All-optical photoacoustic tomography via beam deflection

1. Problem Statement

Photoacoustic Imaging (PAI) combines the high contrast of optical imaging with the deep tissue penetration of ultrasound. However, conventional PAI relies on arrays of piezoelectric pressure transducers to detect acoustic waves. These transducers suffer from inherent limitations:

• Sensitivity vs. Resolution Trade-off: Thin-film transducers (e.g., PVDF) have flat frequency responses but low sensitivity. Crystals (e.g., LiNbO3) and solid transducers (e.g., PZT) have high sensitivity but suffer from mechanical resonances that distort the measured pressure waveforms.
• Fixed Sensitivity: The sensitivity of piezoelectric sensors is fixed by their material properties and pre-amplifier noise, limiting dynamic range.
• Bandwidth Constraints: While optical sensors have high bandwidth, traditional transducers are limited by their mechanical response times.

The authors propose an all-optical approach to overcome these limitations by detecting photoacoustic (PA) waves via optical beam deflection rather than direct pressure measurement. This method aims to reconstruct the initial pressure field from the deflection of probe laser beams caused by density gradients in the acoustic field.

2. Methodology

The proposed framework involves a forward physical model, a novel measurement strategy, and an optimization-based inversion algorithm.

A. Physical Modeling

• Wave Propagation: The acoustic field is modeled using the linear wave equation derived from mass conservation and linearized equations of state in a homogenous compressible fluid.
• Optical Sensing Mechanism: As acoustic waves propagate, they create local density fluctuations that perturb the refractive index of the medium. A probe laser beam traversing this field is deflected.
• Signal Formulation: The deflection angle is proportional to the integral of the density gradient along the beam path. The measured signal ($\beta$) is mathematically formulated as the Radon transform of the pressure gradient field ($\nabla p$), not the pressure field itself.
  $$\beta = \int_{z_L}^{z_U} \mathbf{n}' \cdot \nabla p(\mathbf{x}(z), \tau) \, dz$$
  where $\mathbf{n}'$ is the normal vector of the measurement plane.

B. Inverse Problem Formulation

Reconstructing the initial pressure $p_0(\mathbf{x})$ from limited, noisy gradient measurements is an ill-posed inverse problem. The authors reformulate this as a constrained optimization problem:

1. Preprocessing: The raw sensor data undergoes an inverse Radon transform to recover the directional pressure gradient components ($\partial p / \partial x_m$) on the measurement planes.
2. Optimization Objective: The algorithm seeks to minimize a loss function consisting of:
   • Data Consistency: The difference between the forward-simulated gradients and the preprocessed measured gradients.
   • Regularization: An $L_1$ norm term ($\lambda J$) is applied to the gradient field to promote sparsity and preserve edges (crucial for resolving sharp boundaries).
3. Algorithm: An Iterative Shrinkage-Thresholding Algorithm (ISTA) is used, leveraging the adjoint operator of the wave propagation model to iteratively update the trial pressure gradient field.
4. Final Reconstruction: Once the 3D pressure gradient field ($\nabla \tilde{p}_0$) is reconstructed, the initial pressure field ($\tilde{p}_0$) is recovered by solving the Poisson equation using the Galerkin method (Finite Element Method).

C. Experimental Setup (Simulated)

• Geometry: A cubic domain with sensors arranged on boundary faces.
• Acquisition: A single probe beam and detector are scanned across the boundary, or an array is used. The setup involves rotating the target or sensors to cover a $180^\circ$ range to ensure full angular sampling.
• Simulation: Synthetic data is generated using the $k$-Wave toolbox, incorporating Gaussian noise to mimic experimental conditions.

3. Key Contributions

1. New Imaging Modality: Proposes a fully all-optical PAI system that detects acoustic waves via beam deflection, eliminating the need for piezoelectric transducers and their associated waveform distortions.
2. Mathematical Insight: Establishes that the beam deflection signal corresponds to the Radon transform of the pressure gradient, necessitating a two-step reconstruction (gradient recovery $\to$ pressure integration) rather than direct pressure inversion.
3. Robust Inversion Algorithm: Develops a hybrid reconstruction framework combining:
   • Adjoint-operator-based optimization for solving the ill-posed inverse problem.
   • $L_1$ regularization for edge preservation.
   • Galerkin-based integration to reconstruct the scalar pressure field from the vector gradient field.
4. Performance Benchmarking: Provides a comprehensive numerical validation against conventional transducer-based methods, demonstrating superior stability in noisy environments.

4. Results

The authors validated the method using numerical simulations with three types of sources:

• Shepp-Logan Phantom (65³ grid):
  • The algorithm converged within 50 iterations.
  • The reconstructed pressure field achieved a Peak Signal-to-Noise Ratio (PSNR) significantly higher than the gradient field alone, demonstrating the effectiveness of the integration step.
  • Pearson correlation coefficients between ground truth and prediction exceeded 92% for pressure and 96% for gradients.
• Noise Robustness:
  • Compared to conventional pressure-based detection (band-limited to 0.2–2.0 MHz), the proposed beam-deflection method showed superior Structural Similarity (SSIM) and PSNR across varying noise levels (up to 100% relative error).
  • The method's stability is attributed to the unbiased nature of Gaussian noise and the energy-minimizing properties of the Galerkin formulation, which filters out random fluctuations while preserving structural gradients.
• Complex Structures:
  • Cubic Phantom with Bulge: Successfully resolved sharp edges and tiny protrusions (small tumor analogs) that are often blurred in conventional methods.
  • Spinodal Metamaterials: Accurately reconstructed the lamellar-like microstructure of a 33³ mechanical metamaterial, highlighting the method's potential for non-destructive testing of microscale architectures.

5. Significance and Future Outlook

• Enhanced Sensitivity and Fidelity: By avoiding the mechanical resonances of piezoelectric transducers, this method offers the potential for higher sensitivity (limited only by laser pointing stability) and undistorted waveform capture.
• Broadband Detection: Optical sensors offer bandwidths constrained only by electronics (nanosecond response), potentially capturing higher frequency components than mechanical transducers.
• Clinical and Industrial Potential: The method is compatible with low-cost, portable LED excitation sources, suggesting a pathway toward miniaturized, all-optical clinical imaging systems. It is particularly promising for detecting small tumors and characterizing mechanical metamaterials where high spatial resolution and structural fidelity are critical.
• Limitations & Future Work:
  • Currently limited by computational cost (3D wave simulations) and the need for dense sensor arrays or long acquisition times.
  • The linear Radon model assumes mild refractive index gradients; stronger heterogeneities would require ray-tracing corrections.
  • Future work will focus on experimental validation and exploring deep learning approaches to accelerate the inversion process.

In conclusion, this paper presents a theoretically sound and numerically validated framework for all-optical photoacoustic tomography, offering a promising alternative to traditional transducer-based systems with improved resolution, reduced distortion, and enhanced noise resilience.