Fast, multicolour optical sectioning over extended fields of view with patterned illumination and machine learning

Area of Science:

• Optical microscopy
• Bioimaging
• Machine Learning applications

Background:

• Structured illumination microscopy (SIM) offers high-speed, high-contrast imaging by rejecting out-of-focus light.
• Current SIM techniques are hampered by image reconstruction artefacts and poor performance in low signal-to-noise ratio (SNR) conditions.
• Existing methods struggle with background noise and sample movement, limiting their applicability.

Purpose of the Study:

• To develop a novel structured illumination method for high-contrast, real-time image reconstruction.
• To overcome limitations of existing optical sectioning techniques, specifically artefacts and low SNR performance.
• To create a robust imaging method resilient to background noise and sample motion.

Main Methods:

• Integration of multicolour interferometric pattern generation with advanced machine learning algorithms.
• Development of a real-time image reconstruction pipeline.
• Validation using *in silico* simulations and diverse biological and synthetic specimens.

Main Results:

• Achieved high-contrast, real-time reconstruction of image data.
• Demonstrated robustness against background noise and sample motion.
• Successfully imaged fixed and live biological samples, and synthetic biosystems at 11 Hz across a 44 × 44 μm² field of view, with acquisition speeds exceeding 154 Hz.

Conclusions:

• The combined approach of multicolour interferometric pattern generation and machine learning significantly enhances structured illumination microscopy.
• This method provides robust, high-speed, and high-contrast imaging, overcoming key limitations of previous techniques.
• The validated technique shows broad applicability for advanced bioimaging and biosystem analysis.