NIH Researcher Creates Anatomic Visualization for Noninvasive 3D Bronchial Fly-Throughs
Ronald M. Summers, National Institutes of Health, Bethesda, Maryland

Aided by powerful and flexible computer visualization software, a diagnostic radiologist at the National Institutes of Health, Bethesda, Maryland, is pioneering completely non-invasive human bronchial tree inspection that requires only a few minutes of patient involvement.

NIH Image 1 With a technique he calls virtual bronchoscopy, Ronald M. Summers, M.D., Ph.D., in the NIH Diagnostic Radiology Department, uses computerized tomography (CT) and computer software to create extremely accurate 3D images of patients' bronchial structures, then analyzes them with a fly-through technique.

With degrees in physics and medicine from the University of Pennsylvania, Dr. Summers completed a radiology residency program at the University of Michigan, then went to Duke University where he began to study computerized 3D imaging and visualization techniques.

"During that time," he says, "I was using software to fly through 3D models of different structures of the body. When I came to NIH as a diagnostic radiologist, I began to use IRIS Explorer and a Silicon Graphics computer. Then I started a program to make computerized models."

Dr. Summers' workstation is an SGI Indigo 2 Extreme with 320 megabytes of RAM and a nine gigabyte hard drive running IRIS Explorer(TM) 3.0, the interactive 3D data visualization system from the Numerical Algorithms Group (NAG), Inc., Downers Grove, Illinois.

CT and MRI scanners in his department are routinely used for diagnostic studies. At the present time, for example, Dr. Summers and an associate are studying patients with homozygous familial hypercholesterolemia, using MRI to research the development of atherosclerosis in those patients. Another study involves juvenile dermatomyositis, an inflammatory condition of the muscles, and MRI is assisting in the evaluation of abnormalities in the muscles of affected children.

As the primary collectors of data for computerized visualization, their MRI and CT scanners are linked to the Internet where the output is transmitted to Dr. Summers' workstation. The datasets containing scans range from 70 to 120 megabytes in size.

"I transfer them over our network to the SGI computer," Dr. Summers explains. "Then I strip the header information and generate a 3D uniform lattice to use as IRIS Explorer input. From there, I use a module I wrote for IE that does a region-growing algorithm, a technique in image processing. In brief, I say that a voxel (volume element) in the dataset is my seed, and then I let the region-growing algorithm aggregate adjacent data elements (voxels) in the image."

NIH Image 2 "In my method," Dr. Summers says, "I use a CT dataset of a patient's chest -- perhaps 70 megabytes of data, and it includes multiple cross sections through the imaging area stacked vertically. Each data element in the set consists of the density of the tissue in one very tiny area of the patient. Thus, the 70 megabyte set specifies and includes the density of the tissue at millions of voxels in that patient."

"Then I pick one voxel inside the airway, and I ask the software to aggregate all the voxels that connect to the original one that also are filled with air. That algorithm spreads out through the patient looking for little bits of air in the lungs. When it's done, it has the bronchial tree."

At this point, Dr. Summers sends the data through IRIS Explorer's IsoSurfaceLat module. An isosurface is a surface of constant value in a 3D field. IsoSurfaceLat calculates an isosurface of a scalar 3D lattice. The algorithm used takes edge intersections and knits them into triangle meshes -- usually between 100,000 and 250,000 triangles -- for rendering. This procedure generates a 3D visualization of the surface of the bronchial wall which Dr. Summers then sends to IE's Render module.

The Render module displays geometry data. It is built using the Inventor SceneViewer which allows a large number of different viewing paradigms. There are two major modes in Render -- viewing and selection/picking. In viewing mode, the camera parameters can be changed, and users can toggle between a perspective and an orthographic view. Other features include a Walk viewer for a walk-through allowing motion and a camera pointed with constant eye level, a Plane viewer that lets the user manipulate the camera with respect to the viewing plane, and a Fly viewer that simulates constrained flight through space.

"With this 3D model and the Fly viewer," he continues, "I can move around and look for lesions of the bronchial wall. I also look for areas of narrowing, and I can make measurements of bronchial size. This computerized method allows us to make exact lesion measurements. The measurements cannot be made accurately with the bronchoscope used in standard procedures because of its wide-angle lens."

This procedure has many other advantages," Dr. Summers emphasizes. "We're simply doing a CT scan; no sedation is necessary. The patient just holds his or her breath for 15-20 seconds twice."

Dr. Summers uses a number of IRIS Explorer modules in his work. One is OrthoSlice, a module that extracts a 2D lattice from a 3D lattice in index space. Others include AnimateCamera and WriteMovie, both of which he connects to Render to make movies of fly-throughs.

"I have video capability through my Indigo2 video collaboration option, and a VCR connected to my workstation. As I'm flying around, I can record it on standard VHS videotape. Quite often, I make slides with the DisplayImage module and the WriteImage module using SGI format, then import them into the SGI Showcase presentation package. Then I can generate 35mm slides for presentations. I've also done fly-throughs of the aorta in the patients with homozygous familial hypercholesterolemia, and I've visualized a beating human heart."


No endorsement of any organization, product or service mentioned in this article is intended by the National Institutes of Health or its employees, or should be inferred.
Last modified: Thu Jan 16 14:17:27 1997
[ Posters Home ]
© The Numerical Algorithms Group Ltd, Oxford UK, 1996.