Circular Spectrum Mapping for Intravoxel Fiber
Structures Based on High Angular Resolution Apparent
Diffusion Coefficients
Wang Zhan,
1
Hong Gu,
1
Su Xu,
2
David A. Silbersweig,
1
Emily Stern,
1
and
Yihong Yang
1
*
A method is presented for mapping intravoxel fiber structures
using spectral decomposition onto a circular distribution of
measured apparent diffusion coefficients (ADCs). The zeroth-,
second-, and fourth-order harmonic components of the ADC
distribution on the circle spanned by the major and median
eigenvectors of the diffusion tensor can be used to provide
quantitative indices for isotropic, linear, and fiber-crossing dif-
fusion, respectively. A diffusion-weighted MRI technique with
90 encoding orientations was implemented to estimate the cir-
cular ADC distribution and calculate the circular spectrum. A
digital phantom was used to simulate various diffusion pat-
terns. Comparisons were made between the circular spectrum
and regular DTI-based index maps. The results indicated that
the zeroth- and second-order circular spectrum maps exhibited
a strong consistency with the DTI-based mean diffusivity and
linear indices, respectively, and the fourth-order circular spec-
trum map was able to identify the fiber crossings. MRI experi-
ments were performed on seven healthy human brains using a
3T scanner. The in vivo fourth-order maps showed significantly
higher densities in several brain regions, including the corpus
callosum, cingulum bundle, superior longitudinal fasciculus,
corticospinal tract, and middle cerebellar peduncle, which in-
dicated the existence of fiber crossings in these regions.
Magn Reson Med 49:1077–1088, 2003. © 2003 Wiley-Liss, Inc.
Key words: apparent diffusion coefficient; diffusion tensor im-
aging; fiber crossing; high angular resolution; spectral decom-
position
Diffusion-weighted MRI has been widely used to noninva-
sively investigate the structural organization of biological
tissues. Much attention has been paid to diffusion anisot-
ropy, in which a scalar apparent diffusion coefficient
(ADC) is insufficient to fully describe the diffusion process
(1– 4). The diffusion tensor imaging (DTI) method intro-
duced by Basser et al. (5,6) overcame this difficulty to a
large extent by interpreting the anisotropic diffusion as a
second-order function of the encoding orientation. Much
effort has been directed toward optimizing the DTI acqui-
sition scheme (7–9), and significant progress has been
made in the interpretation of DTI data by various quanti-
tative indices based on the eigenstructure of the diffusion
tensor (10 –12). As a promising application of DTI, “fiber
tracking” techniques have been developed to allow a non-
invasive delineation of the neuronal pathways inside the
brain (13–16). However, a major challenge for DTI tech-
niques is the problem caused by fiber crossings, including
fiber “intersection,” “dispersion,” “kissing,” and “branch-
ing,” whereby multiple fiber compartments share a single
voxel. In this case, the major eigenvector of the tensor can
be substantially biased from the real fiber orientation, pre-
venting fiber-tracking algorithms from following the actual
fiber connection. Moreover, reducing the voxel size does
not remedy this problem completely because of the “pow-
der averaging” effects of multiple-fiber singularity (15).
Some tensor-derived indices, e.g., the fractional anisot-
ropy (FA), are vulnerable to the artifacts associated with
the partial volume effects of crossing fibers (17). To resolve
these problems, Poupon et al. (14,18) proposed an im-
proved fiber-tracking algorithm to estimate a regularized
fascicle trajectory by taking into account prior knowledge
of a local direction map. Wiegell et al. (19) investigated the
fiber crossings in several regions of interest (ROIs) in the
human brain by employing the full eigenstructure of the
diffusion tensor, and found that the planar model spanned
by the major and median eigenvectors of the tensor could
be used to characterize the intersection and dispersion of
fibers. Some geometric measurement (spherical case (CS),
(linear case (CL), and (planar case (CP)) indices, describing
the tensor shape in regard to spherical, linear, and planar
degrees, respectively, were proposed by Westin et al. (20)
and investigated extensively by Alexander et al. (21).
However, the efforts made within the DTI formulation
cannot offer a complete solution to the fiber-crossing prob-
lem, mainly because of the limitation of the tensor model
itself. The CP index, for example, is unable to distinguish
an inherent planar diffusion tensor (e.g., in a thin mem-
brane filled with water) from the planar tensor induced by
an orthogonal intersection of two linear fibers (17). More
importantly, it should be noted that the diffusion tensor is
only a second-order approximation (in terms of mean
square fitting) to the real 3D diffusion process (22), and
thus this model cannot make full use of this intravoxel
non-Gaussian information. Therefore, a model beyond
conventional DTI is essential to solve the problem of fiber
crossing. In fact, a number of studies beyond the DTI
framework have been carried out. An important contribu-
tion came from q-space imaging, a technique in which the
tissue structures are probed by directly calculating the
diffusion displacement distribution in each voxel (23–25).
This method is theoretically rooted in the Fourier-trans-
form relationship between the echo attenuation in q-space
and the diffusion displacement probability distribution.
1
Functional Neuroimaging Laboratory, Department of Psychiatry, Weill Med-
ical College of Cornell University, New York, New York.
2
Department of Medical Physics, Memorial Sloan-Kettering Cancer Center,
New York, New York.
*Correspondence to: Yihong Yang, Ph.D., MRI Physics Unit, Neuroimaging
Research Branch, National Institute on Drug Abuse, NIH, 5500 Nathan Shock
Drive, Rm. 383, Baltimore, MD 21224. E-mail: YiHongYang@intra.nida.nih.gov
Received 12 June 2002; revised 13 January 2003; accepted 15 January 2003.
DOI 10.1002/mrm.10484
Published online in Wiley InterScience (www.interscience.wiley.com).
Magnetic Resonance in Medicine 49:1077–1088 (2003)
© 2003 Wiley-Liss, Inc. 1077