Photoacoustic digital brain and deep-learning-assisted image reconstruction

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Imaging

Background:

• Photoacoustic tomography (PAT) offers high optical contrast and deep penetration for medical imaging.
• Recent applications of PAT in human brain imaging are hindered by skull-induced acoustic attenuation and aberration.
• These distortions degrade photoacoustic signals, complicating accurate image reconstruction.

Purpose of the Study:

• To develop and validate a deep learning approach for correcting skull-induced acoustic aberrations in photoacoustic tomography of the human brain.
• To improve the quality and diagnostic utility of PAT brain images by mitigating signal distortion.

Main Methods:

• Generation of 2D numerical human brain phantoms from 180 MRI and MRA datasets, incorporating scalp, skull, white matter, gray matter, blood vessels, and cerebrospinal fluid.
• Monte Carlo-based optical simulation to determine initial photoacoustic pressure, followed by k-wave acoustic simulations using fluid and viscoelastic models to simulate skull-induced aberrations.
• Training a U-net deep learning model using aberrated photoacoustic sinograms as input and skull-stripped signals as supervision to learn aberration correction.

Main Results:

• The U-net model effectively alleviated acoustic aberrations caused by the skull.
• Significant improvements in the quality of reconstructed PAT human brain images were achieved.
• The corrected images clearly visualized the cerebral artery distribution within the skull.

Conclusions:

• Deep learning-based correction of skull-induced acoustic aberrations is a viable strategy for enhancing PAT brain imaging.
• This method shows promise for improving the visualization of cerebrovascular structures using PAT.
• The U-net approach offers a powerful tool for overcoming a key limitation in applying PAT to human brain imaging.