Identifying out-of-voxel echoes in edited MRS with phase cycle inversion

Abstract

Purpose: To identify the origin of out-of-voxel (OOV) signals based on the coherence transfer pathway (CTP) formalism using signal phase conferred by the acquisition phase cycling scheme. Knowing the CTP driving OOV artifacts enables optimization of crusher gradients to improve their suppression without additional data acquisition.

Theory and methods: A phase cycle systematically changes the phase of RF pulses across the transients of an experiment, encoding phase shifts into the data that can be used to suppress unwanted CTPs. We present a new approach, phase cycle inversion (PCI), which removes the receiver phase originally applied to the stored transients, replacing it with new receiver phases, matching the phase evolutions associated with each unwanted CTP, to identify the OOV signals. We demonstrated the efficacy of PCI using the MEGA-edited PRESS sequence in simulations, phantom and in vivo experiments. Based on these findings, the crusher gradient scheme was optimized.

Results: The simulation results demonstrated that PCI can fully separate signals originating from different CTPs using a complete phase cycling scheme. PCI effectively identified the CTP responsible for OOV signals in phantom experiments and in vivo, though with reduced specificity in vivo due to phase instabilities. Re-optimization of the gradient scheme based on the identified OOV-associated CTP to suppress these signals, resulted in cleaner spectra in six volunteers.

Conclusion: PCI can be broadly applied across pulse sequences and voxel locations, making it a flexible and generalizable approach for diagnosing the CTP origin of OOV signals.

Keywords: Edited MRS; coherence transfer pathways; gradient scheme; out-of-voxel artifacts; phase cycling.

Figure 1.. Edited PRESS pulse sequence with the corresponding complete coherence transfer pathway (CTP) map.

A The pulse sequence diagram of our MEGA-edited PRESS implementation, showing the intended CTP as a green line and an example of an unwanted CTP as a broken black line. Two gradient schemes are shown; gradient pulses included in both are shown in white. Black-filled gradient pulses in our ‘Shared’ gradient scheme are changed to the red pulses in the upgraded scheme, referred to as ‘Two-last, increased-area’ because the duration of these pulses has been increased and there are pulses on two axes in the final delay. B The complete CTP map, where each column corresponds to one of the 81 possible pathways, and each row represents the coherence order following the application of a specific RF pulse (RF_ex, RF_echo1, RF_edit1, RF_echo2, RF_edit2). The acquired signal is conventionally defined by a final coherence order of −1. The broken line in A corresponds to CTP-41 in B.

Figure 2.. MRS voxel positioning.

A Edited spectra were acquired from an isotropic voxel (15 mm)³ voxel in a necked cylindrical phantom. B In vivo edited spectra were acquired from thalamus and medial prefrontal (mPFC) voxels, both (23 × 30 × 23) mm³.

Figure 3.. PCI validation via simulations.

A Ten simulated OOV echoes, each assigned to an arbitrary CTP (from 80 unwanted CTPs), were added to the (three Lorentzian) metabolite signals. B Each signal was modulated by the accumulated phase corresponding to its generative CTP, derived from the complete 1024-step phase cycle of the five RF pulses. C PCI was applied to the individual transients shown in B. The OOV and metabolite signals associated, respectively, with UCTPs and intended CTP were effectively separated. Each CTP is labeled with the vector $\Delta \mathbf{\text{p}}$, which gives the coherence order change across each RF pulse, e.g. the intended signal originates from the CTP characterized by $\Delta \mathbf{\text{p}}=[-\mathrm{1,\; 2,\; 0},-\mathrm{2,\; 0}]$.

Figure 4.. Identification of the CTP responsible for the OOV observed in the phantom edited spectrum.

A Separation of the intended and OOV CTPs using PCI. With this 16-step phase cycling scheme, PCI categorized the CTPs into 16 groups, referred to as PCI rows. All CTPs within a group shared the same receiver phase, and could not be separated with this partial phase cycling scheme. Each PCI row is labeled with a schematic vector containing the elements of $\Delta \mathbf{\text{p}}$ that are shared within the group (where * denotes a lack of specification). B Validation of the CTP identified by PCI. The OOV signal isolated by PCI (in red), which corresponds to CTP-41 $(\Delta \mathbf{\text{p}}=[\mathrm{0,\; 0,\; 0,\; 0},–1\left]\right)$, is overlaid on the signal acquired by switching off all RF pulses apart from the final editing pulse. The strong correspondence in frequency, timing and phase validates the identification of the CTP responsible for the observed OOV echo.

Figure 5.. Identification of the CTP responsible for the OOV observed in the in vivo edited spectrum.

A PCI categorized the CTPs into 16 rows with this partial phase cycling scheme. Each PCI row is labeled with a schematic vector containing the elements of $\Delta \mathbf{\text{p}}$ that are shared within the group (where * denotes a lack of specification). The OOV echoes in the 1.5–2.4 ppm and 3.7–4.3 ppm frequency ranges are most prominent in the 4th PCI row, associated with CTPs characterized by $\Delta \mathbf{\text{p}}=[\mathrm{0,\; 0},*,0,*]$. B Validation of the CTP identified by PCI. The OOV signal isolated as PCI row 4 (in red) is overlaid on the signal acquired by switching off all RF pulses apart from the final editing pulse. The strong correspondence in frequency, timing and phase validates the identification of the CTP responsible for the observed OOV echo.

Figure 6.. Phase stability comparison between phantom and in vivo spectra.

A Overlay of four phantom transients acquired approximately 2 minutes apart (with identical RF pulse phases). B Overlay of four in vivo transients acquired approximately 2 minutes apart (with identical RF pulse phases).

Figure 7.. Comparison between the ‘Shared’ and the ‘Two-last, increased-area’ gradient schemes.

A SUM, and GABA- and GSH-edited difference spectra and their corresponding models and residuals, shown for one volunteer using both gradient schemes (Thalamus above, and mPFC below). Frequency ranges containing OOV signals are highlighted with gray boxes. B OOV signal is quantified using the quality metric QM_OOV for all six subjects, with individual datapoints shown as filled circles, and the corresponding mean ± standard deviation as unfilled squares with error bars. QM_OOV is compared between the two gradient schemes in both regions.