Diagnostic Imaging
October 2003


Supertensor imaging locates complex fibers

New generation of software gives radiologists more diagnostic options

By: C.P. Kaiser

At the time of its invention in the early 1990s, MR diffusion tensor imaging, which measures water molecule diffusion in brain white matter, solved problems associated with standard MR diffusion imaging. Before long, however, it became apparent that even this cutting-edge technology fell short of accurately portraying the complex network of fiber bundles underlying the brain.

Conventional MR diffusion tensor imaging, which uses six tensors, captures and displays data that indicate only the dominant fiber orientation within the region of interest. Small adjacent fibers and branching fibers may not be distinguished. While information about the dominant fibers is valuable for many clinical indications, the picture it gives of the fiber network is incomplete. Researchers have attempted to solve this problem by developing methods to acquire data from up to 250 tensors. Images from these measurements are extremely detailed, showing many fiber bundles per voxel.

Several different names have been ascribed to this new approach, but it is commonly known as supertensor imaging.

"Measuring in six directions essentially oversimplifies the connectivity of white matter tracts," said Dr. A. Gregory Sorensen, director of the HST Martinos Center at Massachusetts General Hospital. "In any given voxel, there are a number of intersecting tracts. This is where supertensor imaging or diffusion spectrum imaging comes in."

In the pons, for example, one fiber bundle goes up/down and another goes side-to-side. While a voxel will have two different fiber populations, a standard tensor description will miss the second bundle. Supertensor imaging collects much more information.

The wealth of data revealed by supertensor imaging has researchers moving beyond surgical applications to nonsurgical uses. Studies have demonstrated that diseases such as Alzheimer's, schizophrenia, and autism have pattern changes in their fractional isotropy, a measure of water's tendency to move along a fiber's restricted path. In diseased or damaged white matter, the integrity of restriction is compromised, allowing water to diffuse isotropically. Researchers hope to use supertensor imaging to uncover fundamental causes of many brain disorders, as well as to document the pathological progression of disease. Investigators are also attempting to map fiber connectivity of the entire normal brain.


Sorensen and colleagues use supertensor imaging in clinical research to predict stroke recovery, to study new white matter connections following disruption, to chart normal brain development, and to uncover the underlying mechanisms of actions involved in postsurgical changes. A cingulotomy, for example, often improves the health of patients suffering from obsessive-compulsive disorder, but researchers are not sure why this operation helps. Sorensen wants to study this patient population with supertensor imaging. He theorizes that if he can follow the fiber bundles up to where the cingulotomy is done and watch the growth of tissue over time, he might be able to determine which connections actually matter to the treatment of OCD.

Knowing this information might help with drug development or in tailoring treatment more precisely. This approach could feasibly work for any number of brain disorders, he said.

Mass General has developed a visualization task card for standard diffusion tensor imaging. It calculates fractional isotropy maps that can be colorized to show different directional movement. Radiologists simply draw a region of interest, such as within the corpus collosum, and the computer searches for the fiber paths intersecting that ROI. Other planes can be subtracted, leaving only the directionality of the desired tracts. These can further be manipulated in 3D. Many surgeons find this helpful to identify tracts, Sorensen said. Even during surgical procedures, tensor data can indicate whether fiber bundles have moved into a desired position. Sorensen's group expects to have an equally interactive platform for supertensor imaging by year's end.

Standard diffusion tensor imaging is clinically feasible because the data can be acquired in minutes, which is not true for supertensor imaging. Researchers have used-alone or in combination-q-space imaging, high angular resolution diffusion-weighted (HARD) imaging, and diffusion spectrum imaging to reduce acquisition time.


A paper presented at the International Society for Magnetic Resonance in Medicine meeting in July detailed a new algorithm that reduced acquisition to 15 minutes. Its inventor, David S. Tuch, Ph.D., a radiology instructor at Harvard Medical School and researcher at Mass General, dubbed the method q-ball imaging.

Traditional q-space imaging uses a 3D cartesian grid (10 x 10 x 10, or 1000 images), which takes a long time to acquire by MR. Tuch's insight was that he needed to sample only on the shell of the sphere rather than the whole sphere-hence, q-ball imaging. Using this method, Tuch can acquire in 15 minutes clinical images measuring in 126 directions. But even 15 minutes is too long, according to Sorensen. Three-T scanners, multiple channels, parallel imaging, and newer algorithmic frameworks will eventually cut the time in half.

"We routinely collect clinical research data, but it will be another year before we can safely use these tools in routine clinical use," he said.

The Mass General team uses q-ball imaging to research stroke. It helps to identify fiber destruction at the core of an infarct or secondary edema versus actual fiber damage. Fractional anisotropy either increases or decreases in early cerebral ischemia and this might be an imaging biomarker that predicts viability, Sorensen said. For tumors, advanced tensor imaging can determine if the mass simply diverges around white matter or infiltrates it and to what degree.

A diffusion abnormality on an MR scan in a region with different fiber populations does not indicate which fibers are affected. A q-ball scan can identify selective damage to particular fiber populations. The diagnosis can be targeted to particular anatomical pathways and not just to a global location. Since a particular region of the brain can have five functional pathways crossing through it-motor, language, speech, etc.-neuroradiologists will know the precise functional system being compromised.

Researchers at Mass General are hopeful that q-ball imaging will be useful for studying schizophrenic patients. In schizophrenia, fiber connections are disrupted and the thalamus plays an important role. Autopsy and MRI show schizophrenic patients have smaller thalami than normal.

"We don't know which nucleus in the thalamus is the important one in terms of treatment, but we think diffusion spectrum imaging can help us sort it out," Sorensen said.

Under the direction of Lawrence R. Frank, Ph.D., an associate professor of radiology at the University of California, San Diego, researchers have reduced supertensor acquisition time to 15 minutes. Frank uses HARD imaging with a modified reconstruction algorithm called spherical harmonic decomposition. Researchers at UCSD use advanced tensor imaging to study a variety of conditions including muscular dystrophy, AIDS, and alcoholism.

Susan F. Tapert, Ph.D., an assistant professor in residence at UCSD, has added HARD imaging to her armamentarium to study addiction in adolescents. She has found a diminished fractional anisotropy in subjects with drinking problems compared with controls, particularly in the back of the corpus callosum. Previous studies of older adults have found overall shrinkage of the corpus callosum by later stage alcoholism.

"In the young population, it could be the quality, not the size, of the corpus callosum that matters," she said.

Tapert and colleagues have found correlations between the integrity of white matter in several parts of the corpus callosum and the ability of these subjects to perform certain thinking and memory tasks.

"Tensor imaging is a very sensitive way to see what is going on in the brain, particularly in patient groups without profound abnormalities yet," she said. "In these patients, we do not see any major shrinkage of gray or white matter. But we are seeing changes in the white matter integrity."


Radiologists will be forced to use tensor imaging and fiber tracking because the detailed information it provides is particularly useful to surgeons, Dr. Chip Truwit, chief of radiology at the University of Minnesota, said at the American Society of Neuroradiology annual meeting in April.

"We shouldn't expect tensor imaging to solve every problem. It can sort out major pathways, but it can't sort out the hand from the foot, at least not today," he said.

Truwit acknowledged that many tensor imaging processes are becoming user-friendly, one of the first steps toward routine clinical use. But for now, researchers across the country and in Europe are developing independent strategies to diminish the acquisition time associated with supertensor imaging.

"We know that diffusion tensor imaging throws away a lot of information associated with the other crisscrossing fibers," said Daniel Alexander, Ph.D., a lecturer in computer science at University College of London. "We also know that we can use higher powered techniques to get at that information."

Alexander and colleagues developed a method called persistent angular structure imaging, which allows them to measure 50 directions in 20 minutes. The trade-off with the shortened acquisition time is a substantial increase in postprocessing.

"There doesn't seem to be one single technique that ameliorates the problems of diffusion tensor imaging without introducing another set of problems," Alexander said.