Laser diode photoacoustic point source detection: machine learning-based denoising and reconstruction.

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Compact, portable, low-cost laser diodes (LDs) are advancing photoacoustic (PA) imaging.
• LD-based PA imaging faces challenges with low signal intensity from conventional transducers.
• Temporal averaging, a common solution, reduces frame rates and increases patient laser exposure.

Purpose of the Study:

• To develop a deep learning method for denoising point source PA radio-frequency (RF) data before beamforming.
• To create a deep learning method for automatic point source reconstruction from noisy pre-beamformed data.
• To combine denoising and reconstruction strategies to improve PA imaging in low signal-to-noise ratio conditions.

Main Methods:

• A deep learning model was proposed to denoise PA RF data, enabling beamforming with very few frames (even one).
• A separate deep learning model was developed for automatic reconstruction of point sources from pre-beamformed, noisy data.
• A combined strategy integrating denoising and reconstruction was employed to enhance performance for low signal-to-noise ratio inputs.

Main Results:

• The proposed deep learning methods effectively denoise PA RF data and reconstruct point sources.
• The combined approach significantly improves PA imaging quality, even with minimal input frames.
• This technique offers a viable solution for enhancing LD-based PA imaging without compromising frame rate or increasing laser exposure.

Conclusions:

• Deep learning offers a powerful solution to overcome signal intensity limitations in LD-based PA imaging.
• The proposed denoising and reconstruction methods enable high-quality PA imaging with reduced data requirements.
• This advancement has the potential to make PA imaging more accessible and efficient.