-resolution dataset, which shows the successful removal of the nuisance signals. In Figure 3 we show the SPICE reconstructed NAA maps. The reconstructed spatio-spectral distribution has very good SNR. Minimal baseline is visible in the spectra. Features such as the ventricles are clearly delineated in the NAA map.

This paper presents a new method (data acquisition, processing, and reconstruction) to enable rapid high-resolution 1H-MRSI using FID acquisitions. To our knowledge, this is the first time MRSI of the brain with isotropic $$$3$$$ mm nominal resolution in $$$15$$$ minutes without water and fat suppression has been demonstrated. The advances in removing nuisance signals from unsuppressed data should be applicable to other spectroscopic studies where no suppression is necessary. Finally, it is possible to extend this acquisition method to incorporate parallel imaging for further acceleration or better pulse design to shorten acquisition delay.