Computational Imaging Group (CIG) is looking for motivated WashU students for the following research projects for Spring 2020. If you are interested in a given project, please apply by emailing your resume to the project supervisor.

Motion-artifact correction in quantitative MRI

Status: available Description: Noninvasive imaging is the holy grail of medical brain diagnostics. One popular technology used for this purpose is quantitative magnetic resonance imaging (MRI). However, quantitative MRI typically requires long acquisition times and suffers from motion artifacts due to patient motion during scanning. Our goal is to build on the recent algorithms developed in CIG to reduce motion artifacts in quantitative MR images using deep learning. Supervision: Xiaojian Xu (xiaojianxu@wustl.edu), Ulugbek Kamilov Areas: Medical Imaging, Learning for Imaging

Machine learning to see through scattering materials

Status: available Description: Thick biological tissues give rise to the multiple scattering of light waves which complicates noninvasive microscopic imaging. The challenge for existing optical microscopes to overcome scattering has limited the high-resolution imaging to shallow depths. In this project, the goal is to use machine learning to overcome the fundamental scattering limit in tissue imaging. CIG has several contributions to the area, but would like to push the technology beyond its current limit. Supervision: Yu Sun (sun.yu@wustl.edu), Ulugbek Kamilov Areas: Learning for Imaging, Computational Microscopy

Computational imaging using the deep image prior

Status: available Description: Deep image prior (DIP) is an elegant strategy that relies on an untrained convolutional neural network to restore an image. CIG has developed a powerful extension to DIP, called DIP-TV, which significantly improves over the original formulation by offering better imaging quality and more stable performance. This project aims to further improve DIP-TV to enable practical imaging. Supervision: Jiaming Liu (jiaming.liu@wustl.edu), Ulugbek Kamilov Areas: Image Restoration, Learning for Imaging

Nonlocal learning for image restoration

Status: available Description: Deep learning has become the method of choice for image restoration. However, prior to the deep-learning revolution, non-local image restoration methods, such as BM3D, were the dominant algorithms in the field. In this project, the goal is to reconcile deep learning with non-local algorithms and come up with a new generation of powerful image restoration methods. Supervision: Xiaojian Xu (xiaojianxu@wustl.edu), Ulugbek Kamilov Areas: Image Restoration, Learning for Imaging

Recovery guarantees for deep learning

Status: available Description: Traditional sampling theory comes with elegant strategies for both data acquisition and signal reconstruction. Sampling is performed at twice the Nyquist rate, while the reconstruction is achieved with simple low-pass filtering. However, deep learning algorithms do not come with such elegant theoretical background, which we intend to fix in this project. Supervision: Yu Sun (sun.yu@wustl.edu), Ulugbek Kamilov Areas: Compressive Imaging, Learning for Imaging

Generative adversarial priors for image reconstruction

Status: available Description: Generative adversarial networks (GANs) provide a powerful mechanism to represent visually important image features. However, GANs require pretty careful handling during training. In this project, our goal is to develop GAN-based algorithms for image reconstruction. In particular, our goal is to combine GANs with physics-based information for improved performance. Supervision: Jiaming Liu (jiaming.liu@wustl.edu), Ulugbek Kamilov Areas: Compressive Imaging, Learning for Imaging

Adaptive algorithms for computational imaging

Status: available Description: Computational imaging relies on advanced algorithms for forming high-quality images. However, such algorithms are typically fixed and have limited capability to adapt to new data. In this project, our goal is to develop the next generation of algorithms that can adapt “on the go” to the changes in the imaging performance. Supervision: Xiaojian Xu (xiaojianxu@wustl.edu), Ulugbek Kamilov Areas: Compressive Imaging, Learning for Imaging

Computational methods for quantitative brain MR imaging

Status: active (Max Torop) Description: Quantitative magnetic resonance imaging (MRI) is a powerful tool for understanding the brain. However, reliable acquisition of high-quality MR images is diicult due to long scan times and patient motion. The goal of this project is to use machine learning to compensate for lost information in MR acquisition. The project is done in collaboration with WashU Medical School. In the context of this project, we are looking for 2 highly-motivated students comfortable coding and with interest in medical imaging. Supervision: Ulugbek Kamilov Areas: Learning for Imaging, Image Restoration

Next generation Cryo-EM imaging

Status: active (Mingyang Xie) Description: The Nobel Prize in Chemistry 2017 was awarded to the inventors of cryo-electron microscopy (Cryo-EM), which enables high-resolution imaging of biomolecules. Our goal is to improve cryo-EM by developing more advanced imaging algorithms that take into account sophisticated structural constraints. This project is a collaborative project with several external collaborators. The student is expected to have a good grasp of optimization and be comfortable coding in Matlab or Python. Supervision: Yu Sun (sun.yu@wustl.edu), Ulugbek Kamilov Areas: Computational Microscopy, Learning for Imaging

Learning for magnetic resonance imaging (MRI)

Status: active (Weijie Gan) Description: Magnetic resonance imaging (MRI) is a medical imaging technique used in radiology to form pictures of the anatomy and the physiological processes of the body. One of the key practical limitations of MRI is the long duration of data acquisition. The goal of this project is to reduce the scan times by using deep-learning-based imaging techniques. Supervision: Jiaming Liu (jiaming.liu@wustl.edu), Ulugbek Kamilov Areas: Medical Imaging, Learning for Imaging

Asynchronous large-scale optimization for imaging

Status: active (Yiran Sun) Description: Stochastic optimization methods have become increasingly important in optimization for solving large-scale problems arising in data analysis. In this project, we study asynchronous variants of such algorithms in the context of large-scale estimation problems arising in image recovery. Asynchronicity is essential for parallel implementations without any locking, which essential for practical acceleration. The goal is to investigate both theoretical convergence and practical performance for real-time imaging. Supervision: Yu Sun (sun.yu@wustl.edu), Ulugbek Kamilov Areas: Large-Scale Optimization

Learning for material imaging with light

Status: active (Ray Wu) Description: Optical imaging has been widely applied in various fields such as material imaging for nano-chip design. Recent works have shown that deep neural networks (DNNs) are able to invert multiple light scattering and to reconstruct a high-quality two-dimensional (2D) image in a very short time, which is attractive for industrial applications. The goal of this project is to design a novel DNN architecture that can reconstruct 3D objects from the light measurements and deploy the model for material imaging. Supervision: Yu Sun (sun.yu@wustl.edu), Ulugbek Kamilov Areas: Computational Microscopy, Learning for Imaging

Adversarially robust classifiers for image reconstruction

Status: active (Eric Tang) Description: There has been a recent interest in training adversarially robust deep neural nets for various applications. In this project, we explore such neural nets for the problem of image reconstruction. The goal is to obtain a single net that can be used for multiple imaging tasks. Supervision: Xiaojian Xu (xiaojianxu@wustl.edu), Ulugbek Kamilov Areas: Learning for Imaging

Generative Adversarial Networks for Image Restoration

Status: active (Ryogo Suzuki) Description: Forming high-quality images of moving objects is an important problem in photography. The motion of the objects results in blur of an unknown shape that leads to the loss of resolution. The goal of this project is to use the generative adversarial neural nets (GANs) for developing new image restoration algorithms. Supervision: Xiaojian Xu (xiaojianxu@wustl.edu), Ulugbek Kamilov Areas: Learning for Imaging, Image Restoration

Large-Scale Optimization for Computational Imaging

Status: active (Yi Jia) Description: Stochastic optimization algorithms (such as SGD) are increasingly popular for dealing with big datasets. However, exact trade-offs between stochastic and batch algorithms is still not well understood. In this project, the goal is to investigate such trade-offs in the context of large-scale imaging, where an image is reconstructed from a large number of measurements. The project will be conducted in Python using the Anaconda Framework. Supervision: Ulugbek Kamilov (kamilov@wustl.edu) Areas: Large-Scale Optimization

Motion artifact reduction in quantitative brain MRI

Status: active (Nina Woythaler) Description: Magnetic resonance imaging (MRI) is a widely used technique for noninvasive brain imaging. While traditionally MRI is a qualitative modality, recently there has been a growing interest in developing methods to yield quantitative images. However, this type of methods typically require longer acquisition times and come with artifacts due to patient motion during scanning. In this project, our goal is to use deep learning to reduce motion artifacts in quantitative MR images. Supervision: Xiaojian Xu (xiaojianxu@wustl.edu), Ulugbek Kamilov Areas: Medical Imaging, Image Restoration

• Deep Single-Molecule Super-Resolution Imaging (S. Xu, Y. Sun, H. Mazidi, M. Lew, U. S. Kamilov)
• Learning for material imaging with light (R. Wu, Y. Sun, U. S. Kamilov)
• Regularization with ConvNets (J. Liu, Y. Sun, U. S. Kamilov)

• Regularization with ConvNets (J. Liu, Y. Sun, U. S. Kamilov)
• Learning for 3D computational imaging (W. Gan, X. Xu, U. S. Kamilov)
• Convolutional dictionary learning for image reconstruction (F. Long, X. Xu, U. S. Kamilov)
• Unsupervised deep image learning (H. Zou, Y. Sun, U. S. Kamilov)
• Distributed algorithms for nonconvex stochastic optimization (M. Abijaoude, U. S. Kamilov)
• Learning for material imaging with light (R. Wu, Y. Sun, U. S. Kamilov)
• Image restoration with GANs (F. Zhai, Y. Sun, U. S. Kamilov)
• Learning for MRI (J. Xing, X. Xu, U. S. Kamilov)
• Learning for material imaging with light (R. Wu, Y. Sun, U. S. Kamilov)
• Dictionary Learning for Fourier Ptychographic Microscopy (S. Xu, X. Xu, U. S. Kamilov)