Optoacoustic Tomography Using Accelerated Sparse Recovery and Coherence Factor Weighting

Abstract

Sparse recovery algorithms have shown great potential to accurately reconstruct images using limited-view optoacoustic (photoacoustic) tomography data sets, but these are computationally expensive. In this paper, we propose an improvement of the fast converging Split Augmented Lagrangian Shrinkage Algorithm method based on least square QR inversion for improving the reconstruction speed. We further show image fidelity improvement when using a coherence factor to weight the reconstruction result. Phantom and in vivo measurements show that the accelerated Split Augmented Lagrangian Shrinkage Algorithm method with coherence factor weighting offers images with reduced artifacts and provides faster convergence compared with existing sparse recovery algorithms.

Keywords: model-based reconstruction; optoacoustic tomography; photoacoustic tomography; sparse recovery method..

Figure 1.

Reference U.S. Air Force phantom printed on white paper with black ink, which was embedded in scattering agar (A). The reconstructed image by L2-norm (B). The ASALSA method (C) and the proposed method (D). The subsects in (B–D) are the zoomed-in regions marked in a red rectangle of (A). The line profiles in the horizontal and vertical directions marked in (B) are represented in (E) and (F), respectively.

Figure 2.

Images reconstructed using 128 transducer positions over 270 degrees. The reconstructed image by L2-norm (A). The ASALSA method (B) and the proposed method (D). The subsects in (A–C) are the zoomed-in regions marked in a red rectangle of Figure 1A. The line profiles in the horizontal and vertical directions marked in Figure 1B are represented in (D) and (E), respectively.

Figure 3.

Images reconstructed using 128 transducer positions over 135 degrees. The reconstructed image by L2-norm (A). ASALSA (B) and the proposed method (D). The subsects in (A–C) are the zoomed-in regions marked in a red rectangle of Figure 1A. The line profiles in the horizontal and vertical directions marked in Figure 1A are represented in (D) and (E), respectively.

Figure 4.

Reconstructed images of the tissue-mimicking agar phantom, which includes a hollow cavity filled with air and 2 high absorbing areas. Reference image of the phantom (A). The reconstructed image by L2-norm (B). The ASALSA method (C) and the proposed method (D). Arrows indicate artifacts caused by reflections or scattering of the acoustic waves, which are significantly reduced with the proposed method.

Figure 5.

Reconstructed images of the mouse kidney from 256 transducer positions over 270 degrees. The reconstructed image by L2-norm (A). The ASALSA method (B) and the proposed method (C). The subsects in (A–C) are the zoomed-in regions marked in a red rectangle of Figure 5A. The line profiles marked by the red line in Figure 5A are represented in (D).

Figure 6.

Reconstructed images of the mouse kidney from 128 signal positions over 135 degrees. The reconstructed image by L2-norm (A). The ASALSA method (B) and the proposed method (C). The subsects in (A–C) are the zoomed-in regions marked in a red rectangle of Figure 6A. The line profiles marked by the red line in Figure 6A are represented in (D).