Clinical application of 3D steady state chemical exchange saturation transfer (CEST) imaging¹ at 3T is hindered by two problems; long acquisition times (~10 mins) and artefacts in the z-spectra interfering with fitting¹. These problems persist because steady state CEST, where preparation and readout of the CEST contrast are interleaved, has not yet been optimised for 3D segmented EPI at 3T. Furthermore, the combination of broad repetitive large off-resonance CEST pulses, long T₂ and short TR render conventional gradient and RF spoiling ineffective². We therefore propose the following improvements to 3D CEST: 1) hexagonal gradient spoiling to reduce artefacts and confounding T2 sensitivity; 2) symmetric z-spectra sampling to maintain a steady state between acquisition of volumes; 3) exchange-rate matched 360° CEST pulses to reduce direct water saturation, T1 sensitivity and increase CNR.A 3D segmented-EPI CEST sequence (see Fig. 1.a for parameters) was implemented in which lower spatial frequencies were acquired last to ensure the presence of steady state CEST contrast in this domain. X and Y gradients, between the CEST and readout modules, were cycled according to a hexagonal pattern, given by $$$G(n)=G_{max} \cdot cos(60\cdot n ) \hat{x} + G_{max} \cdot sin(60 \cdot n ) \hat{y}$$$. This better prevents buildup of unwanted coherence pathways without altering the gradient spoiling moment². The z-spectrum was sampled "center-in" symmetrically, with points in order of descending ppm-offset to avoid the need for T₁ restoration delays between acquired volumes. Two-pool Bloch-McConnell simulations³ were performed to determine the evolution of signal throughout acquisition under the influence of spoiling and determine B1 for optimal contrast. Candidate 3D methods were tested on quantitative ammonium chloride phantoms and on healthy volunteers at 3T (Siemens Verio). A standard 2D CEST sequence⁴ (see Fig. 1.b for parameters) was used as a gold standard for comparison .

Simulations (Fig. 2) revealed two exchange rate regimes: firstly a fast exchange regime which occurs during each CEST pulse, and a slow exchange regime where complete exchange is facilitated by the locked 180° state in the time between CEST pulses (see Fig. 1). By using 360° pulses to purposely avoid saturating the exchange an increase in sensitivity to changes in the exchange rate can be realised (Fig. 2). 360°pulses also reduce direct water saturation and T₁ sensitivity by reducing time spent in the inverted state where increased T₁ recovery occurs. 360° pulses can be matched to different exchange rate regimes by altering the pulse duration whilst maintaining the flip angle. Figure 3 demonstrates that at larger off-resonance frequencies, where the effective field experienced by the water spins reduces and the resulting transverse magnetisation is increased, additional spoiling can be used to accelerate natural T₂ dephasing. This enables steady states to be reached faster and reduces tissue-based T₂ contrast. Through these mechanisms gradient spoiling can contribute to faster acquisition, as k-space is acquired during preparation of the CEST contrast, allowing faster acquisition times without affecting the sensitivity at frequencies critical to CEST.

Figure 4 shows how the reduced buildup of unwanted coherence pathways for hexagonal spoiling significantly reduces artefacts and improves the consistency of the acquired z-spectra, directly benefiting the accuracy of fitting methods. ROI analysis of the phantom (Fig. 5 A,B) demonstrated that even though a smaller slice volume was used for the optimised 3D method it was able to capture a bigger dynamic range amongst compartments. Furthermore the doped compartments in 3D scans had on average 10% higher mean signal and variance was on average 2 times lower for the ROIs (see Fig. 5), indicating higher signal and improved accuracy with the 3D method. Noise outside the phantom was reduced with the 3D method indicating the effectiveness of hexagonal spoiling. In-vivo results (Fig. 5 C,D) also demonstrated a reduction in artefacts for the 3D sequence, with the additional benefit that segmented EPI reduces chemical shift artefacts from fat. Results also showed more homogenous MTR_asymmetry, as expected in healthy volunteers, indicating a reduction in confounding effects as sensitivity to exchange was increased (see Fig. 5 A,B) even with an acquisition time under 3 minutes.

Through the optimisation of B1, implementation of hexagonal gradient spoiling schemes and symmetric z-spectra sampling we were able to acquire whole-brain high-quality 39-offset z-spectra at 3mm isotropic resolution in just 2:59 minutes at 3T, whilst simultaneously improving SNR, CNR, reducing direct water saturation and decreasing sensitivity to T₁ and T₂ confounding effects.