Near-zero photon bioimaging by fusing deep learning and ultralow-light microscopy

Area of Science:

• Biotechnology
• Optical Microscopy
• Artificial Intelligence (AI)

Background:

• Reducing specimen irradiance in optical microscopy is crucial for reliability and reproducibility.
• Low irradiance increases photon sparsity (Poisson noise), limiting image quality.
• Current methods to overcome sparsity require long acquisition times, reducing imaging rates.

Purpose of the Study:

• To introduce a novel bioimaging method operating at significantly reduced irradiance and high speeds.
• To overcome the limitations of photon sparsity in biological imaging.
• To enable high-fidelity imaging with minimal light exposure.

Main Methods:

• Developed a near-zero photon bioimaging technique combining a specialized epifluorescence microscope with ultralow background capabilities.
• Utilized artificial intelligence (AI) algorithms to reconstruct biological images from extremely low photon counts (as low as 0.01 photons per pixel).
• Operated the system at kilohertz (kHz) rates, achieving 10,000-fold lower irradiance than standard microscopy.

Main Results:

• Demonstrated high-fidelity reconstruction of multicellular and subcellular structures from near-zero photons per pixel.
• Achieved kHz imaging rates with significantly reduced specimen irradiance.
• Showcased the method's ability to capture features represented by minimal photon counts.

Conclusions:

• Near-zero photon bioimaging offers a paradigm shift for enhancing optical microscopy reliability and speed.
• The AI-driven approach effectively reconstructs images from sparse photon data.
• This technology has potential applications beyond microscopy, including remote sensing and biomedical imaging.