Deep-learning-based multi-transducer photoacoustic tomography imaging without radius calibration.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Optical Imaging

Background:

• Pulsed laser diodes are crucial excitation sources in photoacoustic tomography (PAT) due to their affordability, small size, and high pulse repetition rates.
• Integrating multiple single-element ultrasound transducers (SUTs) enhances PAT imaging speed.
• Accurate PAT image reconstruction necessitates precise knowledge of each SUT's radius, a process that can be complex and time-consuming.

Purpose of the Study:

• To develop a novel deep learning approach for PAT image reconstruction that eliminates the requirement for SUT radius calibration.
• To improve the accuracy and efficiency of PAT image reconstruction.
• To enhance the peak signal-to-noise ratio (PSNR) without degrading overall image quality.

Main Methods:

• A deep learning model, specifically a fully dense U-Net incorporating a convolutional long short-term memory block, was developed for PAT image reconstruction.
• The convolutional neural network was trained and validated on a test dataset.
• In vivo imaging experiments were conducted to confirm the network's performance in a real-world biological context.

Main Results:

• The proposed deep learning network successfully eliminated the need for radius calibration in PAT image reconstruction.
• A significant improvement in peak signal-to-noise ratio (PSNR) of approximately 73% was achieved.
• The network reconstructed high-quality PAT images without compromising diagnostic information, as validated by in vivo imaging.

Conclusions:

• The developed deep learning method offers a robust solution for simplifying PAT image reconstruction by removing the dependency on transducer radius calibration.
• This approach enhances image quality and imaging speed, making PAT more accessible and efficient.
• The findings demonstrate the potential of deep learning to advance biomedical imaging techniques like photoacoustic tomography.