Home > Seminars > Walter J. Scheirer - Scalable Strategies for Image Analysis in Neuroscience

Walter J. Scheirer - Scalable Strategies for Image Analysis in Neuroscience


9/7/2017 at 3:30PM


9/7/2017 at 4:45PM


140 DeBartolo


College of Engineering close button

Patrick Flynn

Patrick Flynn

VIEW FULL PROFILE Email: flynn@nd.edu
Phone: 574-631-8803
Website: http://www.nd.edu/~flynn
Office: 384A Fitzpatrick Hall
Curriculum Vitae


College of Engineering Duda Family Professor of Engineering
Computer Vision Biometrics Pattern Recognition Computer Graphics and Scientific Visualization Mobile Application Development
Click for more information about Patrick
Add to calendar:
iCal vCal
Mapping the synaptic connectivity of neurons in the brain provides diagrams that frame the structural and computational constraints of neuronal circuits. In combination with physiology, these circuit maps unravel the underlying mechanisms of neuronal computations, and hold much promise for the field artificial intelligence, where new classes of algorithms that mimic the sensory processing and reasoning abilities of biological systems may be discovered. At present, only electron microscopy provides sufficient resolution to visualize the detailed intricacy of neuronal circuits. The current strategy is to generate a large set of serial sections containing all circuit elements and reconstruct the circuit by comprehensive automatic segmentation – a process known as dense reconstruction. Imaging large data sets, however, is presently very time consuming and reconstruction remains prohibitively slow and expensive. As an alternative, this talk introduces an Assisted Reconstruction Technique for Electron Microscopic Interrogation of Structure (ARTEMIS) that circumvents these limitations.
By enhancing the signal of genetically encoded markers expressed in defined circuit elements and quickly mapping them in large volumes of brain tissue, it enables sparse reconstructions of genetically defined circuit motifs. These motifs can be further used as road maps for targeted imaging of tissue subsets at high resolution, thus restricting imaging and segmentation time by enabling directed unsupervised reconstructions. As a proof of principle, this technique has been used to automatically reconstruct and visualize interneurons of different mouse retinas. This approach is not restricted to the retina and can be used to track long-range projections anywhere in the brain, e.g., from the retina to visual recipient layers located several millimeters away. Further, the same approach for unsupervised image processing can be applied to a variant of spectral confocal reflectance microscopy, facilitating the long range tracing of myelinated axons, as well as the automatic assessment of myelin thickness. 
Finally, directions in open set recognition for machine learning targeting imaging problems in neuroscience will be discussed. Unknown structures will appear in all imaging modalities as volumes of brain tissue grow in size. How do we identify unknown data and incorporate it into a reconstruction model? A new class of supervised learning methods that can minimize the risk of the unknown and incrementally learn from new data addresses this. A key enabling algorithmic component is the use of the statistical Extreme Value Theory, which leads to accurate probabilistic estimation in various decision-making problems in computer vision.

Seminar Speaker:

Walter J. Scheirer

Walter J. Scheirer

University of Notre Dame


Walter J. Scheirer, Ph.D. is an Assistant Professor in the Department of Computer Science and Engineering at the University of Notre Dame. Previously, he was a postdoctoral fellow at Harvard University, with affiliations in the School of Engineering and Applied Sciences, Dept. of Molecular and Cellular Biology and Center for Brain Science, and the director of research & development at Securics, Inc., an early stage company producing innovative computer vision-based solutions. He received his Ph.D. from the University of Colorado and his M.S. and B.A. degrees from Lehigh University. Dr. Scheirer has extensive experience in the areas of computer vision, machine learning and image processing. His overarching research interest is the fundamental problem of recognition, including the representations and algorithms supporting solutions to it.