MR Spectroscopy without Water Suppression using the Gradient Impulse Response Function

This paper demonstrates that the Gradient Impulse Response Function (GIRF) can effectively correct eddy current-induced sidebands in non-water-suppressed proton MR spectroscopy, enabling the recovery of metabolite signals and revealing that water suppression typically causes magnetization transfer effects that underestimate metabolite concentrations.

The Gist

The Big Idea: Hearing the Whisper in a Roaring Room

Imagine you are trying to listen to a quiet conversation (the metabolites, which are the tiny chemical signals in your brain) happening in a room where a jet engine is screaming right next to you (the water signal).

In standard brain scans (MRS), scientists usually try to "turn off" the jet engine so they can hear the conversation. They do this by using special radio pulses to suppress the water signal. But this has a few downsides:

1. It takes extra time and energy.
2. The act of "turning off" the engine sometimes accidentally mutes the conversation too (a side effect called magnetization transfer).
3. You lose the ability to use the jet engine's roar as a reference point to measure exactly how loud the conversation is.

The Problem: If you just leave the jet engine running, the noise creates weird, ghostly echoes (called sidebands) that distort the room, making it impossible to hear the conversation clearly.

The Solution: This paper introduces a clever trick called GIRF (Gradient Impulse Response Function). Instead of trying to silence the jet engine, the authors figured out exactly how the engine makes the room vibrate, and then they used a computer to "cancel out" those vibrations in the recording.

---

How It Works: The "Noise-Canceling Headphone" Analogy

Think of the MRI machine like a high-tech recording studio. When the machine switches its magnetic gradients (the "switches" that help take the picture), it causes the metal coils inside the machine to vibrate slightly, like a speaker cone. These vibrations create tiny, shifting magnetic fields that mess up the sound recording.

1. The Calibration (The "Test Drive"):
   Before scanning a person, the researchers do a one-time "test drive" using a water-filled ball (a phantom). They run a specific test to map out exactly how the machine vibrates when it switches. They create a map of the noise (the GIRF). This is like recording the exact frequency and pattern of a jet engine's roar so you know exactly what to expect.

2. The Scan (The "Recording"):
   They scan a person's brain without turning off the water signal. The raw data comes in looking messy, with the jet engine roaring and those ghostly echoes everywhere.

3. The Correction (The "Magic Filter"):
   Using the map they made during the test drive, the computer predicts exactly how the machine's vibrations will distort the signal during the actual scan. It's like having a noise-canceling headphone that knows the exact pattern of the noise before it even happens. The computer subtracts this predicted distortion from the recording.

The Result: The jet engine is still there (the water signal is visible), but the ghostly echoes are gone. Now, the quiet conversation (the metabolites) is crystal clear, and you can use the jet engine's volume as a perfect ruler to measure how loud the conversation is.

---

What They Found

The researchers tested this on 8 people using two different scanning methods (semi-LASER and MEGA-PRESS).

• The Good News: The "noise-canceling" worked beautifully. They could see the brain chemicals clearly without needing to silence the water first. The results were almost identical to the traditional method, but with a bonus: they could use the water signal as a built-in ruler.
• The Surprise: When they compared the new method to the old method, they found that the old method was actually under-reporting the amount of a chemical called Creatine (and others in some scans).
  • Why? Because the old method's "turning off" pulses were accidentally muting the Creatine signal (magnetization transfer).
  • The Takeaway: The new method might actually give us more accurate numbers for what's really happening in the brain.

Why This Matters

• No More "Turn Off" Buttons: We don't need to waste time and energy trying to silence the water.
• Better Accuracy: We avoid accidentally muting the chemicals we are trying to study.
• Built-in Ruler: We can use the massive water signal to calibrate our measurements, making the data more reliable.
• One-Time Setup: You only need to do the "test drive" calibration once for a specific machine. After that, you can use this trick for any brain scan on that machine.

In a Nutshell

This paper is like upgrading from a method where you have to mute the radio to hear a whisper, to a method where you keep the radio on but use a smart algorithm to filter out the static. The result is a clearer picture of the brain's chemistry, with fewer errors and more useful data.

Technical Summary

1. Problem Statement

Proton magnetic resonance spectroscopy ($^1$H-MRS) typically requires water suppression to prevent the overwhelming water signal from obscuring metabolite signals. However, retaining the water signal offers significant advantages:

• Internal Referencing: The water signal can serve as an internal concentration reference.
• Correction & Monitoring: It facilitates automated frequency/phase correction, motion monitoring, and tracking of BOLD (line-narrowing) effects in functional MRS.
• Reduced Artifacts: Removing water suppression RF pulses reduces total RF power deposition (SAR), shortens minimum TR, and eliminates magnetization transfer (MT) effects caused by saturation pulses.

The Core Challenge: Without water suppression, eddy currents and mechanical vibrations induced by gradient switching create time-dependent magnetic field perturbations ($B_0$) during the Free Induction Decay (FID). These perturbations cause antisymmetric sidebands on the water resonance, which distort the spectral baseline and obscure metabolite signals. Existing solutions (e.g., metabolite cycling, two-scan subtraction, or pre-emphasis) are often sequence-specific, increase scan time, or fail to fully correct sidebands away from the isocenter.

2. Methodology

The authors propose a post-processing correction method using the Gradient Impulse Response Function (GIRF) to predict and remove gradient-induced field perturbations.

• GIRF Calibration:

  • A one-time calibration scan is performed on a phantom (NiCl$_2$-doped water) using a 3T Prisma scanner.
  • The method employs an optimized thin-slice approach with triangular input gradients and 2D phase encoding to measure the system's impulse response.
  • The GIRF characterizes the linear time-invariant (LTI) relationship between input gradient waveforms and the resulting magnetic field perturbations (including eddy currents and mechanical vibrations) across spherical harmonic terms (up to first-order).

• Correction Pipeline:

  1. Simulation: Gradient waveforms ($i_j(t)$) for specific sequences (semi-LASER and MEGA-PRESS) are simulated based on the specific voxel orientation and acquisition parameters.
  2. Prediction: The measured GIRF ($H_{j,m}(\omega)$) is applied in the frequency domain to the simulated waveforms to predict the time-dependent magnetic field components ($o_{j,m}(t)$) during the FID.
  3. Phase Calculation: The predicted field is integrated to calculate the accumulated phase error ($\phi(t)$) at the voxel center.
  4. Correction: The measured complex FID signal is corrected by subtracting the predicted phase error, effectively removing the sidebands.
  5. Water Removal: After correction, the strong water resonance is modeled (using a Voigt lineshape) and subtracted from the spectrum, leaving the metabolite signals intact.

• Experimental Setup:

  • Subjects: 8 healthy participants and a SPECTRE phantom.
  • Sequences: Semi-LASER (motor cortex) and MEGA-PRESS (visual cortex) at 3T.
  • Comparison: GIRF-corrected non-water-suppressed (NWS) spectra were compared against standard water-suppressed (WS) spectra.
  • Robustness Check: Semi-LASER scans were repeated after system heating (via a diffusion tensor imaging sequence) to test GIRF stability.

3. Key Contributions

• Generalized Correction: Unlike previous hardware-specific or sequence-specific methods, this approach uses a system-level model (GIRF) that is independent of the spectroscopy sequence. It requires no sequence modification or additional hardware.
• Post-Processing Solution: The correction is applied retrospectively, allowing for the recovery of metabolite signals in NWS data without increasing scan time or complexity during acquisition.
• Quantitative Validation: The study provides a rigorous comparison of metabolite quantification between NWS (corrected) and WS data, isolating the effects of magnetization transfer.

4. Results

• Sideband Suppression: GIRF correction successfully reduced the amplitude of antisymmetric sidebands to levels comparable to the spectral baseline. The resulting NWS spectra closely resembled standard WS spectra.
• Metabolite Quantification:
  • Phantom Data: Excellent agreement was observed between NWS and WS data, with no significant differences in metabolite concentrations (except minor residual eddy current features at specific frequencies, handled by baseline fitting).
  • In Vivo Data: Systematic increases in metabolite concentrations were observed in NWS data compared to WS data.
    • MEGA-PRESS: Total Creatine (tCr) increased by 6.9% (Ratio: 1.069 ± 0.039).
    • Semi-LASER: Larger increases were seen across multiple metabolites, with tCr showing the largest increase of 53.5% (Ratio: 1.535 ± 0.160).
• System Heating: Metabolite ratios remained consistent before and after system heating, indicating the GIRF correction is robust to short-term thermal drifts within a session.
• Cause of Discrepancy: The authors attribute the concentration increases in NWS data to magnetization transfer (MT) effects. In WS acquisitions, saturation pulses on the water pool transfer saturation to metabolites (especially tCr), artificially lowering their measured signal. NWS acquisitions avoid this, revealing the true (higher) concentration.

5. Significance and Conclusion

This study demonstrates that non-water-suppressed MRS is feasible for routine use if gradient-induced sidebands are corrected via GIRF.

• Clinical/Research Impact: The method enables the use of the water signal for internal referencing and real-time monitoring while avoiding the MT artifacts that plague traditional water-suppressed protocols.
• Accuracy: The "elevation" of metabolite concentrations in NWS data is not an error but a correction of the MT bias inherent in WS methods. This suggests that NWS-GIRF may provide more accurate absolute quantification of metabolites like creatine.
• Scalability: While demonstrated at 3T, the approach is theoretically extendable to other field strengths and sequences, provided a GIRF calibration is performed.

Limitations: The method relies on the LTI assumption and a one-time calibration; long-term system drift or hardware changes may require recalibration. Additionally, the water modeling step requires careful handling to avoid baseline distortions.