DeepInMiniscope: Deep learning-powered physics-informed integrated miniscope

Area of Science:

• Biomedical imaging
• Microscopy
• Deep learning applications

Background:

• Mask-based integrated fluorescence microscopy offers compact, snapshot 3D imaging with scalable fields of view (FOV).
• Existing computational reconstruction algorithms struggle with efficiency for large-scale data.
• A need exists for improved reconstruction speed and quality in integrated microscopy.

Purpose of the Study:

• To develop DeepInMiniscope, a miniaturized integrated microscope with a custom optical mask.
• To create an efficient physics-informed deep learning model for rapid 3D object reconstruction.
• To demonstrate enhanced reconstruction quality and speed for large-scale imaging data.

Main Methods:

• Development of a miniaturized integrated microscope with a novel optical mask.
• Implementation of a physics-informed deep learning algorithm for 3D object reconstruction.
• Reconstruction of large volumetric data (4mm x 6mm x 0.6mm).

Main Results:

• DeepInMiniscope achieved significant improvements in reconstruction quality and speed over traditional methods.
• The deep learning model successfully reconstructed large object volumes.
• Near-cellular resolution imaging of neuronal activity in awake mouse cortex was demonstrated.

Conclusions:

• DeepInMiniscope represents a substantial advancement over existing integrated microscopes.
• The system offers scalable, large-FOV, high-speed 3D imaging capabilities.
• This technology holds promise for diverse biomedical research applications requiring compact imaging devices.