Learning-based Shack-Hartmann wavefront sensor for high-order aberration detection.

Area of Science:

• Optical Engineering
• Biomedical Optics
• Machine Learning in Optics

Background:

• Traditional Shack-Hartmann wavefront sensors (SHWS) face limitations in high-order aberration detection, often requiring complex image processing.
• Accurate wavefront sensing is critical for high-resolution imaging in biological applications, particularly with scattering samples.

Purpose of the Study:

• To develop a learning-based SHWS capable of high-order aberration detection without image segmentation or centroiding.
• To enhance the accuracy, speed, and applicability of SHWS for biological imaging.

Main Methods:

• Generated training datasets using Zernike coefficient amplitudes from biological sample aberrations.
• Employed a machine learning model to predict up to 120th Zernike modes from a single SHWS pattern.
• Validated performance on a 300 µm thick mouse brain slice.

Main Results:

• Achieved prediction of 120th Zernike modes within 10.9 ms with 95.56% accuracy on a personal computer.
• Reduced root mean squared error in phase residuals by ~40.54% compared to traditional SHWS.
• Improved Strehl ratio by ~27.31% and demonstrated a 30-40% median improvement in peak-to-background ratio for biological samples.

Conclusions:

• The learning-based SHWS offers a significant advancement in aberration detection accuracy and speed.
• This method provides a simple, compatible, and fast approach for wavefront sensing in adaptive optics systems.
• The technology holds potential for widespread application in advanced biological imaging and microscopy.