Calibrated simulations for dynamic focusing of ultrasound through the temporal window

Technical Summary: Calibrated Simulations for Dynamic Focusing of Ultrasound Through the Temporal Window

Problem Statement
Focused ultrasound (FUS) neuromodulation is emerging as a tool for therapeutic clinical applications and causal brain investigation. A critical challenge in deploying multi-element transducer arrays for dynamic steering is the discrepancy between acoustic simulations and measured focal pressures. While single-focus transducers often rely on a single calibration value, multi-element arrays capable of steering across 3D space introduce complex errors due to mechanical or electrical cross-talk and inaccurate assumptions regarding skull transmission. Existing simulation models often fail to account for these factors, leading to significant mismatches (exceeding 20%) between predicted and actual pressure outputs. Furthermore, the microstructural heterogeneity of the skull, particularly trabecular bone scattering, is difficult to simulate accurately with current imaging resolutions, leading to underestimation of attenuation losses. The paper addresses the need for efficient, accurate calibration methods that bridge the gap between simulation and reality to ensure the safety and precision of personalized neuromodulation treatments.

Methodology
The authors developed a calibration framework combining sparse experimental sampling with optimized acoustic simulations across a 3D steerable subspace.

• Target Space Definition: Relative target coordinates for four bilateral deep brain structures (caudate, amygdala, centromedian nucleus of the thalamus, and subthalamic nucleus) were extracted from 157 subjects. These coordinates defined a truncated cone calibration space capturing 95% of potential targets.
• Simulation-Measurement Calibration:
  • Acoustic Holography: Initial attempts to use holography for source pressure mapping revealed significant mismatches (>20%) due to unmodeled element cross-talk, necessitating direct measurement across the steering space.
  • Sparse Sampling: The authors determined that a 2D calibration plane could be accurately characterized by sampling 50 points with an equidistant 6 mm spacing, fitting the discrepancy data with an anisotropic Gaussian function ($R^2 = 0.83$). For 3D space, Bayesian model selection and elbow detection indicated that 20 points were sufficient to model the discrepancy, a finding that generalized across five transducers with identical configurations.
  • Axisymmetry: While simulations suggested axisymmetric fields, experimental measurements of 8 planes at 45° steps revealed modest differences due to manufacturing variability, suggesting that full 3D characterization is necessary for non-axisymmetric arrays.
• Skull Attenuation Optimization:
  • Ex Vivo Validation: Temporal window fragments were extracted from 5 cadaver skulls (yielding 8 usable fragments) and scanned via Zero Echo Time (ZTE) MRI. These fragments were validated against virtual extractions from 157 live subjects' MRIs, confirming similarity in thickness and incidence angle.
  • Attenuation Coefficient Derivation: Using time-reversal simulations to target 40 locations across the calibration space for each fragment, the authors iteratively modified the homogeneous bone attenuation coefficient ($\alpha_0$). The value minimizing the mean square error between simulated and measured peak pressures was determined to be 15.3 dB/cm/MHz.
  • Error Boundaries: The study calculated the 5th and 95th percentile pressure errors, finding that actual pressures could be 44% lower or 33% higher than simulated values using the optimized coefficient.

Key Results

• Sampling Efficiency: The study established that high-resolution field sampling is not required for calibration; 50 points for 2D planes and 20 points for 3D volumes provide sufficient accuracy to model the simulation-reality discrepancy using anisotropic Gaussian functions.
• Optimized Attenuation: An optimized bone attenuation coefficient of 15.3 dB/cm/MHz was derived for the temporal window, aligning with previous experimental estimates (13.3–16.6 dB/cm/MHz at 1 MHz).
• Discrepancy Characterization: The mismatch between simulation and measurement was found to contain a gradual, low-frequency component, with the largest discrepancies occurring at near-field lateral target points.
• Generalizability: The calibration approach and the optimized attenuation coefficient were shown to be applicable across different array designs (concentric circle, Fermat spiral, grid) and multiple transducer units.

Significance and Claims
The paper claims that this calibration framework and the derived optimized temporal window coefficients offer a practical solution to improve the accuracy of reporting and safety assessments for personalized neuromodulation. By moving away from single, conservative estimates or uncalibrated simulations, the authors propose a data-driven approach that:

1. Enhances Safety: Capturing the actual peak pressure rather than the intended focal position ensures conservative safety calibrations for parameters like Mechanical Index (MI) and spatial peak pulse average intensity ($I_{SPPA}$).
2. Improves Accuracy: The optimized attenuation coefficient accounts for the scattering losses inherent in bone microstructure that are typically underestimated in simulations.
3. Facilitates Scalability: The sparse sampling method (20–50 points) makes calibration feasible for individual devices deployed in the field, addressing the logistical challenges of characterizing every point in a steering range.

The authors emphasize that while thermal derivatives are also critical for safety, the calibrated pressure fields derived here can be directly integrated into thermal simulations (e.g., Pennes' bioheat equation) by adjusting the absorption coefficient to represent the non-scattering fraction of attenuation. The study concludes that clearly specifying these attenuation methods is essential for guiding replication efforts and ensuring the safety of steerable ultrasound arrays in human applications.