Self-Supervised Upsampling for Reconstructions With Generalized Enhancement in Photoacoustic Computed Tomography

Area of Science:

• Biomedical imaging
• Medical physics
• Computational imaging

Background:

• Photoacoustic computed tomography (PACT) is a promising hybrid imaging modality.
• Limited detector numbers in PACT cause spatial aliasing and streak artifacts.
• Increasing detectors is costly and often impractical.

Purpose of the Study:

• To develop a novel self-supervised learning approach to overcome PACT under-sampling challenges.
• To improve image quality and reduce artifacts in PACT reconstructions.
• To provide a cost-efficient solution for PACT data acquisition.

Main Methods:

• A self-supervised neural network was trained on downsampled PACT channel data.
• The method leverages data similarity at various downsampling rates for interpolation.
• No densely-sampled ground truth data is required for training.

Main Results:

• The method successfully performed accurate interpolation of under-sampled PACT data.
• Validation was confirmed through simulations, phantom experiments, and animal studies.
• The technique effectively reduced spatial aliasing and streak artifacts.

Conclusions:

• The proposed self-supervised learning approach offers an effective solution for PACT under-sampling.
• This method enhances PACT capabilities in a cost-efficient manner.
• The technique is applicable across multiple PACT systems and imaging scenarios.