Authors: Tayler Fearn, Caroline Torgensen, Vern Hart
Mentors: Vern Hart
Insitution: Utah Valley University
Coherent diffraction imaging (CDI) is a newly developed modality used to measure phase shifts introduced by fine-scale structures in cells. These phase shifts can be used to distinguish healthy and malignant cells, providing a diagnostic marker for early cancer detection. However, this process, in which diffracted light interferes with incident light, requires collecting scattered photons at large angles, representing high spatial frequencies and short wavelengths. The highest frequencies, needed to reconstruct small details in cells for improved image quality, occur at distances of several centimeters from the central bright fringe. As such, these signals are faint and difficult to collect experimentally. We propose the use of deep learning to synthetically extrapolate diffraction patterns at large distances, where measurements are difficult. In prototyping this method, we will present results produced by a generative adversarial network (GAN), trained using existing data of watercolor paintings to preform style transfer and image extrapolation. This will be an essential step in working towards the larger goal of developing GAN’s that can accurately extrapolate diffraction images.