Image Processing and Analysis

Our core work uses organ and tissue-level diagnostic medical images in segmenting and measuring structure and function, image registration, and tracking and quantifying motion and deformation. Continuing to explore how our approaches can be integrated or economized is always at the forefront of our thinking. For example, how can image intensity and image-derived feature information be combined to develop more robust segmentation and registration algorithms or how can parameters found at a higher level of abstraction (e.g., strain in an infarcted region of the left ventricle) be used to guide the extraction of useful low level image features?

However, we also expect to be drawn in new and exciting directions based on our exposure to emerging collaborations with our colleagues in MR spectroscopy and physics, different clinical areas related to image-guided intervention, and structural and functional imaging at the cellular and molecular level. While we will be able to dovetail some of these efforts with methodological approaches we are already developing, we expect to be drawn to entirely new directions. These include tracking multiple nonrigid moving objects with complex evolving relationships (e.g., cell body motion and tubule growth), estimating statistical mixtures of biochemical information represented at each voxel in a variety of image datasets attempting to probe biologically meaningful information (e.g., MR spectroscopy, molecular imaging using fluorescent and/or radiolabeled probes), and designing close-to-real-time updating strategies regarding tissue and tool movement in interventional procedures.

Dr. James Duncan’s research efforts have focused on computer vision, image processing, and medical imaging, with an emphasis on biomedical image analysis. These efforts have included the segmentation of deformable structures from 3D image data, the tracking of non-rigid motion/deformation from spatiotemporal images, and the development of strategies for image-guided intervention/surgery. Frequently, the analysis strategies employ models based on geometrical and physical/biomechanical information to constrain the range of possible solutions in the presence of uncertainty. The laboratory has concentrated on several problem areas within neuroimaging-based structure/function analysis, cardiac function analysis, and radiotherapy-based cancer treatment. Newer work includes cellular and molecular image analysis from microscopy images. The research endeavors aim to find unifying mathematical and algorithmic principles to address this range of problems. Examples of research performed in the laboratory include:

• Using a level set-based strategy to segment neuroanatomical structures.
• Employing biomechanical models and Bayesian estimation to predict local left ventricular deformation.
• Applying spatial activation constraints to estimate fMRI activations.

Dr. Xenophon Papademetris’s research is concentrated on three main areas: (i) brain deformation modeling and compensation during neurosurgical interventions, along with complex 3D visualization of multimodal image data for surgical planning and guidance; (ii) Medical Image Analysis software development, primarily through the BioImage Suite project (see https://bioimagesuiteweb.github.io/webapp/); and (iii) vascular image analysis, focusing on monitoring angiogenesis and the quantification of the growth of tissue-engineered vessels.

Dr. Lawrence Staib’s work concerns the development and application of algorithms for the analysis of biomedical images for quantification of structure and function. Structural image analysis methods of interest include statistical and geometric deformable models for segmentation and model-based shape measurement and comparison. Diffusion weighted magnetic resonance image analysis also provides structural information, and important problems here include white matter parcellation, quantification, and fiber tracking. Groups of subjects can be characterized both structurally and functionally through statistical characterization and classification using spatial patterns of structural and functional parameters.

Dr. Hemant Tagare works on theoretical and practical problems in biomedical image analysis. His research focuses on image segmentation, non-rigid registration, 3D reconstruction, and shape theory. In segmentation, his work involves cardiac ultrasound and brain MRI segmentation, especially the development of new methodologies that utilize machine learning and advanced numerical optimization techniques. In non-rigid registration, he is developing an axiomatic framework for registration, which is applied to brain deformation and atlas building. His 3D reconstruction research concentrates on reconstructing protein structures from Cryogenic Electron Microscopy. In shape theory, his work addresses questions regarding the topology and geometry of affine shape spaces and non-rigid shape comparison.

This package is a combination of two other packages--ADOLC and Revolve--written by Andreas Griewank. ADOLC and Revolve are separate libraries and in this package I combined them together, since both are necessary for IRAD (above). To do this I had to add a few new classes and make some modifications to ADOLC. Changes that I've made to ADOLC are documented in the file CHANGES located in the top level directory.

Download: adcheck-0.99.tar.gz

This is a software library written in C++ for multiquadric radial basis function interpolation (3D only) using the fast multipole algorithm. It was originally written as a part of IRAD but I later rewrote it to be a separate, independent library. It is essentially an implementation of the following two papers:

• Beatson, R. K., Cherrie, J. B., and Mouat, C. T. (1999).
  Fast fitting of radial basis functions: Methods based on preconditioned GMRES iteration.
  Advances in Computational Mathematics, 11:253--270.
• Cherrie, J. B., Beatson, R. K., and Newsam, G. N. (2002).
  Fast evaluation of radial basis functions: Methods for generalized multiquadrics in Rⁿ.
  SIAM Journal on Scientific Computing, 23(5):1549--1571.

Download: fmi-0.99.1.tar.gz

BioImage Suite Web is an open-source web-based medical image analysis set of tools. It runs in any modern web browser (Chrome preferred) and can be used to perform a variety of tasks, including image annotation, registration and preprocessing. For more information, see the main website and the online manual.