Radio-frequency pulse design in local rotating frame in magnetic resonance imaging

This paper proposes a local rotating frame formalism for MRI radio-frequency pulse design that simplifies magnetization dynamics by nullifying the total longitudinal field in each voxel, thereby offering new theoretical insights and significantly reducing computational time for iterative and multi-coil pulse optimization.

The Gist

The Big Picture: A New Way to Watch Spins Dance

Imagine you are trying to choreograph a dance for a massive crowd of people (the atomic spins in your body) to create a specific picture (an MRI image). In a standard MRI, you use radio waves (the music) and magnetic gradients (the dance floor instructions) to tell the crowd where to move.

Usually, scientists try to calculate this dance while the crowd is spinning wildly because of the Earth's magnetic field and the MRI machine's main magnet. It's like trying to teach a dance routine while everyone is on a rapidly spinning merry-go-round. The math gets messy, the calculations take a long time, and it's hard to predict exactly how the dancers will react when the music gets loud (large "tip angles").

The Author's Solution:
Seung-Kyun Lee proposes a clever trick: Change the perspective.

Instead of watching the dancers from a stationary spot while they spin on the merry-go-round, imagine you hop onto the merry-go-round yourself. But here's the twist: you spin at the exact same speed as the dancers in your specific spot. Suddenly, relative to you, the dancers aren't spinning wildly anymore. They are standing still, waiting for your instructions.

This is the "Local Rotating Frame." By mathematically hopping onto this spinning frame, the author removes the "noise" of the strong magnetic field. The problem becomes simpler, slower, and much easier to solve.

---

Key Concepts Explained with Analogies

1. The "Local Rotating Frame" (The Personal Dance Floor)

In a standard MRI, the magnetic field changes depending on where you are in the machine (like a gradient).

• The Old Way: You calculate the dance for the whole room at once, accounting for the fact that the floor is tilting and spinning differently in every corner. It's chaotic.
• The New Way: The author says, "Let's pretend the floor is flat and still for each dancer individually." We mathematically cancel out the spinning effect of the magnetic field for every single voxel (tiny 3D pixel) in the image.
• The Result: The radio waves (the music) now look like they are rotating at different speeds for different dancers, but the dancers themselves are calm. This makes the math much simpler because we don't have to fight the "spinning" force anymore.

2. The "Stereographic Projection" (Flattening the Ball)

The paper uses a mathematical trick called a "Riccati form" or "stereographic projection."

• The Analogy: Imagine the magnetization of a spin is a ball. Usually, we track the ball's position in 3D space (up/down, left/right, forward/back). It's hard to solve equations for a ball rolling on a sphere.
• The Trick: The author projects that 3D ball onto a flat 2D piece of paper (like projecting the Earth's surface onto a flat map).
• Why it helps: On this flat map, the complex, non-linear rules of the spin dance turn into a much simpler, almost straight-line relationship. It turns a messy, curved problem into a clean, linear one that is easier to solve.

3. The "Residual Phase" (The Leftover Spin)

When you do a slice-selective pulse (telling only a specific slice of the body to dance), the spins don't just stop perfectly; they often wobble a little bit at the end, creating a "residual phase" (a leftover spin).

• The Old Problem: Scientists usually fix this by guessing and checking, adjusting the gradient magnets after the fact.
• The New Insight: Using the new frame, the author derived a formula that predicts exactly how much this wobble will happen based on how hard you pushed the dance (the tip angle).
• The Benefit: You can now calculate the perfect "rewind" magnet adjustment mathematically before you even start the scan, ensuring a cleaner image.

4. Parallel Transmit (The Orchestra)

Modern MRI machines often have multiple radio coils (like an orchestra with many instruments) to fix image distortions. Designing the music for all these instruments at once is incredibly hard.

• The Iterative Fix: The author shows that because the math is simpler in the new frame, you can use a "guess-and-check" loop much faster.
  1. Guess the music.
  2. Simulate the dance.
  3. See where the dancers are out of sync.
  4. Adjust the music.
• The Speed Boost: Because the simulation is faster (see below), you can run this loop many more times in the same amount of time, leading to a much better final result.

5. The Speed Boost (The Time Machine)

This is perhaps the most practical claim of the paper.

• The Problem: Simulating how spins move in a strong magnetic field is like running a high-speed video game. To get it right, you have to update the frame rate thousands of times per second. If you miss a frame, the simulation crashes or becomes inaccurate.
• The Solution: In the "Local Rotating Frame," the "background noise" (the strong magnetic field) is gone. The spins move slowly and calmly.
• The Analogy: It's like switching from filming a hummingbird's wings (which requires a super-fast, expensive camera) to filming a turtle walking (which you can film with a standard camera).
• The Result: The author demonstrates that this method can make the computer simulation 4 times faster without losing accuracy. This is huge for "Optimal Control," where the computer has to run thousands of simulations to find the perfect pulse.

Summary of Claims

The paper does not claim to invent a new MRI machine or a new medical treatment. Instead, it claims to have found a better mathematical lens through which to view the physics of MRI.

1. Simplification: By changing the frame of reference, the complex equations governing spin motion become simpler and more linear.
2. Insight: This new view explains why certain existing methods work better than expected and provides a formula to predict "wobble" (residual phase) in slice selection.
3. Speed: It drastically reduces the time needed to simulate these pulses, which is critical for designing complex pulses for modern, multi-coil MRI machines.
4. Accuracy: It allows for better design of pulses that flip spins by 90 degrees (a standard MRI task) and helps in designing pulses for larger flips (180 degrees) by stacking them together.

In short, the author didn't change the music or the dancers; they just found a better way to watch the show, making it easier to write the choreography and faster to rehearse it.

Technical Summary

1. Problem Statement

In Magnetic Resonance Imaging (MRI), spatially selective Radio-Frequency (RF) pulse design involves determining RF waveforms that, when applied synchronously with gradient fields, produce a specific spatial magnetization profile.

• Conventional Approach: Standard design methods operate in a global rotating frame that cancels the static magnetic field ($B_0$). In this frame, the dynamics of nuclear spin magnetization are governed by the combined effects of time-varying gradient fields and RF fields.
• Challenges:
  • Complexity: The Bloch equations in the global frame are complex, especially for large tip angles (non-linear regime) and multidimensional excitations.
  • Simulation Speed: Iterative design methods (e.g., optimal control, parallel transmit optimization) require repeated Bloch simulations. These simulations are computationally expensive because the strong gradient fields necessitate very small time steps to maintain numerical accuracy, particularly for spins far from the isocenter (out-of-slice voxels).
  • Theoretical Limitations: Existing linear approximations (like Small Tip Angle theory) often fail or require restrictive conditions (e.g., specific gradient trajectories or Hermitian symmetry) to be valid for tip angles exceeding 30°.

2. Methodology: The Local Rotating Frame

The author proposes a novel theoretical framework based on a local rotating frame transformation.

• Definition of the Frame: Unlike the global frame which rotates at a fixed Larmor frequency ($\omega_0 = -\gamma B_0$), the local rotating frame rotates at a position- and time-dependent angular rate determined by the total longitudinal magnetic field ($B_z = B_0 + G(t) \cdot r + \delta B_0$).
• Transformation: The phase of the frame is defined as the integral of the instantaneous Larmor frequency:
  $$-\phi(r, t) = -\gamma \int_0^t (B_0 + G(t') \cdot r + \delta B_0) dt'$$
• Effect on Dynamics:
  • In this new frame, the longitudinal magnetic field is effectively zero for every voxel.
  • The gradient field is "integrated out" of the differential equations.
  • The remaining dynamics are governed solely by the transverse RF fields. While the apparent RF field appears to rotate at a voxel-dependent rate, the spin dynamics become slower and simpler because the dominant gradient term is removed.
• Mathematical Formulation:
  • The Bloch equation is transformed into a simpler form involving only transverse fields.
  • The magnetization is mapped to a complex variable $f$ (stereographic projection) or spinor parameters ($\alpha, \beta$), converting the Bloch equation into a Riccati form or a series of Cayley-Klein parameters.
  • This leads to a series expansion solution where the first term represents a linear response, and higher-order terms represent non-linear corrections.

3. Key Contributions

A. Theoretical Insights and Derivations

1. Alternative Derivation of SLR Transform: The paper provides a new derivation of the Shinnar-Le-Roux (SLR) algorithm. In the local rotating frame, the separation of gradient and RF effects is exact via frame transformation, removing the need for the "hard pulse approximation" traditionally used in SLR derivations.
2. Analytical Calculation of Residual Phase: The author derives an analytical expression for the residual phase winding in slice-selective excitation at large tip angles (up to 90°).
   • Using a second-order expansion, the paper shows that the residual phase slope depends quadratically on the tip angle ($\theta_t$).
   • This allows for the precise calculation of the necessary rewinder gradient area to correct phase errors without empirical tuning.
3. Validity of Linear Approximation: The study demonstrates that the linear approximation (Small Tip Angle theory) remains a valid leading-order approximation for tip angles up to 90° in this frame, provided the non-linear terms are treated as corrections. This challenges the conventional view that linear theory fails significantly above 30°.

B. Practical Applications in Parallel Transmit (pTx)

1. Iterative Inversion for Multi-Coil Design: The framework enables an iterative inversion method for designing RF pulses for parallel transmit systems.
   • The method starts with a linear solution (Fourier design) and iteratively adds corrective RF pulses based on the error between the simulated and target magnetization.
   • It successfully designs phase-coherent 2D excitation pulses for 8-channel systems with tip angles up to 90°.
2. Fast Bloch Simulation:
   • By eliminating the strong gradient field from the numerical integration, the method allows for much larger time steps ($\Delta t$) without accumulating numerical error.
   • Numerical error in the local frame grows as $\theta_t^2 / N$ (where $N$ is the number of steps), whereas in the global frame, error grows rapidly in regions with high gradients (out-of-slice).
   • This results in a ~4x speedup in simulation accuracy for a given time step, or the ability to use coarser time steps for equivalent accuracy.

C. Optimization Acceleration

• The fast simulation capability is applied to optimal control-based RF pulse design.
• In a 180° inversion pulse optimization scenario, the use of the local rotating frame for the line-search subroutines resulted in a 2.5-fold reduction in total optimization time compared to conventional global frame simulations.

4. Results

• Simulation Accuracy: In 2D excitation simulations (8-channel pTx), the iterative method converged effectively for 60° and 90° tip angles. For 120°, convergence was marginal, indicating the need for more rigorous optimal control algorithms for very large angles.
• Phase Correction: The analytical formula for residual phase slope (Eq. 33) matched Bloch simulation data closely for Hamming-windowed sinc pulses with time-bandwidth products of 4 and 12, accurately predicting the non-linear phase dispersion.
• Performance Gains:
  • Speed: Method 2 (Local Frame) was ~4x faster than Method 1 (Global Frame) for equivalent accuracy.
  • Optimization: Total CPU time for optimal control design was reduced by ~60% (2.5x speedup).

5. Significance and Impact

• Unified Framework: The local rotating frame provides a general, algorithm-agnostic framework that simplifies the mathematics of RF pulse design, making it applicable to various methods (SLR, iterative inversion, optimal control).
• Enabling High-Field MRI: As MRI moves to higher fields (7T and above), $B_1$ inhomogeneity and specific absorption rate (SAR) constraints make patient-specific, multi-coil pulse design critical. The speed and accuracy improvements offered by this method are essential for making such designs clinically feasible.
• Theoretical Clarity: The work clarifies why linear design methods often work better than expected for intermediate tip angles and provides a rigorous mathematical basis for correcting non-linear phase errors in slice selection.
• Future Utility: The framework facilitates the integration of relaxation ($T_1, T_2$) into simulations and can be combined with hardware acceleration (GPUs) to further speed up the design of complex, high-resolution spatial pulses.

In conclusion, Seung-Kyun Lee's work transforms the problem of RF pulse design by shifting the reference frame to one where the dominant gradient field is mathematically removed. This yields significant theoretical insights into non-linear spin dynamics and offers substantial practical benefits in the speed and accuracy of designing complex pulses for modern parallel transmit MRI systems.