MSD-Net: Multi-scale dense convolutional neural network for photoacoustic image reconstruction with sparse data.

Area of Science:

• Biomedical Imaging
• Medical Technology
• Artificial Intelligence in Medicine

Background:

• Photoacoustic imaging (PAI) is a hybrid technology merging optical and ultrasound imaging strengths.
• Clinical PAI faces challenges like sparse sampling and limited views, causing artifacts and blurring in image reconstruction.
• Standard reconstruction methods struggle with incomplete data, limiting PAI's clinical utility.

Purpose of the Study:

• To develop an advanced convolutional neural network (CNN) for correcting artifacts in 2D photoacoustic tomography (PAT).
• To enhance image reconstruction quality and speed in PAI using a novel deep learning architecture.

Main Methods:

• Proposed a multi-scale dense UNet (MSD-Net) architecture, integrating multi-scale information fusion and dense connections.
• Utilized simulated and in vivo datasets for experimental validation of the proposed method.
• Focused on artifact correction in 2D photoacoustic tomography reconstruction.

Main Results:

• MSD-Net demonstrated superior performance in correcting artifacts compared to standard methods.
• The proposed architecture achieved improved image reconstruction quality from incomplete photoacoustic data.
• Experimental results showed enhanced reconstruction speed alongside improved accuracy.

Conclusions:

• The developed MSD-Net effectively addresses artifacts in 2D PAT reconstruction.
• This deep learning approach offers a promising solution for improving PAI clinical applications.
• MSD-Net enhances image quality and efficiency in photoacoustic tomography.