CNN-based approach for 3D artifact correction of intensity diffraction tomography images

Area of Science:

• * Biomedical imaging
• * Developmental biology
• * Computational imaging

Background:

• * Limited-angle tomographic imaging frequently causes axial elongation artifacts in 3D reconstructions.
• * Accurately recovering the original 3D morphology of biological specimens is challenging due to inherent developmental changes.
• * Intensity-only tomographic data presents unique difficulties for artifact correction.

Purpose of the Study:

• * To present a novel approach for correcting 3D artifacts in intensity-only tomographic reconstructions.
• * To address the issue of axial elongation artifacts in biological samples.
• * To enable more accurate 3D shape recovery for developmental studies.

Main Methods:

• * Development of a novel network architecture combining volumetric and 3D finite object approaches.
• * Application of the framework to time-lapse imaging data of mouse preimplantation embryos.
• * Non-supervised pipeline designed for generalized use across different biological samples and imaging conditions.

Main Results:

• * Successful correction of axial elongation artifacts in 3D reconstructions.
• * Recovery of spherical shapes in developing mouse embryos.
• * Demonstrated efficacy of the combined volumetric and 3D finite object network architecture.

Conclusions:

• * The novel method effectively corrects 3D artifacts in tomographic reconstructions of biological samples.
• * This approach facilitates accurate 3D shape recovery, crucial for understanding developmental processes.
• * The generalized non-supervised pipeline offers potential for diverse biological imaging applications.