Gordon L. Kindlmann

alternate text
Assistant Professor, Department of Computer Science
Assistant Professor, Fellow, Computation Institute
University of Chicago

1100 E. 58th Street
Ryerson Hall 161-B
Chicago, IL 60637

Current CV


I research scientific visualization and image analysis to improve the biomedical applications of three-dimensional imaging modalities (like MRI and CT). My past research simplified the work of making informative direct volume renderings, inspired by traditional techniques of edge detection. I continue to explore ways of translating mathematical principles of image processing and computer vision to practical methods of detecting, measuring, and understanding biological and anatomical structure in modern imaging data.

My doctoral and post-doctoral work focused on diffusion MRI, including data inspection, fiber tractography, feature detection, and tensor analysis. Recent work is focusing on building image analysis tools that work usefully on the variety of imaging modalities studied by colleagues in the Biological Sciences Division (microCT, clinical CT, spectral MRI). I am also involved in the creation of the Diderot language, which will simplify the work of translating analysis and visualization algorithms into high-performance GPU-based computation. All my research software is open-source, which is vital for creating reproducible methods of computational science.



Recent Publications

C. Chiw, G. Kindlmann, J. Reppy, L. Samuels, N Seltzer. (alphabetical order) Diderot: A Parallel DSL for Image Analysis and Visualization. To appear: ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI '12), June 2012.

Introduces Diderot, a domain-specific language specialized for parallel computation on continuous tensor fields (including scalar and vector fields), supporting idiomatic mathematical expressions like ∇ for gradient, · for dot products, ⊗ for tensor (outer) product, and π for pi. Ongoing work is developing the OpenCL (GPU) backend of the compiler.

T. Schultz, G. L. Kindlmann. Superquadric Glyphs for Symmetric Second-Order Tensors. IEEE Trans. on Visualization and Computer Graphics, Nov/Dec 2010, 16(6):1595-1604

Uses symmetry and continuity guidelines to broaden the representational abilities of superquadric tensor glyphs to all symmetric tensors 2nd-order tensors. Concavity on the glyph surface represents a sign change between eigenvalues, and a halo represents tensor trace. Applications include Hessians, stress tensors, and rate-of-deformation tensors. An interactive OpenGL-based implementation is based on a palette of pre-computed base glyphs.

T. Schultz, C-F Westin, G. Kindlmann. Multi-Diffusion-Tensor Fitting via Spherical Deconvolution: A Unifying Framework. Lecture Notes in Computer Science Volume 6361, Proc. MICCAI 2010, pp 674-681.

Multi-tensor models address the limitations of the single diffusionn tensor in situations of partial voluming and fiber crossings. However, selection of a suitable number of fibers and numerical difficulties in model fitting have limited their practical use. This paper addresses both problems by making spherical deconvolution an efficient part of the fitting process.

P. Savadjiev, G. L. Kindlmann, S. Bouix, M. E. Shenton, C-F Westin. Local white matter geometry from diffusion tensor gradients. Neuroimage, Feb 2010, 49(4):3175-86.

Introduces a mathematical framework for computing geometrical properties of white matter fibers directly from diffusion tensor fields, by isolating the portion of the gradient of the tensor field corresponding to local variation in tensor orientation, and projecting it onto a coordinate frame of tensor eigenvectors. Results are shown in a group study on schizophrenia.

Gordon L. Kindlmann, R. San Jose Estepar, Stephen M. Smith, C.-F. Westin, Sampling and Visualizing Creases with Scale-Space Particles. IEEE Trans. on Visualization and Computer Graphics, Nov/Dec 2009, 15(6):1415-1424

Expands the utility of particle systems from sampling pre-segmented closed surfaces, to performing the work of isolating and sampling ridge and valley features, in scale-space, in unsegmented image data. Includes an efficient means of performing scale-space interpolation in high-resolution 3D images.