Three-dimensional reconstructing undersampled photoacoustic microscopy images using deep learning.

Area of Science:

• Biomedical Imaging
• Optical Microscopy
• Machine Learning

Background:

• Photoacoustic microscopy (PAM) speed is limited by spatial sampling density and data size.
• Undersampling methods increase imaging speed by reducing scanning points but sacrifice spatial resolution.
• Reconstructing 3D PAM data from undersampled scans is challenging.

Purpose of the Study:

• To develop a deep learning-based method for fully reconstructing undersampled 3D PAM data.
• To improve the imaging speed and reduce the data size of PAM systems.
• To evaluate the performance of the proposed reconstruction method against traditional interpolation techniques.

Main Methods:

• A novel deep learning model was developed to reconstruct undersampled 3D photoacoustic microscopy data.
• The method considers data points, data size, and the 3D nature of PAM data.
• Quantitative analyses were performed at various undersampling ratios.

Main Results:

• The deep learning method achieved robust reconstruction of undersampled 3D PAM data.
• The proposed method outperformed interpolation-based reconstruction methods.
• Achieved 80-times faster imaging speed and 800-times lower data size compared to conventional methods.

Conclusions:

• The deep learning-based reconstruction method significantly enhances PAM system performance.
• This approach effectively shortens imaging time and reduces data processing requirements.
• The method is suitable for experimental conditions, offering a practical solution for high-speed 3D PAM.