Walsh mode based neural network for adaptive optics in two-photon microscopy

Thesis

Microscopy is a vital tool in biomedical research. In deep tissue two-photon imaging, scattering induces severe wavefront distortions, which degrade image quality. Adaptive optics is used to restore image quality; however, conventional continuous modes are insufficient for compensating complex wavefronts.

This thesis introduces neural-network–based sensorless adaptive optics methods using pixelated Walsh modes for complex wavefront compensation. These methods are designed to be robust and more efficient than conventional algorithms in improving image quality. The neural networks exploit the unique geometrical properties of Walsh modes and, in the machine-learning-assisted wavefront-sensorless adaptive optics control method (MLAO), also utilize aberration structure information extracted from extended images. Both approaches can outperform the conventional 2N+1 algorithm and the sequential 3N algorithm in efficiency for image quality improvement while maintaining robust performance under various conditions.

Neural-network-based sensorless adaptive optics methods using Walsh modes show significant potential for deep biomedical imaging. Their high efficiency reduces aberration correction time and specimen exposure while maintaining high image quality, and their robust performance ensures versatility across diverse imaging conditions. These results indicate that combining Walsh modes with neural networks provides a powerful framework for advancing sensorless adaptive optics in complex wavefront compensation.

Authors: Xiao, Y

• DOI: 10.5287/ora-w4vwdxdjx
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: Walsh function; neural network; adaptive optics