Using convolutional neural network denoising to reduce ambiguity in X-ray coherent diffraction imaging.

Area of Science:

• Coherent Diffraction Imaging (CDI)
• Computational Imaging
• Image Reconstruction

Background:

• Coherent Diffraction Imaging (CDI) often yields ambiguous reconstructions due to inherent data limitations.
• Inconsistent image results arise from varying initial conditions in conventional CDI algorithms.
• Developing methods for reliable image reconstruction is crucial for advancing CDI applications.

Purpose of the Study:

• To introduce a novel method for mitigating image ambiguities in Coherent Diffraction Imaging (CDI).
• To enhance the consistency and reliability of reconstructed images obtained from a single diffraction pattern.
• To leverage deep learning techniques for improved image reconstruction in CDI.

Main Methods:

• Implementation of the Noise2Noise approach combined with neural networks.
• Application of the methodology to hundreds of ambiguous reconstructed images from CDI.
• Utilizing singular value decomposition (SVD) analysis for comparison and validation.

Main Results:

• Significant reduction of ambiguous features, treating them as inter-reconstruction noise.
• Post-Noise2Noise treated images closely approximate the average of multiple reconstructions.
• Demonstrated consistency and reliability in image reconstructions after applying the novel method.

Conclusions:

• The Noise2Noise approach effectively addresses image ambiguities in CDI.
• Neural network integration provides a robust solution for consistent image reconstruction.
• This method offers a pathway to more reliable and interpretable results in coherent diffraction imaging.