Robust deep learning optical autofocus system applied to automated multiwell plate single molecule localization...

Area of Science:

• Microscopy and Imaging Technologies
• Machine Learning Applications
• Optical Systems Engineering

Background:

• Microscopy experiments, especially those with long acquisition times or automated scanning, are susceptible to defocusing due to thermal or mechanical drift.
• Maintaining precise focus is critical for high-resolution imaging techniques like single molecule localization microscopy (SMLM).

Purpose of the Study:

• To develop a robust, long-range optical autofocus system for microscopy that compensates for environmental and mechanical fluctuations.
• To achieve high axial precision over a wide defocus range for automated imaging applications.

Main Methods:

• Utilized a convolutional neural network (CNN) trained over multiple days to adapt to temporal fluctuations in the optical system.
• Implemented orthogonal optical readouts with separate CNN training data to balance axial precision and autofocus range.
• Characterized system performance using a 1.3 numerical aperture objective lens with a defocus range of +/-100 micrometers.

Main Results:

• The autofocus system demonstrated accuracy well within the 600 nm depth of field.
• The system effectively compensated for thermal and mechanical drift over extended periods.
• Achieved a balance between axial precision and a broad autofocus range.

Conclusions:

• The developed machine learning-based autofocus system provides a robust solution for maintaining focus in demanding microscopy applications.
• The system is suitable for automated slide scanning, multiwell plate imaging, and long-duration experiments.
• Demonstrated successful application in automated multiwell plate single molecule localization microscopy.