Electronic Theses and Dissertations



Document Type


Degree Name

Master of Science


Electrical & Computer Engineering

Committee Chair

Chrysanthe Preza

Committee Member

Mohammadreza Davoodi

Committee Member

Xiaofei Zhang


Recent developments using deep learning (DL) super-resolution in structured-illumination microscopy (SIM), have improved speed in two-dimensional (2D) image restoration and minimized the impact of noise. We have extended this 2D DL technique to 3D by augmenting the 2D convolutional layers to 3D convolutional layers in a 3D U-Net DL network. We demonstrate experimentally that this extension improves lateral and axial resolution in the final 3D restoration compared to the resolution achieved by axially stacking the outputs of the 2D U-Net. To achieve this, we performed 3D processing on data acquired using 3D Structured Illumination Microscopy (3D-SIM) of subcellular biological samples. This is accomplished by splitting the data for training, validation, and testing followed by training a 3D reconstruction machine learning algorithm. We verify lateral and axial super-resolution improvement in 3D U-Net output by analyzing super-resolution quantitative performance metrics and intensity plots of our test data compared to ground truth images obtained from the traditional 3D-SIM restoration. These comparison results are consistent with those that have been reported using conventional techniques for 2D and 3D processing of a similar dataset.


Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest


Open Access