Related Experiment Video
Updated: Aug 5, 2026
08:31
Three-dimensional Optical-resolution Photoacoustic Microscopy
Published on: May 3, 2011
LSMS-Net: Resolving multiple scattering via physics-constrained deep learning for photoacoustic imaging in porous
Wenyi Xu1, Renzhe Bi2, Weiya Xie3
1Institute of Acoustics, School of Physics Science and Engineering, Tongji University, Shanghai 200092, China; A*STAR Skin Research Laboratories (A*SRL), Singapore 138669, Singapore.
Ultrasonics
|July 29, 2026
Summary
Photoacoustic imaging faces challenges in heterogeneous media due to scattering. A new deep learning framework, LSMS-Net, improves deep-tissue imaging of porous structures by addressing scattering effects.
Area of Science:
- Biomedical optics
- Medical imaging physics
- Computational imaging
Background:
- Photoacoustic imaging (PAI) of porous heterogeneous media is limited by optical absorbers acting as acoustic scatterers.
- This dual role causes multiple scattering, wavefront distortion, and frequency-dependent attenuation, degrading image quality.
- Conventional reconstruction algorithms struggle to decouple these effects, leading to blurred deep-tissue images and loss of microstructural details.
Purpose of the Study:
- To develop a novel deep learning framework for high-fidelity photoacoustic imaging of porous heterogeneous media.
- To overcome the limitations of conventional algorithms in handling multiple scattering and wavefront distortion.
- To enable accurate reconstruction of deep microstructural details for potential clinical applications.
Main Methods:
- Proposed a physics-constrained deep learning framework: Layer-Stripping Multi-Scattering Network (LSMS-Net).
- LSMS-Net employs a recursive, top-down strategy to progressively correct raw measurements layer by layer.
- Integrated a Ghost Network for reverberation, an Adaptive Scattering Point Spread Function Layer for distortion, and a Spectral Consistency Constraint for attenuation.
Main Results:
- LSMS-Net accurately reconstructs deep porous structures in numerically simulated and phantom experiments.
- The framework effectively addresses multiple scattering, wavefront distortion, and frequency-dependent attenuation.
- Achieved high-fidelity imaging, preserving high-spatial-frequency microstructural details even under strong scattering conditions.
Conclusions:
- LSMS-Net demonstrates superior performance in photoacoustic imaging of complex scattering media.
- The proposed framework shows significant potential for clinical translation, particularly in diagnosing bone-related diseases.
- Physics-constrained deep learning offers a powerful approach to overcome fundamental challenges in deep-tissue photoacoustic imaging.
