Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GPAIR: Gaussian-Kernel-Based Ultrafast 3D Photoacoustic Iterative Reconstruction

Yibing Wang, Shuang Li, Tingting Huang, Yu Zhang, Chulhong Kim, Seongwook Choi, Changhui Li

TL;DR

The paper tackles the slow speed of iterative reconstruction in 3D PACT caused by heavy wave propagation modeling and voxel discretization errors. It proposes GPAIR, a Gaussian-kernel-based ultrafast IR framework with a differentiable forward model, adaptive sub-grid timing via ASSA, and physical/anatomical priors (NPC and VCR) implemented on GPU with Triton. The method achieves orders-of-magnitude acceleration (up to ~872×) and sub-second reconstructions for 8.4-million-voxel volumes, validated on synthetic vascular data and in vivo datasets, with improved PSNR/SSIM and CNR over state-of-the-art methods. This work enables near-real-time 3D PACT and provides modular tools for rapid forward modeling and visualization, advancing clinical feasibility of volumetric PA imaging.

Abstract

Although the iterative reconstruction (IR) algorithm can substantially correct reconstruction artifacts in photoacoustic (PA) computed tomography (PACT), it suffers from long reconstruction times, especially for large-scale three-dimensional (3D) imaging in which IR takes hundreds of seconds to hours. The computing burden severely limits the practical applicability of IR algorithms. In this work, we proposed an ultrafast IR method for 3D PACT, called Gaussian-kernel-based Ultrafast 3D Photoacoustic Iterative Reconstruction (GPAIR), which achieves orders-of-magnitude acceleration in computing. GPAIR transforms traditional spatial grids with continuous isotropic Gaussian kernels. By deriving analytical closed-form expression for pressure waves and implementing powerful GPU-accelerated differentiable Triton operators, GPAIR demonstrates extraordinary ultrafast sub-second reconstruction speed for 3D targets containing 8.4 million voxels in animal experiments. This revolutionary ultrafast image reconstruction enables near-real-time large-scale 3D PA reconstruction, significantly advancing 3D PACT toward clinical applications.

GPAIR: Gaussian-Kernel-Based Ultrafast 3D Photoacoustic Iterative Reconstruction

TL;DR

The paper tackles the slow speed of iterative reconstruction in 3D PACT caused by heavy wave propagation modeling and voxel discretization errors. It proposes GPAIR, a Gaussian-kernel-based ultrafast IR framework with a differentiable forward model, adaptive sub-grid timing via ASSA, and physical/anatomical priors (NPC and VCR) implemented on GPU with Triton. The method achieves orders-of-magnitude acceleration (up to ~872×) and sub-second reconstructions for 8.4-million-voxel volumes, validated on synthetic vascular data and in vivo datasets, with improved PSNR/SSIM and CNR over state-of-the-art methods. This work enables near-real-time 3D PACT and provides modular tools for rapid forward modeling and visualization, advancing clinical feasibility of volumetric PA imaging.

Abstract

Although the iterative reconstruction (IR) algorithm can substantially correct reconstruction artifacts in photoacoustic (PA) computed tomography (PACT), it suffers from long reconstruction times, especially for large-scale three-dimensional (3D) imaging in which IR takes hundreds of seconds to hours. The computing burden severely limits the practical applicability of IR algorithms. In this work, we proposed an ultrafast IR method for 3D PACT, called Gaussian-kernel-based Ultrafast 3D Photoacoustic Iterative Reconstruction (GPAIR), which achieves orders-of-magnitude acceleration in computing. GPAIR transforms traditional spatial grids with continuous isotropic Gaussian kernels. By deriving analytical closed-form expression for pressure waves and implementing powerful GPU-accelerated differentiable Triton operators, GPAIR demonstrates extraordinary ultrafast sub-second reconstruction speed for 3D targets containing 8.4 million voxels in animal experiments. This revolutionary ultrafast image reconstruction enables near-real-time large-scale 3D PA reconstruction, significantly advancing 3D PACT toward clinical applications.
Paper Structure (11 sections, 21 equations, 4 figures, 2 tables, 2 algorithms)

This paper contains 11 sections, 21 equations, 4 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Results
  3. Principle of Gaussian-kernel-based continuous reconstruction
  4. Validation on synthetic vascular data
  5. Validation on in vivo experimental results
  6. Computational performance analysis
  7. Discussion
  8. Methods
  9. Gaussian-kernel-based distribution for the continuous source discretization
  10. Adaptive supersampling alignment for the differentiable model operator
  11. Gradient-based reconstruction framework with physical constraints and anatomical priors

Figures (4)

  • Figure 1: Schematic principle of GPAIR. (a) Gaussian-Kernel-Based Discretization (GKD): transforming discrete voxels into continuous Gaussian spheres. (b) Differentiable Physics Modeling: using Triton GPU kernels for efficient point propagation. (c) Forward Operator $\mathcal{A}(\mathbf{x})$: showing the analytical Gaussian waveform and temporal width. (d) Optimization Framework (GPAIR): the iterative loop with NPC ($\mathbf{x}=\mathbf{z}^2$), forward model, VCR regularization, and Adam optimizer. (e) Vessel Continuity Regularization (VCR): enhancing vessel connectivity using Hessian and TV priors. (f) Adaptive Supersampling Alignment (ASSA): correcting discretization misalignment by upsampling ($f_s=20$ MHz vs. $f_s^{up}=80$ MHz).
  • Figure 2: Validation using simulation study under two system configurations. (a)--(c) Planar array results with 64, 256, and 1024 sensors, respectively. (d)--(f) Hemispherical array results with 64, 256, and 1024 sensors, respectively. Rows: z-MAP, x-MAP, y-MAP, y-Slice; Columns: MB-PD, GPAIR. (g), (h) 3D volumetric rendering of MB-PD and GPAIR results with planar 256 and 1024 sensors. (i), (j) Planar and hemi-spherical array simulation geometry with GT. Scale bar: 10 mm.
  • Figure 3: In vivo 3D reconstruction results for mouse brain, rat kidney, and rat liver. (a)--(c) Reconstruction comparison between MB-PD and GPAIR for each organ. Rows: z-MAP, x-MAP, y-MAP, y-Slice. Scale bar: 5 mm. (d) Schematic of the experimental system with a 1024-element hemispherical detector array. (e) 3D volumetric rendering of GPAIR reconstructions for mouse brain, rat kidney, and rat liver.
  • Figure 4: Quantitative performance metrics comparison between GPAIR and other methods. (a) Computational speedup factors for planar simulations, hemispherical simulations, and in vivo experiments. Dashed lines indicate 200$\times$ and 800$\times$ levels. (b) Absolute reconstruction time for simulation experiments (log scale), with 30 s threshold marked. (c) Absolute reconstruction time for in vivo experiments (log scale), with 1 s threshold marked. (d)--(f) Quantitative image quality metrics (PSNR, SSIM, and MSE) for simulation experiments (Planar array) under varying sensor numbers.