Deep Learning Empowered Fresnel-based Lensless Fluorescence Microscopy.

Area of Science:

• Neuroscience
• Optical Engineering
• Computational Imaging

Background:

• Miniaturized fluorescence microscopy is crucial for in-vivo neuroscience.
• Lensless imaging offers a path toward next-generation miniaturized microscopes.
• Deep Learning (DL) has shown promise in computational imaging but is underexplored in fluorescence microscopy.

Purpose of the Study:

• To develop and evaluate a miniaturized lensless fluorescence microscope prototype.
• To assess the efficacy of Deep Learning (DL) for enhancing image quality in lensless fluorescence microscopy.
• To explore the potential of Fresnel amplitude masks in advanced microscopy.

Main Methods:

• Developed a lensless fluorescence microscope prototype utilizing an optimized Fresnel amplitude mask.
• Generated a computational dataset based on experimental system calibration for DL evaluation.
• Applied DL algorithms to process and reconstruct images from the microscope prototype.

Main Results:

• Achieved high-quality imaging with a structural similarity index of 89% using the DL-assisted method.
• Reduced the least absolute error by 63% compared to classical imaging models.
• Demonstrated the state-of-the-art performance of the prototype, highlighting the potential of amplitude masks.

Conclusions:

• The DL-assisted lensless fluorescence microscope provides high-quality imaging suitable for in-vivo studies.
• Fresnel amplitude masks are critical for developing robust, flat, miniaturized microscopes.
• This technology advances long-term brain circuit and disease studies in freely moving animal models.