Advances on Medical Imaging and Computing.
ABSTRACT In this article, we present some advances on medical imaging and computing at the National Laboratory of Pattern Recognition
(NLPR) in the Chinese Academy of Sciences. The first part is computational neuroanatomy. Several novel methods on segmentations
of brain tissue and anatomical substructures, brain image registration, and shape analysis are presented. The second part
consists of brain connectivity, which includes anatomical connectivity based on diffusion tensor imaging (DTI), functional
and effective connectivity with functional magnetic resonance imaging (fMRI). It focuses on abnormal patterns of brain connectivity
of patients with various brain disorders compared with matched normal controls. Finally, some prospects and future research
directions in this field are also given.
- SourceAvailable from: Tianzi Jiang[Show abstract] [Hide abstract]
ABSTRACT: So far, resting state functional connectivity (RSFC) has been per- formed mainly by seed correlation analysis (SCA) on functional MRI (fMRI) studies. In previous studies, the seeds are usually selected on the basis of prior anatomical information or previously performed activation maps. In this paper, we proposed a novel way to select the desired seeds by taking the natures of resting state data into account. The proposed approach is based on the meas- urement of regional homogeneity (ReHo) of brain regions. Using this technique, 2 locations showing higher ReHo in the cerebellum (i.e. the bilateral anterior inferior cerebellum, AICb) were identified and used as the seeds for RSFC pat- terns studies. We found that the bilateral AICb show significant functional con- nectivity with the bilateral thalamus, the bilateral hippocampus, the precuneus, the temporal lobe and the prefrontal lobe. Further, the differences of RSFC pat- terns between the bilateral AICb were ascertained by a random effect paired t- test. These findings may improve our understanding of cerebellar involvement in motor and a variety of non-motor functions.Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2004, 7th International Conference Saint-Malo, France, September 26-29, 2004, Proceedings, Part II; 01/2004
- [Show abstract] [Hide abstract]
ABSTRACT: Current analysis of diffusion tensor imaging (DTI) is based mostly on a region of interest (ROI) in an image dataset, which is specified by users. This method is not always reliable, however, because of the uncertainty of manual specification. We introduce an improved fiber-based scheme rather than an ROI-based analysis to study in DTI datasets of 31 normal subjects the asymmetry of the cingulum, which is one of the most prominent white matter fiber tracts of the limbic system. The present method can automatically extract the quantitative anisotropy properties along the cingulum bundles from tractography. Moreover, statistical analysis was carried out after anatomic correspondence specific to the cingulum across subjects was established, rather than the traditional whole-brain registration. The main merit of our method compared to existing counterparts is that to find such anatomic correspondence in cingulum, a scale-invariant parameterization method by arc-angle was proposed. It can give a continuous and exact description on any segment of cingulum. More interestingly, a significant left-greater-than-right asymmetry pattern was obtained in most segments of cingulum bundle (-50-25 degrees), except in the most posterior portion of cingulum (25-50 degrees).Human Brain Mapping 03/2005; 24(2):92-8. · 6.88 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Brain imaging data are generally used to determine which brain regions are most active in an experimental paradigm or in a group of subjects. Theoretical considerations suggest that it would also be of interest to know which set of brain regions are most interactive in a given task or group of subjects. A subset of regions that are much more strongly interactive among themselves than with the rest of the brain is called here a functional cluster. Functional clustering can be assessed by calculating for each subset of brain regions a measure, the cluster index, obtained by dividing the statistical dependence within the subset by that between the subset and rest of the brain. A cluster index value near 1 indicates a homogeneous system, while a high cluster index indicates that a subset of brain regions forms a distinct functional cluster. Within a functional cluster, individual brain regions are ranked at the center or at the periphery according to their statistical dependence with the rest of that cluster. The applicability of this approach has been tested on PET data obtained from normal and schizophrenic subjects performing a set of cognitive tasks. Analysis of the data reveals evidence of functional clustering. A comparative evaluation of which regions are more peripheral or more central suggests distinct differences between the two groups of subjects. We consider the applicability of this analysis to data obtained with imaging modalities offering higher temporal resolution than PET.NeuroImage 03/1998; 7(2):133-49. · 6.25 Impact Factor
Y. Liu, T. Jiang, and C. Zhang (Eds.): CVIBA 2005, LNCS 3765, pp. 13 – 23, 2005.
© Springer-Verlag Berlin Heidelberg 2005
Advances on Medical Imaging and Computing
Tianzi Jiang, Xiaobo Li, Gaolong Gong, Meng Liang, Lixia Tian, Fuchun Li,
Yong He, Yufeng Zang, Chaozhe Zhu, Shuyu Li, and Songyuan Tang
National Laboratory of Pattern Recognition, Institute of Automation,
Chinese Academy of Sciences, Beijing 100080, P. R. China
Abstract. In this article, we present some advances on medical imaging and
computing at the National Laboratory of Pattern Recognition (NLPR) in the
Chinese Academy of Sciences. The first part is computational neuroanatomy.
Several novel methods on segmentations of brain tissue and anatomical
substructures, brain image registration, and shape analysis are presented. The
second part consists of brain connectivity, which includes anatomical
connectivity based on diffusion tensor imaging (DTI), functional and effective
connectivity with functional magnetic resonance imaging (fMRI). It focuses on
abnormal patterns of brain connectivity of patients with various brain disorders
compared with matched normal controls. Finally, some prospects and future
research directions in this field are also given.
It is well known that information technology and biomedical technology are two of
the hottest sciences in twenty-first century. Medical imaging is the convergence of
them. Under such a trend, the Division of Medical Imaging and Computing (MIC) at
the National Laboratory of Pattern Recognition (NLPR) in the Chinese Academy of
Sciences was established in 2001. It is a new group that brings together
multidisciplinary expertise in computer science, mathematics, physics, medical
imaging, medicine, and neuroscience. It pursues a scientifically coherent program of
internationally competitive research in Quantitative Imaging, Image and Signal
Computing, and Medical Computer Vision - building on established strengths in these
areas. Three major fields have been involved. One is Computational Neuroanatomy
(especially on relationships between anatomical abnormalities and mental diseases)
including brain tissue segmentation of MR images, intra- and inter-modality image
registration, automatic lesion detection and segmentation, brain structure
segmentation, registration and shape analysis. The second one is Brain Connectivity,
which includes detection of brain activation regions and functional connectivity
analysis with functional Magnetic Resonance Imaging (fMRI), Diffusion Tensor
Imaging (DTI) based white matter bundle tracking and analysis, and studies on brain
connectivity abnormalities of patients with mental diseases with fMRI and DTI. The
third one is the imaging genome. The motivation is to understand genetic bases for
various anatomical and functional abnormalities of patients with brain diseases and
14 T. Jiang et al.
disorders based on neuroimages. We will introduce the historical achievements,
current progress, and future plans of these research fields in the follow on sections
2 Computational Neuroanatomy
In this part, we focus on human brain morphemetry with Magnetic Resonance
Imaging (MRI), especially on brain image and lesion segmentation, registration, and
2.1 Image Segmentation
Brain Tissue Segmentation: Brain tissue segmentation is an important preprocessing
step in many medical research and clinical applications. However, intensity
inhomogeneities in MR images, which can change the absolute intensity for a given
tissue class in different locations, are a major obstacle to any automatic methods for
MR image segmentation and make it difficult to obtain accurate segmentation results.
In order to address this issue, we proposed a novel method called Multi-context fuzzy
clustering (MCFC) based on a local image model for classifying 2D and 3D MR data
into tissues of white matter, gray matter, and cerebral spinal fluid automatically .
Experimental results on both real MR images and simulated volumetric MR data
show that the MCFC outperforms the classic fuzzy c-means (FCM) as well as other
segmentation methods that deal with intensity inhomogeneities, as shown in Fig. 1.
(a) (b) (c)
Fig. 1. 3-D renderings of the segmentation results of WM (a) FCM segmentation (b) MCFC
segmentation (c) true model
Another related work done in the MIC is Pixon based image segmentation. Markov
random fields (MRF)-based methods are of great importance in image segmentation,
for their ability to model a prior belief about the continuity of image features such as
region labels, textures, edges. However, the main disadvantage of MRF-based
methods is that the objective function associated with most nontrivial MRF problems
is extremely nonconvex, which makes the corresponding minimization problem very
time consuming. We combined a pixon-based image model with a Markov random
field (MRF) model under a Bayesian framework . The anisotropic diffusion
equation was successfully used to form the pixons in our new pixon scheme.
Experimental results demonstrate that the proposed method performs fairly well and
Advances on Medical Imaging and Computing 15
computational costs decrease dramatically compared with the pixel-based MRF
Brain Sub-Structure Segmentation: Accurate volumetric segmentation of brain
structures, such as the brain ventricles, is needed for some clinic applications. In
recent years, the active-models-based segmentation methods have been extensively
studied and widely employed in medical image segmentation and have achieved
considerable success. Unfortunately, the current techniques are going to be trapped in
undesired minimum due to the image noise and pseudoedges. We proposed a parallel
genetic algorithm-based active model method and applied it to segment the lateral
ventricles from magnetic resonance brain images . First, an objective function was
defined. Then one instance surface was extracted using the finite-difference method-
based active model and used to initialize the first generation of a parallel genetic
algorithm. Finally, the parallel genetic algorithm was employed to refine the result.
We demonstrate that the method successfully overcomes numerical instability and is
capable of generating an accurate and robust anatomic descriptor for complex objects
in the human brain, such as the lateral ventricles. This is first time in literature to
introduce clustering distributed computing in medical image analysis. It is very
promising and cheap to increase the speed, which is especially needed for some real
time clinic applications.
Active shape models (ASM) was proposed by Cootes  as shape and appearance
models. The method makes full use of priori shape and appearance knowledge of
object and has the ability to deform within some constraints. Rather than representing
the image structure using intensity gradients, we extracted local edge features for each
landmark using steerable filters in , which provided richer edge information than
intensity. We proposed a machine learning algorithm based on AdaBoost, which
selected a small number of critical features from a larger set and can yield extremely
efficient classifiers. These non-linear classifiers were used, instead of the linear
Mahalanobis distance, to find optimal displacements for landmarks by searching
along the direction perpendicular to each landmark. These features give more accurate
and reliable matching between models and new images than modeling image intensity
alone. Experimental results demonstrated the ability of this improved method to
accurately align and locate edge features.
2.2 Image Registration
Image registration is a key component of computational neuroanatomy. In terms of
satisfying the technical requirements of robustness and accuracy with minimal user
interaction, rigid registration has been considered by many works in the field to be a
solved problem. Now the research focus of medical image registration has been
shifted to the non-rigid registration. Neuroscientists and clinicians are in dire need of
the automatically medical image registration tools to process intra-subject, inter-
subject and atlas registration. The method of non-rigid medical image registration
usually include physics-based and geometry-based. We have made our effort on both
Physics-Based Method: We developed a fast fluid registration method in , which
was based on the physics rule of fluid mechanics, and developed another non-rigid
16 T. Jiang et al.
medical image registration algorithm, by assuming that the displacement fields were
constrained by Maxwell model of viscoelasticity. In Fig. 3, applications of the
proposed method to synthetic images and inter-subject registration of brain
anatomical structure images illustrate the high efficiency and accuracy.
A non-rigid Medical Image Registration by Viscoelastic Model was presented in
, by assuming the local shape variations were satisfied the property of Maxwell
model of viscoelasticity, the deformable fields were constrained by the corresponding
partial differential equations. Experimental results showed that the performance of
proposed method were satisfactory in accuracy and speed.
Fig. 2. Top: the template slices; and Bottom: the corresponding slices of reference
Geometry-Based Method: Non-rigid registration of medical image by linear singular
blending techniques was proposed by Tang and Jiang . A free-form deformation
was based on a LSB B-Spline, which can enhance the shape-control capability of B-
Spline. The experiment results indicate that the method is much better to describe the
deformation than the affine algorithm and B-Spline techniques.
2.3 Shape Analysis
Statistical Shape Analysis (SSA) is a powerful tool for noninvasive studies of
pathophysiology and diagnosis of brain diseases. It is another key component of
computational neuroanatomy. The population-based shape analysis not only reveals
the difference between the healthy and diseased subjects, but also provides the
dynamic variations of the patients’ brain structures over time. We proposed a new
method which incorporated shape-based landmarks into parameterization of banana-
like 3D brain structures to address this problem . First, the triangulated surface of
the object was obtained and two landmarks were extracted from the mesh, i.e. the
ends of the banana-like object. Then the surface was parameterized by creating a
continuous and bijective mapping from the surface to a spherical surface based on a
heat conduction model. The correspondence of shapes was achieved by mapping the
two landmarks to the north and south poles of the sphere. The approach was applied
to the parameterization of lateral ventricle and a multiresolution shape representation
was obtained by using the Discrete Fourier Transform, as shown in Fig.3.
Advances on Medical Imaging and Computing 17
Fig. 3. Parameterization of lateral ventricle and a multiresolution shape representation obtained
by using the Discrete Fourier Transform, with Fourier coefficient 9, 13, 29, 49, 81, 1257, and
3 Diffusion Tensor Imaging
Diffusion tensor imaging, as a relatively new MRI technique, provides information
about the random displacement of water molecules in the brain tissue. Using this
information, ones could investigate the white matter characteristic and the anatomical
connections between different regions non-invasively. Our research efforts cover a
wide range from developing new approaches to fiber tracking and the white matter
analysis using DTI.
(a) (b) (c)
Fig. 4. (a) The reconstructed cingulum tract. (b) The statistical results for asymmetry of right-
handers. (c) The statistical results for the effects of handerness and side in cinulum.
Analysis of white matter from DTI is mostly based on region of interesting (ROI)
in image data set, which is specified by user. However, this method is not always
reliable because of the uncertainty of manual specification. In , we developed an
improved fiber-based scheme rather than ROI-based analysis to study the cingulum,
the most prominent white matter fiber tract of the limbic system. In this work,
cingulum bundles were first reconstructed by fiber-tracking algorithm and then were
parameterized by arc-angle, which was scale-invariant. All fibers centered at a
common origin, and anatomic correspondence across subjects in cingulum was
18 T. Jiang et al.
established. This method was used to explore the asymmetry of cingulum in DTI
datasets of right-hander normal subjects. As in Fig. 4, statistical results show a
significant left-greater-than-right asymmetry pattern in most segments of cingulum
bundle, except the most posterior portion.
4 Functional Brain Connectivity
fMRI is an important functional brain imaging technique. Since its advent in the early
1990’s, a variety of analytic methods have been developed. This is one important
research field in the MIC. Recently, what we have been mostly concerning about are
methods for steady-state (including resting-state) fMRI data and connectivity.
4.1 fMRI Activation Detection
Both the spatial and temporal information of fMRI data have been considered in our
research on fMRI activation detection. The temporal information is typically the time
variant characteristics of the hemodynamic responses, and the spatial information is
the fMRI activated regions typically occur in clusters of several spatially contiguous
voxels [11, 12]. Another research direction was how to model the trial to trial
variability of the hemodynamic responses in human brain. Since the typical model-
based methods for fMRI data analysis, for instance, the General Linear Model (GLM)
and the deconvolution method, are based on the following assumption: the
hemodynamic responses are same across trials, i.e., the trial-to-trial variability is
considered as noise. When hemodynamic responses vary from trial to trail , an
alternative approach is needed to include the trial-to-trial variability.
A region-growing and a split-merge based method have been proposed for fMRI
activation detection [14, 15]. Main feature of the region-growing and split-merge
based methods was that they can utilize the neighboring information of each time
series. As for the second aim, an optimization based framework for single trial
variability was proposed in . The main features of this proposed method were as
follows: (1) The trial-to-trial variability was modeled as meaningful signal rather than
assuming that same HR was evoked in each trial; (2) Since the proposed method was
a constrained optimization based general framework, it could be extended by utilizing
of prior knowledge of HR; (3) The traditional deconvolution method could be
included into our method as a special case.
4.2 Regional Homogeneity (ReHo)
As a result that no specific stimulus was given for resting-state, we proposed a data-
driven method call regional homogeneity (ReHo) in . ReHo assumed that voxels
within a functional cluster should share similar characteristics and such similarity
could vary from state to state. By this method, we found significant higher ReHo in
bilateral primary motor cortex (M1) during unilateral finger tapping. Higher ReHo
was also detected in posterior cingulate cortex (PCC) during resting-state.
Advances on Medical Imaging and Computing 19
4.3 Functional Connectivity
A large quantity of fMRI studies traditionally focused on identifying activated regions
of the brain during an experimental task. However, brain function may be as a result
of information integration among brain regions, described in terms of functional
connectivity  or effective connectivity . Recently, increasing attention has
been focused on detecting interregional connectivity, especially in a resting state. Our
research efforts cover a wide range from developing new methodology to using
established methods for the understanding of the resting brain mechanism and clinical
validation. Some representative contributions were as follows.
All-To-All Functional Connectivity : We developed a new approach based on
graph theory taking into account n-to-1 connectivity using 1-to-1 connectivity
measures instead of conventional pairwise connectivity. It can better reflect the joint
interactions among multiple brain regions. With it, a large organized functional
connectivity network related to motor function in the resting brain with fMRI was
shown. More importantly, we found that such a network can be modulated from a
conscious resting state to a voluntary movement state by the measure.
ReHo and Functional Connectivity : So far, the resting state functional
connectivity (RSFC) has been performed mainly by seed correlation analysis (SCA)
on fMRI studies. On the basis of our previous work based on the ReHo ,
functional network in the resting brain was detected by using the model-free method.
Our method identified a parietal-frontal network of brain regions utilizing only resting
state data. We proposed the ReHo to serve as the selection of the seed before
functional connectivity based on the SCA is performed . It provides a novel way
to select the desired seeds by taking the natures of resting state data into account,
compared with the previous studies about the selection of the seed utilizing prior
anatomical information  or previously performed activation maps . Using this
technique, the bilateral anterior inferior cerebellum (AICb) showing higher ReHo
were identified and used as the seeds for resting state functional connectivity patterns
studies. The results show that the bilateral AICb has significant functional
connectivity with the bilateral thalamus, the bilateral hippocampus, the precuneus, the
temporal lobe, and the prefrontal lobe.
5 Current Activities
Much of our current work on computational neuroanatomy is concerned with
structural abnormalities in human brain. This is essential for the research of mental
diseases. Under the consideration of the limited resolution of the existing imaging
sensors and the low contrast of brain structures to their peri-structures, robust tools for
the identification of such structures are highly expected, so that the structures can be
quantitatively and statistically analyzed. The research on geometrical fairing mapping
using Mean Curvature Flows is now undergoing. It is applicable for arbitrary genus
surfaces and avoids shape shrinkage in discrete space. The global geometry of the
original object and its area are theoretically preserved. This mapping method has the
potential applicability in skeleton regularization and brain surface matching.
20 T. Jiang et al.
Our previous study on DTI was confined to right-handers without considering the
role of handedness. It should be interesting to ascertain the relationship of the
cingulum microstructure with the handedness and side. We therefore recruited another
group of left-handed healthy subjects to examine this issue with the same method
mentioned above. The statistical results also showed a remarkable left-greater-than-
right asymmetry pattern of anisotropy in most segments of cingulum bundles except
the most posterior segment. Higher anisotropy of the right-hander than the left-hander
was found in the bilateral cingulum bundles. However, no significant handedness-by-
side interaction was observed. Besides the applications to brain research based on
DTI, a lot of computational issues are also under our consideration. The tensor model
is explicit, however is too simple to characterize the property in some complicated
regions such as fiber crossing regions and boundary regions between different brain
tissues. More suitable models are under construction in our group.
Fig. 5. Highly discriminative regions identified by Fisher brain
Discriminative analysis on fMRI has also been concerned. A discriminative model
of attention deficit hyperactivity disorder (ADHD) was newly presented on the basis
of multivariate pattern classification and fMRI . This model consists of two parts,
a classifier and an intuitive representation of discriminative pattern of brain function
between patients and normal controls. Regional homogeneity (ReHo), a measure of
brain function at resting-state, is used here as a feature of classification. Fisher
discriminative analysis (FDA) is performed on the features of training samples and a
linear classifier is generated. The classifier is also compared with linear support
vector machine (SVM) and Batch Perceptron. Our classifier outperforms its
counterparts significantly. Fisher brain, the optimal projective-direction vector in
FDA, is used to represent the discriminative pattern. As shown in Fig. 5, some
abnormal brain regions identified by Fisher brain, like prefrontal cortex and anterior
cingulate cortex, are well consistent with that reported in neuroimaging studies on
ADHD. Moreover, some less reported but highly discriminative regions are also
Advances on Medical Imaging and Computing 21
identified. The discriminative model has potential ability to improve current diagnosis
and treatment evaluation of ADHD.
We also apply discriminative analysis on DTI. We use two-dimensional histogram
of ADC and FA to discriminate Neuromyelitis optica (NMO) from healthy subjects.
The correct recognition rate reaches 85%, which is much higher than that of based on
the traditional FA histogram (50%) and ADC histogram (73%). The results indicate
that our method based on combined feature from two-dimensional histogram is more
effective for classification than that of based on one type of feature. Furthermore,
some discriminative regions that contribute most to separating the patients with NMO
and normal controls can be obtained based on our method. It implies that NMO has
diffuse damages of brain tissue on such a small scale that conventional MRI cannot
detect it. This challenges the classic notion of a sparing of the brain tissue in the
course of NMO. In addition, our method based on two-dimensional histogram can
also be used in other brain tissue, especially the diffuse damage of brain tissue, such
as multiple sclerosis.
The application of functional connectivity analysis in ADHD is undergoing. We
investigate the difference of functional connectivity between the Attention-
Deficit/Hyperactivity Disorder (ADHD) children and the normal controls in Flanker
task and resting state, respectively. In Flanker task, we found that ADHD children
show enhanced functional connectivity between dACC and several other brain areas
as shown in Fig. 6. Such an enhanced functional connectivity pattern may suggest that
children with ADHD need greater effort, and accordingly a wider network of brain
areas to complete the same task as the normal controls do. In a resting state, we found
that the ADHD patients exhibited more significant functional connectivity of the
dACC with the dACC, as well as with the left medial frontal cortex/ventral anterior
cingulate cortex, bilateral thalamus, bilateral cerebellum, bilateral insula, right
superior frontal cortex, and brainstem (pons), and only within the brainstem (medulla)
did the controls exhibit more significant connectivity than the patients. More
information is available at http://www.nlpr.ia.ac.cn/jiangtz.
Fig. 6. The brain regions showing significant between-group differences in the Flanker task
(p<0.05 corrected, upper row) and in the resting state (p<0.05 & cluster size >600mm3, lower
22 T. Jiang et al.
The application of ReHo in AD is being evaluated. We used resting-state fMRI to
examine LFFs activity in terms of regional homogeneity (ReHo) to explore the
pathophysiology of dementia of the Alzheimer type (DAT). Compared with healthy
controls, DAT subjects showed decreased ReHo in the posterior cingulate cortex
(PCC) and increased ReHo in the right inferior temporal cortex and bilateral occipital
lobe. In addition, examination of the behavioral correlates revealed significant
positive correlation of PCC ReHo versus Mini-Mental State Examination score. Our
finding, together with a recent fMRI result that DAT subjects showed decreased
resting-state activity in the PCC and hippocampus explored by using a non-pure
resting-state paradigm, suggests that PCC LFFs activity measured using resting-state
fMRI may be a promising marker for characterization and detection of early DAT.
6 Conclusions and Future Directions
We have developed various techniques to detect the anatomical and functional
abnormalities of human brain with neurological and psychiatric diseases. We have
been applying various modern neuroimaging techniques to combat the neurological
and psychiatric diseases, especially Alzheimer's Diseases and Schizophrenia. A long-
term goal of the MIC is to find the early markers based on neuroimages and genome
datasets for the neurological and psychiatric diseases. It would be very interesting to
identify the genetic basis of the anatomical and functional abnormalities of human
brain with neurological and psychiatric diseases. In fact, several publications have
been available and a new field - imaging genomics, named by Hariri and Weinberger,
has emerged . It is at its infant stage and we expect a lot of important progress
can be made in the future.
This work was partially supported by the Natural Science Foundation of China, Grant
Nos. 30425004 and 60121302, and the National Key Basic Research and
Development Program (973), Grant No. 2004CB318107.
1. Zhu CZ, Jiang TZ: Multi-context Fuzzy Clustering for Separation of Brain Tissues in MR
Images. NeuroImage 18 (2003) 685–696
2. Yang FG, Jiang TZ: Pixon-Based Image Segmentation With Markov Random Fields.
IEEE Trans Imag Proc 12 (2003) 1552–1559
3. Fan Y, Jiang TZ, Evans DJ: Volumetric Segmentation of Brain Images Using Parallel
Genetic Algorithm. IEEE Trans Med Imaging 21(2002) 904–909
4. Cootes TF, Taylor, CJ, Cooper, DH, Graham, J: Active shape models- their training and
application. Computer vision and image understanding 61(1995) 38–59
5. Li SY, Zhu LT, Jiang TZ (2004): Active Shape Model Segmentation Using Local Edge
Structures and AdaBoost. In proceedings of MIAR 2004, 121–128, Beijing, China
Advances on Medical Imaging and Computing 23
6. Tang SY, Jiang TZ: Fast Nonrigid Medical Inage Registration by Fluid Model. In
proceedings of ACCV’04, vol. II, (2004a) 914-919, Jeju, Korea
7. Tang SY, Jiang TZ: Nonrigid Registration of Medical image By Maxwell Model of
Viscoelasticity. In proceedings of ISBI'04, (2004b) 1443–1446, Arlington, USA
8. Tang SY, Jiang TZ: Non-rigid Registration of Medical Image by Linear Singular Blending
Techniques. Pattern Recogn Lett 25(2004c) 399–405
9. Zhu LT, Jiang TZ: Parameterization of 3D Brain Structures for Statistical Shape Analysis.
In proceedings of SPIE Medical Imaging’04, (2004) 14-17, California, USA
10. Gong GL, Jiang TZ, Zhu CZ, Zang YF, Wang F, Xie S, Xiao JX, Guo XM: Asymmetry
Analysis of Cingulum Based on Scale-Invariant Parameterization by Diffusion Tensor.
Hum Brain Mapp 24(2005) 92-98
11. Tononi G, McIntosh AR, Russell DP, Edelman GM: Functional clustering: identifying
strongly interactive brain regions in neuroimaging data. Neuroimage 7(1998) 133–149
12. Katanoda K, Matsuda Y, Sugishita M: A spatial-temporal regression model for the
analysis of functional MRI data. Neuroimage 17(2002) 1415–1428
13. Duann JR, Jung TP, Kuo WJ, Yeh TC, Makeig S, Hsieh JC, Sejnowski JT: Single-Trial
Variability in Event-Related BOLD Signals. NeuroImage 15(2002) 823-835
14. Lu YL, Jiang TZ, Zang YF: A Split-merge Based Region Growing Method for the
Analysis of fMRI Data. Hum Brain Mapp 22(2004) 271–279
15. Lu YL, Jiang TZ, Zang YF: Region Growing Method for the Analysis of fMRI Data.
NeuroImage 20(2003) 455–465
16. Lu YL, Jiang TZ, Zang YF: Single-Trial Variable Model for Event-Related fMRI Data
Analysis, IEEE Transactions on Medical Imaging 24(2005) 236-245
17. Zang YF, Jiang TZ, Lu YL, He Y, Lixia Tian: Regional Honogeneity Based Approach to
fMRI Data Analysis. NeuroImage 22 (2004) 394–400
18. Friston KJ, Frith CD, Liddle PF, Frackowiak RS: Functional connectivity: the principal
component analysis of large (PET) data sets. J Cereb Blood Flow Metab 13(1993a) 5–14
19. Friston KJ, Frith CD, Frackowiak RS: Time-dependent changes in effective connectivity
measured with PET. Hum Brain Mapp 1(1993b) 69–80
20. Jiang TZ, He Y, Zang YF, Weng XC: Modulation of Functional Connectivity During the
Rest State and the Task State. Hum Brain Mapp 22 (2004) 63–71
21. He Y, Zang YF, Jiang TZ, Liang M, Gong GL: Detecting Functional Connectivity of the
Cerebellum Using Low Frequency Fluctuations (LFFs). In proceedings of MICCAI'04,
vol. II, (2004) 907-915, St Malo, France
22. Stein T, Moritz C, Quigley M, Cordes D, Haughton V, Meyerand E: Functional
connectivity in the thalamus and hippocampus studied with fMRI. Am J Neuroradiol
23. Zhu C, Zang Y, Liang M, Tian L, He Y, Li X, Sui M, Wang Y, Jiang T: Discriminative
Analysis of Brain Function at Resting-state for Attention-Deficit/Hyperactivity Disorder.
In proceedings of MICCAI'05, USA, (2005) in press.
24. Hariri AR, Weinberger DR: Imaging genomics. Br Med Bull. 65(2003) 259-70.