MicroWorkshop | Computational Imaging – July 15, 2020
Emrah Bostan | Computational Microscopy: From iterative optimization to machine learning and back again
Deep neural networks have emerged as effective tools for computational imaging including quantitative phase microscopy of transparent samples. Traditionally, the phase reconstruction inverse problem is solved by minimizing a least-squares loss that is based on the physics of the problem. This physics-based optimization approach is fundamental to phase imaging and has the immediate advantage that prior assumptions on the images can be directly integrated through regularization. To reconstruct phase from intensity, current approaches, however, rely on supervised learning with training examples; consequently, their performance is sensitive to a match of training and imaging settings. In this talk, we present a new hybrid approach to phase microscopy by using an untrained deep neural network within a physics-based inverse problem formulation. Our approach does not require any training data and simultaneously reconstructs the sought phase and pupil-plane aberrations by fitting the weights of the network to the captured images. We will demonstrate experimental quantitative phase reconstructions from through-focus images blindly (i.e. no explicit knowledge of the aberrations).
Emrah Bostan’s Bio
Emrah Bostan is an assistant professor at the Informatics Institute at the University of Amsterdam (UvA). Before joining the faculty at UvA, he was a post-doctoral researcher in the Computational Imaging Lab at UC Berkeley under the direction of Laura Waller. He was also affiliated with the Berkeley Artificial Intelligence Research Laboratory and Berkeley Center for Computational Imaging during his time in Berkeley. He received his M.Sc. and Ph.D. degrees at École polytechnique fédérale de Lausanne (EPFL) advised by Michael Unser. In broad terms, his research combines optical physics, computer vision, and machine learning to co-design hardware and computation elements of next-generation imaging systems. for acquiring, reconstructing, analyzing, understanding, and displaying high-dimensional visual data.
Katie Bouman | Designing the Future of Black Hole Imaging
This talk will present the techniques used to produce the first image of a black hole from the Event Horizon Telescope, as well as discuss machine learning methods currently being developed to design the next generation telescope. The Event Horizon Telescope is a network of telescopes scattered across the globe that resolved structure on the scale of a black hole’s event horizon for the first time in 2019. To recover more information from the black hole, there is an ongoing effort to design the next generation telescope array by choosing the locations of additional telescopes to add to the network. In this talk, I will discuss a new physics-constrained, fully differentiable, autoencoder that we have developed to optimize the telescope design. This framework allows us to optimize the placement of telescopes, even when sensor correlations and atmospheric noise present unique challenges. We demonstrate results broadly consistent with expectation and draw attention to particular structures preferred in the telescope array geometry that can be leveraged to plan future observations and design array expansions.
Katie Bouman’s Bio
Katie Bouman is a Rosenberg Scholar and Assistant Professor of Computing and Mathematical Sciences (CMS) and Electrical Engineering at Caltech in Pasadena, California. Her research focuses on computational imaging: designing systems that tightly integrate algorithm and sensor design, making it possible to observe phenomena previously difficult or impossible to measure with traditional approaches. Her group at Caltech combines ideas from signal processing, computer vision, machine learning, and physics to find and exploit hidden signals for both scientific discovery and technological innovation. Prior to starting at Caltech, she was a postdoctoral fellow with the Event Horizon Telescope, which published the first picture of a black hole in April of 2019. She received her Ph.D. in Electrical Engineering and Computer Science at the Massachusetts Institute of Technology (MIT) in 2017. She previously received a B.S.E. in Electrical Engineering from the University of Michigan, Ann Arbor, MI in 2011, and an S.M. degree in Electrical Engineering and Computer Science from MIT in 2013.
Roarke Horstmeyer | Rapid coronavirus testing with computational microscopy
Rapid testing for the coronavirus and a better understanding of its resulting effects are immediately pressing challenges. One effective strategy that we have recently identified is based upon high-resolution imaging of peripheral blood smears via digital optical microscopes and subsequent computational analysis of blood cell morphology from acquired image data. Our very preliminary results demonstrate that it is possible to achieve approximately 85-90% sensitivity and specificity of COVID-19 diagnosis, with the potential to be higher, by computationally analyzing the scanned images from a smear created by a drop of blood. Our machine learning network is also highlighting morphological features of interest that may drive future intuition regarding the interaction of this new disease with blood cells. This short talk will summarize these recent findings and discuss new imaging strategies that can hopefully improve diagnostic accuracy.
Roarke Horstmeyer’s Bio
Roarke Horstmeyer is an assistant professor within Duke’s Biomedical Engineering Department. He develops microscopes, cameras and computer algorithms for a wide range of applications, from forming 3D reconstructions of organisms to detecting neural activity deep within tissue. His areas of interest include optics, signal processing, optimization and neuroscience. Most recently, Dr. Horstmeyer was a guest professor at the University of Erlangen in Germany and an Einstein postdoctoral fellow at Charitè Medical School in Berlin. Prior to his time in Germany, Dr. Horstmeyer earned a Ph.D. from Caltech’s electrical engineering department in 2016, a master of science degree from the MIT Media Lab in 2011, and a bachelor’s degree in physics and Japanese from Duke University in 2006.