Visitor Presentations

This page contains information about presentations made by visiting faculty. Some of these presentations are available as PPT/PDF upon request. Contact here.

Maximum-likelihood Segmentation of Ultrasound Images by Tunneling Descent

Hemant D. Tagare
Dept. of Diagnostic Radiology
Dept. of Biomedical Engineering, Yale University
Talk given at NLM on July 25, 2005.


Ultrasound images are notoriously difficult to segment with classical segmentation algorithms like active contours. The problem is that the energy function of active contours exhibits multiple local minima in ultrasound images. When the active contour evolves under gradient descent it gets trapped in these local minima, giving wrong segmentations. In this talk I will propose an alternate evolution strategy for active contours called tunneling descent. Tunneling descent is a deterministic evolution strategy that "tunnels out of" spurious local minima. It is designed to be a replacement for gradient descent. When used in a maximum-likelihood formulation, tunnel descent can successfully find the endocardium in short axis ultrasound images without manual parameter tweaking. Using tunneling descent, we have segmented over 100 short axis images with literally the same algorithm. I will present validation results for the segmentation as well.

Medical Image Databases: Retrieval by Content

Hemant D. Tagare
Dept. of Diagnostic Radiology
Dept. of Electrical Engineering, Yale University
Talk given at NLM on April 26, 2002.


One of the challenges we face as medical imaging goes digital is that of archiving and retrieving digital images. Standard text-based databases fall short because they are unable to deal with the richness of image content (images are hard to describe by text). In this talk I will discuss some of the challenges in designing a truely image-aware database in which images are retrieved by their content. During the first half of the talk I will review the requirements of medical and biological image databases. During the second half, I will discuss some the open technical problems in the indexing of medical image databases.

An Introduction to Shape Spaces and Shape Statistics

by Hemant D. Tagare
Dept. of Diagnostic Radiology
Dept. of Electrical Engineering, Yale University
Talk given at NLM on April 25, 2002.


Shape spaces are spaces of shapes of points. Shape statistics is concerned with developing statistics for them. These two problems - that of defining shape, and that of measuring it - are classic in image processing. Shape spaces are a natural mathematical setting for these problems. Unfortunately much of the theory of shape spaces is not easily accessible to engineers and image processors. To overcome this, I will introduce a simple shape space which can be visualized in 3-d. And with the help of it, I will introduce the basic concepts of Shape Spaces and Statistics. I will also refer to some applications of Shape spaces in medical imaging. This is a very informal and tutorial talk. It does not use anything more complicated than vector analysis in 3-d.

Active Contours and Deformable Templates: A Quick Tour

by Dr. Hemant D. Tagare, Associate Professor, Dept. of Radiology and
Electrical Engineering, Yale University
Talk given at NLM on September 21, 2000.


This is an informal tutorial on medical image segmentation for non-specialists. I will review two common techniques used for segmenting medical images: active contours and deformable templates. Both of these try to evolve a curve or a surface from an initial guess to nearby edges (or ridges) in the image. The evolution is usually guided by image and shape dependent forces, which are called external and internal forces.

I will introduce the following basic ideas in active contours and deformable templates:

  • Design of external forces:
    • Edge-seeking and Ridge-seeking active contours. Gradient Estimation. Maximum-likelihood type active contours.
  • Design of internal forces:
    • Smoothness and Shape dependent forces.
  • Bias and stability of active contours.
  • Numerical algorithms for evolution of active contours:
    • B-spline active contours
    • Dynamic Programming
    • Level set/Front Propagation

I will present examples from my work and the literature.

Cervical and Lumbar Spine Image Analysis

by Dr. R. Joe Stanley, Assistant Professor, Dept. of Electrical and Computer Engineering, University of Missouri-Rolla
Talk given at NLM on October 6, 2000.


This research investigates image processing and pattern recognition techniques used for anterior osteophyte classification in cervical spine x-ray images. The specific objectives for this research are to:

  • investigate histogram-based cervical spine image enhancement algorithms
  • evaluate several edge operators for highlighting vertebra boundaries
  • examine a B-spline algorithm for approximating vertebra boundaries from manually chosen point sets
  • explore radius of curvature-based features for anterior osteophyte classification
  • examine four different pattern recognition algorithms for anterior osteophyte classification using the radius of curvature-based features
  • look at the applicability of the cervical spine image enhancement and vertebra boundary determination techniques to lumbar spine images
  • propose future directions for cervical spine image analysis, and
  • propose image analysis techniques for lumbar spine images.

Customized Active Shape Models for Segmentation of Cervical and Lumbar Spine Vertebrae

by Dr. Hamed Sari-Sarraf, Assistant Professor, Dr. Sunanda Mitra, Professor, Gilberto Zamora, and Abraham Tezmol, all of the Dept. of Electrical Engineering, Texas Tech University

This report details work done by Texas Tech researchers in collaboration with CEB for the segmentation of cervical and lumbar spine x-ray images.

To download the report, go to and take links Research Projects and Customized Active Shape Model for Segmentation of Cervical and Lumbar Spine Vertebrae.