Learning a Filtered Backprojection Reconstruction Method for Photoacoustic Computed Tomography With Hemispherical...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Imaging

Background:

• Photoacoustic computed tomography (PACT) utilizes hemispherical measurement apertures for 3D imaging, particularly in breast imaging.
• Data acquired using these geometries is termed half-scan data, representing half of a full spherical acquisition.
• While half-scan data allows for unique and stable reconstruction, a closed-form solution has been lacking, necessitating computationally intensive iterative methods.

Purpose of the Study:

• To develop a computationally efficient reconstruction method for 3D PACT using half-scan data.
• To address the limitations of slow, optimization-based iterative reconstruction techniques.
• To introduce a learning-based filtered backprojection (FBP) approach for approximating the unknown filtering operation in half-scan PACT.

Main Methods:

• A novel learning-based filtered backprojection (FBP) method, termed the half-scan FBP method, was developed.
• A machine learning approach was employed to approximate the explicit filtering operation required for half-scan data reconstruction.
• The method was validated using extensive numerical simulations with 3D breast PACT models and phantoms, followed by application to experimental in vivo breast PACT data.

Main Results:

• The half-scan FBP method successfully reconstructed accurate 3D images from half-scan PACT data.
• The developed method demonstrated significant computational speed-up compared to traditional iterative reconstruction techniques.
• Reconstruction accuracy was maintained even when the method was applied to data dissimilar to the training dataset, indicating robustness.

Conclusions:

• The learning-based half-scan FBP method provides an accurate and computationally efficient solution for 3D PACT image reconstruction from limited-aperture data.
• This advancement offers a practical alternative to iterative methods, potentially improving the clinical applicability of PACT, especially for breast imaging.
• The method's robustness suggests its potential for broader applications in PACT where data acquisition is constrained.