Xiaoying Tang
Pronouns: She, her, hers
Research Mentor(s): Dawen Cai
Co-Presenter:
Research Mentor School/College/Department: Cell and Developmental Biology; Biophysics; Neuroscience Graduate Program / Medicine
Presentation Date: April 20
Presentation Type: Oral5
Session: Session 1 – 10am – 10:50am
Room: Breakout Room 4
Authors: Logan Walker
Presenter: 4
Abstract
Recent advances in deep learning enabled the enhancement of image resolution across different microscopy modalities. Fluorescence confocal microscopy as an important imaging modality plays an essential role in biomedical research. While it provides optical sectioning capability to acquire high resolution images in 3D from highly scattered samples, such resolution is difficult to achieve deep inside the tissue due to the limitation of working distance of the high resolving power objectives. Alternatively, we may use a low resolution objective to image deep into the tissue with little distortion. Meanwhile a resolution boosting AI will be able to improve the image resolution. To meet such need, I created a deep neural network and trained following the generative adversarial network (GAN) framework: a generative model (U-net with residual blocks and skip connections) which enhances the input low-resolution image, and a discriminative model (CNN architecture) which returns an adversarial loss to the resolution-enhanced image.
Engineering
