Conformal convolutional neural network (CCNN) for single-shot sensorless wavefront sensing.

Area of Science:

• Optics and Photonics
• Biomedical Imaging
• Machine Learning

Background:

• Wavefront sensing is crucial for deep tissue imaging, enabling spatial light modulators to correct distortions and improve image quality.
• Convolutional neural network (CNN) methods offer speed advantages in sensorless wavefront sensing through single-shot measurements.
• Traditional CNNs struggle with the circular features of point-spread functions (PSFs), limiting their accuracy.

Purpose of the Study:

• To introduce a novel conformal convolutional neural network (CCNN) for enhanced wavefront sensing in deep tissue imaging.
• To address the limitations of traditional CNNs in accurately processing circular PSF features.
• To improve the efficiency and accuracy of sensorless wavefront sensing techniques.

Main Methods:

• Developed a CCNN that utilizes conformal mapping to transform circular PSF features into rectangular ones.
• Pre-processing step reduces the number of convolutional filters required to represent circular features.
• Implemented and tested the CCNN against traditional CNNs in both simulations and experimental setups.

Main Results:

• The CCNN demonstrated a wavefront sensing accuracy improvement exceeding 15% compared to traditional CNNs.
• Conformal mapping enabled more efficient recognition of PSF features by the neural network.
• Experimental validation confirmed the accuracy gains achieved through the CCNN approach.

Conclusions:

• The proposed CCNN significantly enhances wavefront sensing accuracy for deep tissue imaging applications.
• The conformal mapping technique offers a more efficient way to process circular features in CNNs.
• The CCNN method shows great promise for advancing high-speed, high-quality deep tissue imaging.