Deep learning for precise axial localization of trapped microspheres in reflective optical systems

Area of Science:

• Optical Measurement
• Biophysics
• Machine Learning

Background:

• High-precision axial localization is critical for micro-nanometer optical measurements.
• Existing methods face challenges in calibration efficiency, accuracy, and measurement complexity, particularly in reflective illumination systems due to poor imaging detail clarity.

Purpose of the Study:

• To develop an improved method for high-precision axial localization of microspheres.
• To address the limitations of current techniques, especially in reflective illumination systems.
• To enhance the accuracy and convenience of measurements in various optical setups.

Main Methods:

• Development of a trained residual neural network (ResNet) model.
• Implementation of a convenient data acquisition strategy.
• Application of the method to both reflective and transmission illumination optical tweezers platforms.

Main Results:

• The developed method significantly improves axial localization precision for microspheres.
• The technique is effective in both reflective and transmission illumination systems.
• The method establishes a reliable "positioning point" by leveraging unique sample signal characteristics, reducing systematic errors and enhancing inter-sample localization precision.

Conclusions:

• The trained residual neural network coupled with a novel data acquisition strategy offers a robust solution for precise axial localization.
• This advancement provides greater convenience for measurements in solution environments.
• The method offers higher-order guarantees for force spectroscopy and applications like super-resolution microscopy and surface mechanical property analysis of cells and materials.