Recover Biological Structure from Sparse-View Diffraction Images with Neural Volumetric Prior.

Area of Science:

• Biophysics
• Optical Imaging
• Computational Biology

Background:

• Label-free optical microscopy allows studying cellular metabolism in native environments.
• Current 3D reconstruction methods require numerous 2D images, limiting dynamic biological sample imaging.
• Reconstructing semi-transparent samples from sparse views presents significant challenges due to limited viewing angles and measurements.

Purpose of the Study:

• To develop a novel method for high-fidelity volumetric reconstruction of semi-transparent biological samples from sparse-view microscopic images.
• To overcome the limitations of existing optical tomography techniques in capturing dynamic biological processes.
• To enable real-time 3D imaging of label-free biological samples.

Main Methods:

• Development of Neural Volumetric Prior (NVP), integrating explicit and implicit neural representations.
• Incorporation of the physical prior of diffractive optics into the reconstruction model.
• Validation using both simulated data and experimentally captured microscopic images.

Main Results:

• NVP significantly reduces the number of required images by approximately 50-fold.
• Processing time is reduced by 3-fold compared to previous methods.
• State-of-the-art performance is maintained with sparse-view data.

Conclusions:

• NVP is the first technique enabling volumetric reconstruction of label-free biological samples from sparse-view microscopic images.
• This advancement paves the way for real-time 3D imaging of dynamically changing biological samples.
• NVP enhances the capability to study cellular dynamics in their native environments non-destructively.