Implicit neural representation for sparse-view photoacoustic computed tomography.

Area of Science:

• Biomedical Imaging
• Computational Imaging
• Medical Physics

Background:

• High-quality photoacoustic computed tomography (PACT) imaging necessitates high-channel systems to prevent aliasing artifacts.
• Sparse-view acquisition in PACT can lead to artifacts, prompting research into model-based (MB) and deep learning-based reconstruction methods.
• Existing discrete representation methods for PACT reconstruction are ill-conditioned, susceptible to errors, and worsen with higher resolution.

Purpose of the Study:

• To propose an implicit neural representation (INR) framework for PACT image reconstruction using ring transducer arrays.
• To address the ill-conditioning and error-proneness of discrete representation methods in PACT.
• To improve image fidelity and artifact suppression in sparse-view PACT acquisition.

Main Methods:

• Representing the initial heat distribution as a continuous function using a multi-layer perceptron (MLP).
• Training the MLP weights in a self-supervised manner by minimizing the discrepancy between measured and predicted photoacoustic (PA) signals.
• Utilizing the trained network to map PA images by inputting spatial coordinates.

Main Results:

• The INR method demonstrated superior performance over universal back-projection and MB methods in preserving image fidelity and suppressing artifacts under identical acquisition conditions.
• Experimental data showed that the INR method improved signal-to-noise ratio (generalized contrast-to-noise ratio) by 1.1-24.0 dB (0.037-0.716) compared to other methods.
• Simulation and phantom experiments validated the effectiveness of the INR approach.

Conclusions:

• The INR framework offers significant value for high-quality PACT image reconstruction, particularly with sparse data acquisition.
• INR demonstrates potential for reducing the overall complexity of PACT systems.
• This continuous representation approach effectively mitigates artifacts and enhances image quality in PACT.