Medical Image Processing Laboratory
Research

[NOTE: THIS PAGE IS STILL UNDER CONSTRUCTION PLEASE VISIT US AGAIN TO LEARN MORE ABOUT US AS WE COMPLETE IT]

Overview

The Medical Image Processing Laboratory pursues research on the improvement of medical diagnosis and therapy via imaging. The laboratory is internationally recognized for its work in the area of registration, image-guided surgery, image segmentation, and image acquisition. In the area of registration, the laboratory is pursuing several lines of research. The first is the development of devices and techniques that permit the use of pre-operative image information during surgery. Patented, FDA-approved technology developed in cooperation with surgeons in the Vanderbilt Medical School permits the sub-millimetric registration (spatial realignment) of pre-operative images with patient anatomy during surgery. This process provides the surgeon with critical guidance information that reduces risk and permits better targeting of therapy. Registration methods are also being developed for fusing information acquired with various imaging modalities such as computed tomography (CT), magnetic resonance imaging (MR), positron emission tomography (PET), or ultrasound. Algorithms that permit the three-dimensional warping of one brain onto another are used to create anatomic and functional atlases. Such atlases permit the comparison of populations and the discovery of differences, both anatomic and functional, between normal and diseased subjects. Segmentation techniques are being developed to automate the analysis of medical images. Areas of applications include volumetric and shape measurements of brain structures and substructures for diagnosis purposes or for therapy assessment. Segmentation methods are also being developed for the automatic labeling of radiation sensitive structures. The methodology being developed will make it possible to increase the amount of radiation delivered to cancerous tissues while decreasing radiation delivery to non-target tissues. In the area of image acquisition, the laboratory has developed methods for the correction of geometric distortions in MR images and is developing the methodology required to correct artifacts related to patient motion during the scanning procedure. Currently active research projects include

· Non-rigid registration

· Retrospective evaluation of rigid-body registration methods

· Automatic segmentation and analysis of PET images

· 2D and 3D deformable models for image segmentation

· Development of MRI simulators

Non-rigid registration

Non-rigid registration involves the computation of transformations that permit the elastic deformation or warping of image volumes. The first application of this type of algorithms is atlas-based segmentation and labeling. In this approach one reference image volume (the atlas) is segmented and regions of interest are labeled once and for all. The atlas is then warped onto a new patient's scan. This permits the automatic transfer of the labels from the atlas to the new volume. The second application is the creation of statistical atlases. Information ranging from structure shape to function is extracted from individual image volumes and captured in the atlas. As the number of studies increases, this process permits the creation of statistical information and the comparison of populations. The Medical Image Processing laboratory is conducting research in both these areas. The figure below illustrates results that have been obtained with such a warping algorithm. The image on the left is one slice in a 3D MR image volume of one subject. The image on the right is the corresponding slice in another subject. The image in the middle shows the results obtained when the left volume is warped to make it similar to the right one.

To see a movie showing how the left volume is slowly deformed, play the following mpeg file.

[Home] [People] [Research] [Publications] [News & Links]

The Retrospective Registration Evaluation Project

This NIH-supported project is designed to compare retrospective CT-MR and PET-MR registration techniques used by a number of groups. It involves the use of an FTP database to allow the downloading of image volumes on which the registrations are to be performed. The idea is that the collaborating groups perform registrations on the image volumes, using their own retrospective techniques, and we at Vanderbilt evaluate the accuracy of these transformations by means of our own prospective, marker-based technique. This section is still under construction but to know more, follow this link.

[Home] [People] [Research] [Publications] [News & Links]

Occupancy of Extrastriatal D2 receptors by Clozapine

The responsibility of the laboratory in this NIH-sponsored project is to develop rigid and non-rigid registration methods for the spatial alignment of serial PET (Positron Emission Tomography) images and MR images and for the automatic segmentation and labeling of these PET images. The figure below shows one set of MR (top) and PET (bottom) registered brain images with segmented structures highlighted in red.

[Home] [People] [Research] [Publications] [News & Links]

2D and 3D Segmentation with Deformable Models

The laboratory is conducting research to develop automatic and robust methods for the 2D and 3D segmentation of medical images. A current focus of research is the use of geometric deformable models for the 3D segmentation of the liver in CT images. Liver surfaces created with these techniques are then used for surgical guidance through surface-based registration of pre-operative images with surface points acquired intra-operatively. The figure below shows a few segmentation examples. Each pair of images shows an initial contour (top images) and the final segmentation results obtained after deformation (bottom images)

To see short movies that show the deformation of the contours for each of these examples, click on one of the following links: movie1, movie2, movie3, movie4, movie5. A movie that shows a 3D liver segmentation example can be seen at movie6.

[Home] [People] [Research] [Publications] [News & Links]

MRI Simulator

Comparisons of MRI acquisition protocols and registration/segmentation algorithms require gold standard image sets that reflect a wide variety of protocols and must include correctly segmented anatomy. Segmentation of clinical images requires many hours of expert time, is rarely precise, and is infeasible for large image sets. Simulated MR images provide a partial solution to these problems. Our group has constructed an MRI simulator that provides realistic images for arbitrary pulse sequences executed in the presence of static field inhomogeneities including those due to magnetic susceptibility, errors in the applied field, and chemical shift. The system provides object-specific inhomogeneity patterns from first principles and propagates the consequent frequency offsets and intravoxel dephasing through the acquisition protocols to produce images with realistic artifacts using the 3D digital brain phantom introduced by the McConnell Brain Imaging Centre. The two figures below represent output of our simulator. The first represents an image produced with no distortion and second is an image with an artifical, but easily visualized static-field inhomogeneity pattern.

[Home] [People] [Research] [Publications] [News & Links]