End-to-end Res-Unet based reconstruction algorithm for photoacoustic imaging.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Photoacoustic imaging (PAI) is a powerful biomedical imaging modality.
• Reconstructing initial pressure distribution from photoacoustic signals is a critical inverse problem in PAI.
• Deep neural networks show promise for addressing complex inverse problems in PAI.

Purpose of the Study:

• To design and train an end-to-end Res-Unet deep learning model for solving the PAI inverse problem.
• To evaluate the performance of the Res-Unet against a model-resolution-based regularization algorithm (MRR).
• To compare the Res-Unet with the state-of-the-art Unet++ architecture in simulation experiments.

Main Methods:

• Development of a residual block-integrated Unet (Res-Unet) architecture.
• Training the Res-Unet model using photoacoustic (PA) signal data.
• Comparative analysis using numerical and physical phantom experiments against MRR.
• Performance evaluation using metrics such as Pearson correlation and peak signal-to-noise ratio (PSNR).

Main Results:

• The Res-Unet achieved over 95% improvement in Pearson correlation and 39% in peak signal-to-noise ratio compared to the MRR algorithm.
• In simulation experiments, the Res-Unet outperformed the Unet++ architecture by over 18% in PSNR.
• The proposed Res-Unet demonstrates superior accuracy and image quality in reconstructing initial pressure distributions for PAI.

Conclusions:

• The Res-Unet is a highly effective deep learning model for solving the inverse problem in photoacoustic imaging.
• This approach significantly enhances the quality of reconstructed images in PAI.
• The Res-Unet offers a promising advancement for PAI applications, outperforming existing methods.