Deep Learning-Powered Bessel-Beam Multiparametric Photoacoustic Microscopy.

Area of Science:

• Biomedical Optics
• Neuroimaging
• Medical Physics

Background:

• Multi-parametric photoacoustic microscopy (PAM) is vital for live mouse brain imaging, quantifying hemoglobin concentration, oxygen saturation (sO2), and cerebral blood flow (CBF).
• Conventional Gaussian-beam PAM faces limitations in quantitative accuracy due to a restricted depth of focus, impacting hemodynamic measurements.

Purpose of the Study:

• To develop an advanced PAM system for high-resolution, quantitative, and depth-independent hemodynamic imaging of the live mouse brain.
• To overcome the depth-of-focus limitations of traditional Gaussian-beam PAM.

Main Methods:

• Integration of Bessel-beam excitation with a conditional generative adversarial network (cGAN)-based deep learning approach.
• Development of a combined hardware-software solution for multi-parametric PAM.
• Comparative analysis against conventional Gaussian-beam multi-parametric PAM.

Main Results:

• The novel cGAN-powered Bessel-beam PAM achieved high-resolution quantitative imaging of hemoglobin, sO2, and CBF over an extended depth range in the live mouse brain.
• The new system demonstrated significantly reduced errors (13-58 times lower) compared to the conventional Gaussian-beam system.
• The system effectively addresses the strict focusing requirements for accurate hemodynamic measurements.

Conclusions:

• The developed deep learning-powered Bessel-beam multi-parametric PAM system enhances quantitative accuracy and depth penetration for brain hemodynamic imaging.
• This technology holds promise for large-field functional brain recording, especially over uneven surfaces, and applications like tumor imaging.