Knowledge about the main inhibitory neurotransmitter γ-aminobutyric acid (GABA) of the brain can give important insight in human brain function. This is valuable not only from a fundamental point of view, but also from a clinical perspective, since imbalances in excitatory and inhibitory processes are connected to several diseases like epilepsy, Parkinson, autoimmune inflammation and stroke [1].

Several studies have demonstrated significant changes in GABA levels between patient populations, however, with substantial variances of GABA levels within groups. In this work we analyzed the physiologic variations in GABA levels of the visual cortex in resting state using a fast and high SNR method for accurate detection of GABA.

Three healthy volunteers were scanned at a high field 7T system (Philips Healthcare, Cleveland, USA). For validation purposes, a phantom (ping-pong ball, with a diameter of 4 cm) containing water-dissolved GABA (6.6mM) and creatine (48mM) was constructed and scanned.

The set-up used was a half-cylinder open transmit coil with 8 transmit channels, combined with 2x16ch receive arrays covering the back of the head (Figure 1). Data acquisition was performed with a MEGA-sLASER [2] sequence with GABA editing, enhanced with macromolecule suppression and FOCI pulses for better B₁ performance [3]. The sequence parameters were: single voxel, TE/TR= 74/5000 ms, in-vivo voxel size of 40x30x30 mm³, phantom voxel size of 17x20x17 mm³, spectral bandwidth: 4000 Hz, a total acquisition time of 2:46 min (16 averages) and 5:30 min (32 averages for bootstrapping). The spectroscopy volume inside the phantom was 6x smaller than the voxel volume in-vivo, due to phantom size limits. This is corrected, by selecting a GABA phantom concentration 6x larger than in-vivo concentration, maintaining a realistic SNR comparison between phantom and volunteers.

An in-house Matlab-based fitting model was used to fit the data. Bootstrapping was performed by selecting different sets of 16 odd-even pairs from a total of 32 pairs (32 averages). For each selected combination, the resulting spectrum was fitted. This was repeated 500 times.

The spectral fitting of creatine and GABA is shown in Figure 2. The SNR and goodness of fit (assessed using the Cramer-Rao lower bounds) is calculated for increasing number of spectral averages. The SNR (Figure 2, middle column) increases with the number of averages. The goodness of fit (Figure 2, right column) stabilizes after 8 averages for creatine and after 16 averages for GABA. The time to acquire 16 averages is 2:46.

Bootstrapping shows the variation of multiple spectroscopy measurements in a phantom (Figure 3, left). The mean and standard deviation of the bootstrap for six different measurements is displayed. The bootstrapping shows a variation for creatine of less than 0.5%, which is similar the total system stability. GABA shows a variation of less than 3% both within bootstraps and between measurements. When measuring in volunteers (Figure 3, right), around 1% variation is found for creatine, which is a factor of 10 higher than the variation for the phantom data. For GABA, a variation within bootstraps is in the order of 8-12%. The difference in GABA concentration between individuals is up to 30%.

The SNR improves with the number of averages. The goodness of fit stabilizes for GABA at around 16 averages (2:46 min). For the phantom data, the variation of the GABA measurements is less than 3% within and between scans. However, for comparable measurements in volunteers a much larger variation is found. Also, the variation between consecutive measurements (10%-30%) is much larger than the variation within the scan (8-12%). This shows that fluctuations of GABA signals are substantially larger than those caused by intrinsic system noise and measurement instabilities. This implies that in the visual cortex at resting state, GABA physiology fluctuates up to 30%.

In conclusion, the results indicate that stability and fitting accuracy of the sensitivity optimized method employed here is sufficient to detect concentration changes of GABA in the visual cortex higher than 3%, within a fast scan time of less than 3 minutes. This brings us closer to more real-time measurements of GABA levels. The in-vivo measurements suggest that the large variation of GABA (up to 30% in this work) between individuals is mainly of physiologic origin. Moreover, individual GABA levels can vary within a temporal resolution of 3 minutes. This brings new insight in the interpretation of GABA spectroscopy measurements.