Multi-angle phase aberration correction for holographic tomography by dual-output U-Net

Area of Science:

• Optics and Photonics
• Computational Imaging
• Biomedical Engineering

Background:

• Holographic tomography reconstructs internal structures using multi-angle complex amplitude fields.
• Optical aberrations limit spatial resolution and imaging quality in tomographic reconstructions.
• Existing aberration correction methods are costly and unreliable.

Purpose of the Study:

• To develop an accurate and robust aberration correction strategy for holographic tomography.
• To enhance the spatial resolution and imaging quality of refractive index distribution retrieval.
• To overcome the limitations of traditional aberration correction techniques.

Main Methods:

• A dual-output-branch 3D convolutional neural network was proposed for end-to-end learning.
• The network optimizes complex-amplitude distributions from multi-angle illumination.
• Inputs include multi-angle phase maps and masks; outputs are Zernike coefficients and aberration phase maps.

Main Results:

• The proposed method achieved accurate aberration correction for tomographic imaging.
• Reconstructed refractive index distributions showed a Structural Similarity Index (SSIM) of 0.9929.
• Microscopic details were effectively preserved in the reconstructed images.

Conclusions:

• The dual-output neural network offers an effective and robust aberration-compensation strategy.
• It unifies aberration correction across multiple illumination angles, eliminating per-angle correction needs.
• The method is suitable for tasks requiring joint optimization across multiple datasets.