Paul Thompson's Research Publications

Tensor Visualization of Brain Variability

SIGGRAPH Computer Graphics Conference, 2001, Los Angeles, CA

John Bacheller, Paul M. Thompson, Colin J. Holmes, Arthur W. Toga

1Laboratory of Neuro Imaging, Department of Neurology, Division of Brain Mapping, UCLA School of Medicine, Los Angeles, California 90095


Description. No two brains are exactly alike, and brain structure varies dramatically even among normal individuals. To visualize such variability, our brain mapping research group at the UCLA Laboratory of Neuro Imaging developed novel supercomputing algorithms to compute and visualize how brain structure varies across normal individuals, based on a database of MRI scans from 20 subjects. In these images, the ellipsoids represent the amount of anatomical variation across subjects. Pink colors show greatest anatomical variation in the most uniquely human brain areas, controlling language and logical reasoning. The blue spheres in the middle represent brain areas with little interpersonal variability; these control sensation and motor function. Ellipsoids are also most elongated along directions in which variability is greatest. The bubble-like rendering plots a probability field, indicating through clustering and varying levels of transparency how likely it is to find a brain structure in each coordinate location in a normal population.

Novelty. These tensor visualizations are novel, as they draw upon high-resolution brain imaging, as well as tensor concepts from Gaussian field theory and fluid dynamics, to help determine when a brain region in a new subject is abnormally situated, or whether such a variation is normal. This helps in the early diagnosis of Alzheimer's Disease and schizophrenia, and in the evaluation of new treatment approaches. The application area is novel, as it is the first time brain variations in an image database have been encoded for disease detection in new subjects.

Computation. Novel mathematics and advanced rendering power are both indispensable in visualizing such complex tensor datasets. In addition to the medical applications, these tensor images are computed from mathematical mappings with 0.9 trillion parameters and rendered on a Silicon Graphics RealityMonster Supercomputer with thirty two MIPS R10000 CPUs, requiring 9 hours parallel CPU time. The navigation and visualization of these high-dimensional datasets represents a key challenge in biomedical computing today, and the resulting visualizations are empowering new diagnostic applications in neuroscience and medicine.

Related Publications

  • More about Tensors


  • Disease-Specific Brain Atlases

  • other research areas

  • (back to main list)

    Contact Information

  • Mail:

    Paul Thompson, Ph.D.
    Assistant Professor of Neurology
    Laboratory of Neuro Imaging and Division of Brain Mapping
    710 Westwood Plaza, UCLA School of Medicine
    Westwood, Los Angeles CA 90095-1761, USA.

  • E-mail:
  • Tel: (310)206-2101
  • Fax: (310)206-5518