In ¹H magnetic resonance spectroscopy (MRS), macromolecular resonances (MM) overlap with metabolites resulting in inaccurate quantification of the metabolites due to baseline distortion. This effect becomes even more severe in case of short echo times (TE). Previously, single and double inversion recovery techniques^1-3 have been developed in order to address this problem. The main idea is the utilization of the large difference of the T₁ relaxation times between the MM and the metabolites. At ultra-high-field (UHF) frequency and phase alignment based on metabolite cycling (MC) have been shown to be beneficial for the spectral quality of metabolite spectra^4-5. Thus, the purpose of this study was to develop a metabolite-cycled adiabatic double-inversion recovery sequence applicable at 9.4T and to investigate the impact of including MM into quantification of metabolite-cycled 9.4T spectra of the human brain.

The inversion pulse for metabolite nulling (InvP) is required to fulfill two criteria: 1^st) the bandwidth of the InvP is large enough for inversion of the metabolite signals 2^nd) the InvP is robust against B1+ inhomogeneity. For this purpose, a novel type of adiabatic full passage pulse (AFP) was designed. Its amplitude modulation (AM; normalized) was constructed using three Gaussian pulses and its frequency modulation (FM) using a hyperbolic tangent (Fig.1). In particular, the n_i^th sample-point of the pulse amplitude and frequency are given by the following equations:

$$AM(n_i) =α_1\cdot e^{-(\frac{n_i-β_1}{γ_1})^2}+ α_2\cdot e^{-(\frac{n_i-β_2}{γ_2})^2}+ α_3\cdot e^{-(\frac{n_i-β_3}{γ_3})^2}$$

α1=0.590, α2=0.515, α3=-0.207, β1=0.304, β2=-0.004, β3=1.050, γ1=1.420, γ2= 0.615. γ3=1.099

$$FM(n_i) =κ_1\cdot tanh(λ_1\cdot n_i)$$

κ1= 1017 Hz, λ1= -1.89

$$$n_1=-π,n_2= n_1+δn,…,n_{N-1}=n_{N-2}+ δn,n_N=π$$$

$$$ δn=\frac{2π}{N}$$$, N is the number of sample-points.

The behavior of the InvP was simulated for different durations and B₁⁺ inhomogeneity levels. In order to compensate for the different T₁ relaxation times of distinct metabolites at 9.4T ranging from 1000ms to 2000ms⁶ a double-inversion recovery scheme was used. The inversion scheme was implemented along with an MC-STEAM⁴ sequence (Fig 2) enabling correct frequency and phase alignment of all individual spectral averages for spectra with and without metabolite-nulling. The optimum values for the recovery times TI1 and TI2 were calculated using Bloch simulations and ensured sufficient suppression of the metabolites.

This scheme was tested on three healthy volunteers on a 9.4 Tesla scanner (SIEMENS, Germany). MM (metabolite nulled) and metabolite spectra were acquired from a voxel placed in the occipital lobe (TE/TM/TR: 8/50/5000ms. B1+ of InvP=22μΤ, freq. offset of InvP=-850Hz). The individual measured metabolite-nulled spectra were averaged to calculate a MM baseline "template"(Fig. 3). Post-processing was performed using MATLAB (The MathWorks,Inc.) scripts and included the following steps: 1)zero filling, 2)frequency and phase alignment based on unsuppressed water, 3)averaging 4)eddy current correction and 5)coil combination. Only for the MM spectra, the data were apodized with an exponential function (14Hz) for noise reduction. Metabolite spectra acquired without double-inversion were analysed with LCModel⁷ using a simulated basis set^8-10 consisting of 19 metabolites either with or without the MM ‘template’. In the case that MM baseline was introduced in the basis-set, the hidden control paramater of LCmodel dkntmn was set to 0.3. Moreover, an additional singlet was included at 1.73ppm to compensate for a peak systematically appearing at this frequency in our data.

The resulting MM baselines were consistent among the three healthy volunteers except of amplitude differences of M4 resonance (Fig.3). The MM baseline was free of metabolite contamination demonstrating the insensitivity of this method and pulse against a range of T1 relaxation times and B1+ inhomogeneity. The MM template is similar to other published studies^11-13, but the high spectral resolution reveals a splitting of the M5 resonance (Fig 3). Moreover, one additional MM double peak (MX1) was detected (~2.55-2.75ppm). MX1 appears also in other studies¹² but hasn't been reported. These findings were realizable due to UHF and frequency alignment based on MC spectra. The LCModel calculations demonstrate that inclusion of MM template leads to quite flat spline baseline correction without significant residuals (Fig. 4). Contrarily, the baseline calculated by LCModel without prior-knowledge of the MM baseline is quite arbitrary introducing potentials for quantification errors. Furthermore, LCModel quantification without measured MM results in altered metabolite concentrations (in line with Ref.13) with over- or underestimation of certain metabolites being due to the spline baseline's high degree of freedom (Fig. 5).In this study we demonstrated an optimised AFP and double-inversion scheme for metabolite nulling at 9.4T. This is the first study where MC-STEAM is combined with a double-inversion technique. The results showed the advantages of UHF and MC as well as the necessity of the inclusion of MM baseline in the basis-set.