Deep learning for photoacoustic tomography from sparse data.

Area of Science:

• Medical Imaging
• Computational Imaging
• Biomedical Engineering

Background:

• Fast and accurate image reconstruction is crucial for computed tomography (CT).
• The sparse data problem in photoacoustic tomography (PAT) presents significant reconstruction challenges.
• Existing iterative methods can be computationally intensive.

Purpose of the Study:

• To develop a direct and highly efficient deep learning-based reconstruction algorithm for PAT.
• To address the challenges of sparse data reconstruction in PAT.
• To achieve image reconstruction quality comparable to state-of-the-art methods.

Main Methods:

• A deep convolutional neural network (CNN) was developed for image reconstruction.
• The CNN architecture incorporates the PAT filtered backprojection algorithm in its initial layer, followed by a U-net architecture.
• Network weights are optimized using training data prior to reconstruction.

Main Results:

• The deep learning approach enables image reconstruction through a single CNN evaluation.
• This method avoids time-consuming forward and adjoint problem solutions.
• Numerical results show image quality comparable to iterative approaches for sparse data PAT.

Conclusions:

• Deep learning offers a fast and efficient solution for PAT image reconstruction from sparse data.
• The proposed CNN-based method achieves high-quality reconstructions.
• This approach has the potential to significantly advance PAT imaging capabilities.