Journal: Biomedical Optics Express

Article Title: Ultrafast filtered back-projection for photoacoustic computed tomography

doi: 10.1364/BOE.540622

Figure Lengend Snippet: Flow chart of the proposed GPU-based FBP implementation. The dashed boxes and arrows denote that the corresponding operations are one-time tasks.

Article Snippet: Compared with the regular CPU-based FBP implementation programmed with MATLAB, the proposed FBP implementation greatly improves the computation efficiency by 439 times, and it only takes 0.38 ms to reconstruct a 2D PA image of 512 × 512 pixels.

Techniques:

Journal: Biomedical Optics Express

Article Title: Ultrafast filtered back-projection for photoacoustic computed tomography

doi: 10.1364/BOE.540622

Figure Lengend Snippet: Blood vessel simulation evaluating the reconstruction accuracy and computation efficiency of the proposed GPU-based FBP implementation. (a)–(c) PA images of the numerical blood vessel phantom reconstructed with the regular MATLAB-based, the regular C++-based, and the proposed GPU-based FBP implementations, respectively. (d)–(f) The differences of (a)–(c). (g) Total image reconstruction time for (a)–(c). (h) Total image reconstruction time under different 2D imaging settings.

Article Snippet: Compared with the regular CPU-based FBP implementation programmed with MATLAB, the proposed FBP implementation greatly improves the computation efficiency by 439 times, and it only takes 0.38 ms to reconstruct a 2D PA image of 512 × 512 pixels.

Techniques: Imaging

Journal: Biomedical Optics Express

Article Title: Ultrafast filtered back-projection for photoacoustic computed tomography

doi: 10.1364/BOE.540622

Figure Lengend Snippet: Multi-sphere simulation demonstrating the superiority of the proposed FBP implementation in reconstruction accuracy over existing FBP implementations in the 2D imaging scenario. (a) The x - y cross-sectional image of the 3D multi-sphere phantom. (b)–(e) Absorption distributions of the multi-sphere phantom reconstructed using different FBP implementations. (f)–(j) Zoomed-in images of (a)–(e) concerning the region inside the red dashed box shown in (a). Reconstruction errors of (b)–(e) and (g)–(j) are shown below the corresponding panels. (k) and (l) Intensity profiles of (a)–(e) concerning the horizontal and vertical blue dashed lines shown in (a), respectively. The intensity profiles of (d) and (e) are scaled intentionally to match the intensity levels of (a)–(c) for better comparison. For convenience, PA images reconstructed with the FBP implementation proposed in this work and the FBP implementation developed by Yuan et al . share the color bar with the ground truth.

Article Snippet: Compared with the regular CPU-based FBP implementation programmed with MATLAB, the proposed FBP implementation greatly improves the computation efficiency by 439 times, and it only takes 0.38 ms to reconstruct a 2D PA image of 512 × 512 pixels.

Techniques: Imaging, Comparison

Journal: Biomedical Optics Express

Article Title: Ultrafast filtered back-projection for photoacoustic computed tomography

doi: 10.1364/BOE.540622

Figure Lengend Snippet: Heart vasculature simulation demonstrating the superiority of the proposed FBP implementation in reconstruction accuracy over existing FBP implementations in the 3D imaging scenario. (a) Maximum intensity projection (MIP) of the 3D heart vasculature phantom along the x -axis. (b)–(d) MIP of the 3D absorption distributions of the heart vasculature phantom reconstructed using different FBP implementations. (e)–(h) The y - z cross-sectional images of the 3D heart vasculature phantom and the reconstructed 3D absorption distributions. Reconstruction errors of (b)–(d) and (f)–(h) are shown below the corresponding panels.

Article Snippet: Compared with the regular CPU-based FBP implementation programmed with MATLAB, the proposed FBP implementation greatly improves the computation efficiency by 439 times, and it only takes 0.38 ms to reconstruct a 2D PA image of 512 × 512 pixels.

Techniques: Imaging

Journal: Biomedical Optics Express

Article Title: Ultrafast filtered back-projection for photoacoustic computed tomography

doi: 10.1364/BOE.540622

Figure Lengend Snippet: Experimental study validating the superiority of the proposed FBP implementation in reconstruction accuracy over existing FBP implementations in the practical imaging scenario. (a)–(d) Absorption distributions of the 3D-printed blood vessel phantom reconstructed using different FBP implementations. (e)–(h) Zoomed-in images of (a)–(d) concerning the region inside the red dashed box shown in (a). (i) Intensity profiles of (a)–(d) concerning the blue dashed line shown in (a). The intensity profiles of (c) and (d) are scaled intentionally to match the intensity levels of (a) and (b) for better comparison. For convenience, PA images reconstructed with the FBP implementation proposed in this work and the FBP implementation developed by Yuan et al . share the same color bar.

Article Snippet: Compared with the regular CPU-based FBP implementation programmed with MATLAB, the proposed FBP implementation greatly improves the computation efficiency by 439 times, and it only takes 0.38 ms to reconstruct a 2D PA image of 512 × 512 pixels.

Techniques: Imaging, Comparison