Tolga Tasdizen

Associate Professor, Electrical and Computer Engineering Department
Adjunct Faculty Member, School of Computing
Adjunct Faculty Member, Department of Neurology

Curriculum Vitae
Publications Page


I am an Associate Professor in the Electrical and Computer Engineering Department at the University of Utah. I am also a USTAR faculty member at the Scientific Computing and Imaging Institute and a member of the Utah Center for Neuroimage Analysis and the Center for Advanced Microscopy. My research interests are in the general areas of image processing and pattern recognition. More specifically we study image neighborhoods and context for image registration and segmentation; and variational methods and level-sets for volume restoration, reconstruction and segmentation. Over the last five years, the main driving application of my research has been neural circuit reconstruction (connectomics) from large-scale electron and confocal microscopy image datasets. This application poses challenges in image registration, pattern recognition and 3D segmentation, which are very well aligned with my research interests.

Image registration: My research group has developed algorithms for assembling 3D volumes from hundreds of thousands of 2D images of serial sections. In electron microscopy, our approach uses the well-known shift property of the Fourier transform for a computationally feasible solution to correcting distortions between sections due to the cutting process as well as the imaging optics. For confocal microscopy of neurons, we have recently developed a novel algorithm that combines axon tracing and a landmark based alignment approach that is well suited to the sparse nature of these images. These tools are combined in our NCR Toolset, which is freely available along with an easy to use graphical interface. Our main related publications in this area are:

T. Tasdizen, P. Koshevoy, B. C. Grimm, J. R. Anderson, B. W. Jones, C. B. Watt, R. T. Whitaker and R. E. Marc, "Automatic mosaicking and volume assembly for high-throughput serial-section transmission electron microscopy," Journal of Neuroscience Methods, 193(1), 132-44, 2010.

L. Hogrebe, A. Paiva, E. Jurrus, C. Christensen, M. Bridge, J. Korenberg and T. Tasdizen, ”Trace Driven Registration Of Neuron Confocal Microscopy Stacks,” IEEE International Symposium on Biomedical Imaging (ISBI): From Nano to Macro, 2011.

L. Hogrebe, A. R.C. Paiva, E. Jurrus, C. Christensen, M. Bridge, J.R. Korenberg, P. R. Hof, B. Roysam, T. Tasdizen, Serial Section Registration of Axonal Confocal Microscopy Datasets for Long Range Neural Circuit Reconstruction, Journal of Neuroscience Methods 207, pp. 200-210, 2012.

M. L. Berlanga, S. Phan, E. A. Bushong, S. Lamont, S. Wu, O. Kwon, B. S. Phung, M. Terada, T. Tasdizen, E. Martone and M. H. Ellisman, ”Three-dimensional reconstruction of serial mouse brain sections using high-resolution large-scale mosaics,” Frontiers in Neuroscience Methods, March 2011.

J.R. Anderson, B.W. Jones, J-H Yang, M.V. Shaw, C.B. Watt, P. Koshevoy, J. Spaltenstein, E. Jurrus, Kannan U V, R. Whitaker, D. Mastronarde , T. Tasdizen, R.E. Marc. "A Computational Framework for Ultrastructural Mapping of Neural Circuitry" PLoS Biology, March, vol. 7, no. 3, pp. e74, 2009.

Pattern Recognition: Electron microscopy images of neural tissue are typically hard to segment due to their properties. The staining used in electron microscopy is non-selective, i.e. it highlights all cellular and intracellular membranes present in the tissue sample. Therefore, a successful segmentation of cells first requires the differentiation of cellular membranes from intracellular membranes, which look similar at the local scale but are differentiable using non-local contextual cues. We have developed a series artificial neural network classifier that uses non-local context to provide as good or better accuracy compared to other approaches to the cell membrane detection problem. A significant advantage of our method over previous approaches is the sequential stage design that allows for efficient training. See the following papers and Liz Jurrus' webpage for more details.

E. Jurrus, A. R. C. Paiva, S. Watanabe, J. R. Anderson, B. W. Jones, R. T. Whitaker, E. M. Jorgensen, R. E. Marc and T. Tasdizen, "Detection of neuron membranes in electron microscopy images using a serial neural network architecture," Medical Image Analysis, 14:6, 770-83, 2010

E. Jurrus, S. Watanabe, A. R. C. Paiva, M. Ellisman, E. M. Jorgensen and T. Tasdizen, ”Neuron Reconstruction from Electron Microscopy Images,” Neuroinformatics 2012.

S. M. Seyedhosseini, R. Kumar, E. Jurrus, R. Guily, M. Ellisman, H. Pfister and T. Tasdizen, ”Detection of Neuron Membranes in Electron Microscopy Images using Multi-scale Context and Radon-like Features,” accepted to MICCAI 2011.

T Liu, S. M. Seyedhosseini, E Jurrus, M Ellisman and T Tasdizen, Watershed Merge Tree Classification for Electron Microscopy Image Segmentation, ICPR 2012

Also, for problems ranging from filtering to recognition, the conventional wisdom in image processing is to use a filter bank hand designed to match the problem at hand. However, filters are functions of image neighborhoods and evidence suggests they can be used directly as a feature representation. For instance, we have used image neighborhoods directly to classify brain tissues in magnetic resonance images and to segment general textured images. Unfortunately, a problem with representing context from large areas using image neighborhoods is the curse of dimensionality. We have used principal component analysis of image neighborhoods to develop a novel version of the non-local means image restoration algorithm that outperforms the original algorithm. Similarly, for cell membrane detection, we are currently utilizing a multi-scale image neighborhood approach to develop a method that will surpass the accuracy of state-of-the-art approaches.

T. Tasdizen, ”Principal Neighborhood Dictionaries for Non-local Means Image Denoising,” IEEE Transactions on Image Processing, vol. 18, no. 12., pp. 2649-60, December 2009. Top third accessed paper in IEEE Xplore in November 2009.

S. Gerber, T. Tasdizen and R. T. Whitaker, ”Dimensionality Reduction and Principal Surfaces via Kernel Map Manifolds,” ICCV 2009.

S. Gerber, T. Tasdizen, S. Joshi and R. T. Whitaker, ”On the Manifold Structure of the Space of Brain Images” MICCAI 2009.

S. Gerber, T. Tasdizen, T. Fletcher, S. Joshi and R. T. Whitaker, Manifold Modeling for Brain Population Analysis, Medical Image Analysis, Volume 14, Issue 5, Special Issue on the 12th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), October 2010, Pages 643-653. Best paper of the special issue award.

Suyash P. Awate, T. Tasdizen, Norman L. Foster and Ross T. Whitaker, “Adaptive, Nonparametric Markov Modeling for Unsupervised, MRI Brain-Tissue Classification,” Medical Image Analysis, Vol. 10, Num. 5, Pages 726-739, 2006.

S. P. Awate, T. Tasdizen an R. T. Whitaker, “Unsupervised Texture Segmentation with Nonparametric Neighborhood Statistics,” ECCV 2006.

3D segmentation: Even state-of-the art approaches can’t provide perfect cell membrane detection results. Gaps may exist between adjacent cells or a single cell can have false positive membranes inside, resulting in over-segmentation. Combining the detected membranes with shape-based regularization can further reduce these errors. We developed a novel three-dimensional surface processing method that is based on optimization of energy function defined on the field of surface normal vectors. Our approach couples a partial differential equation (PDE) that minimizes the surface normal energy and another PDE that links the surface to the changes in the surface normal vectors resulting from the first PDE.  We have applied this approach to smoothing and enhancement of geometrical models, and reconstruction of objects from laser range finder data. While our original methods were developed for models represented as level-set functions, other researchers have extended our methods to surface meshes. Some related publications are:

T. Tasdizen and R. T. Whitaker, “Higher-order Nonlinear Priors for Surface Reconstruction,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 26, Num. 7, Pages 878-891, July 2004.

T. Tasdizen, R. T. Whitaker, Paul Burchard and Stanley Osher, “Geometric Surface Processing via Normal Maps,” ACM Transactions on Graphics, Vol. 22, Num. 4, Pages1012-1033, October 2003.


Contact Information

Best way to get in touch with me is via email.
Office  WEB 3887
Phone  (801) 581-3539
Fax      (801) 585-6513
Mail     Scientific Computing and Imaging Institute
            University of Utah
            72 S. Central Campus Drive, 3750 WEB
            Salt Lake City, 84112, USA