Toward Real-Time Giga-Voxel Optoacoustic/Photoacoustic Microscopy: GPU-Accelerated Fourier Reconstruction with Quasi-3D Implementation

Abstract

We propose a GPU-accelerated implementation of frequency-domain synthetic aperture focusing technique (SAFT) employing truncated regularized inverse k-space interpolation. Our implementation achieves sub-1s reconstruction time for data sizes of up to 100 M voxels, providing more than a tenfold decrease in reconstruction time as compared to CPU-based SAFT. We provide an empirical model that can be used to predict the execution time of quasi-3D reconstruction for any data size given the specifications of the computing system.

Quantitative Characterization of Age-Related Changes in Peripheral Vessels of a Human Palm Using Raster-Scan Optoacoustic Angiography

“…The first stage of the OA image analysis included preprocessing. Two-dimensional delay-and-sum image reconstruction in the frequency domain [29] was applied to the raw OA B-scans in two planes XZ and YZ. The reconstructed 3D images were further Frangifiltered [30] using the "vesselfilter" function from the k-Wave software package [31].…”

Evaluation of 10 current image reconstruction algorithms for linear array photoacoustic imaging

“…The reconstruction algorithms were implemented and run on a PC with 12th Intel(R) Core (TM) i9‐12900KF 3.19 GHz CPU with 32 Gigabytes of RAM. The computation time for reconstruction changes nonlinearly with the number of pixels to be reconstructed [97]. For example, the computation time for reconstructing an image with the size 128 × 128, 256 × 256, 512 × 512 or 1024 × 1024 using UBP is 0.05, 0.16, 0.59, or 2.26 s. Although the reconstruction time was calculated using a CPU based PC, it can be reduced significantly by utilizing a GPU.…”

Optoacoustic angiography for noninvasive monitoring of experimental tumor angiogenesis

“…For each received array of 3D data, band-pass filtering was carried out in the frequency range of 1-80 MHz [29] to suppress high frequency noise and low frequency signal component. Then a 2D reconstruction was applied with delay and sum in two orthogonal XZ and YZ planes [30]. The Hilbert transform was applied to the reconstructed and filtered data to represent the data as a maximum intensity projection (MIP).…”