All Thayer Events

PhD Thesis Defense: Rubio Shang

Nov

11

Thursday
9:00am - 11:00am EST

Videoconference

For Info on how to attend this video conference, please email rubio.shang.TH@dartmouth.edu.

Computational and Deep-learning Approaches for Improved Image Reconstruction and Analysis

Abstract

Computational imaging is developed to combine imaging system design and algorithm development. The imaging system design is for the ease to build the imaging forward model and for the ease to solve the imaging inverse problem mathematically. The algorithm development is for better image reconstruction and analysis with the prior information from the imaging system design. It can achieve imaging tasks that cannot be achieved in conventional imaging, such as high-image-quality imaging, phase imaging, super-resolution imaging, ultrafast imaging, large space-bandwidth product imaging etc.

In this work, several computational imaging approaches are proposed for improved image reconstruction and analysis with the special imaging system design and advanced algorithm development. A sparsity-based photoacoustic image reconstruction approach is developed for the reconstruction of the initial pressure distribution in a linear-array ultrasound transducer and direct measurement of the imaging forward model. The compressed ultrafast holography is proposed for reconstruction of the optical field of dynamic objects at an ultrafast imaging rate. A two-step-training deep learning framework is proposed for computational imaging to learn the prior information of the imaging system through training and then further optimize the image estimated from the learnt imaging system priors. A Bayesian convolutional neural network is proposed for uncertainty quantification so that the accuracy of the predictions from the deep learning networks can be quantified in practical imaging applications where the ground truths are unknown. The developed computational imaging approaches can be applied to more diverse imaging models beyond the imaging applications in this work.

Thesis Committee

  • Geoffrey P. Luke, PhD (Chair)
  • Brian W. Pogue, PhD
  • Anne E. Gelb, PhD
  • Jinyang Liang, PhD (External)

Contact

For more information, contact Theresa Fuller at theresa.d.fuller@dartmouth.edu.