Scan-less machine-learning-enabled incoherent microscopy for minimally-invasive deep-brain imaging.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Optical Imaging

Background:

• Deep-brain microscopy is limited by bulky probes, causing surgical trauma and restricted resolution.
• Existing methods struggle to achieve high-quality imaging deep within brain tissue.

Purpose of the Study:

• To develop a minimally invasive, high-resolution deep-brain microscopy technique.
• To replace traditional microscope objectives with ultra-thin fiber optics for in vivo imaging.

Main Methods:

• Utilized an ultra-thin multi-mode fiber (cannula) as a replacement for microscope objectives.
• Developed a self-consistent deep neural network (DNN) for image reconstruction from fiber-optic signals.
• Employed a novel ensemble method averaging DNN outputs for improved image quality.
• Demonstrated imaging of green-fluorescent-protein (GFP) labelled neurons and GCaMP-labelled C. elegans.

Main Results:

• Achieved single-cell resolution (< 10μm) at depths up to 1.4 mm.
• Demonstrated depth sectioning resolution of 40 μm and a field of view of 200 μm.
• Successfully performed dynamic imaging of moving organisms.

Conclusions:

• The fiber-optic cannula combined with DNNs significantly simplifies deep-brain microscopy.
• This approach offers a promising alternative for in vivo neural circuit investigation.
• The method overcomes limitations of current deep-brain imaging technologies.