Open Access

Physiological basis and image processing in functional magnetic resonance imaging: Neuronal and motor activity in brain

BioMedical Engineering OnLine20043:13

DOI: 10.1186/1475-925X-3-13

Received: 12 January 2004

Accepted: 05 May 2004

Published: 05 May 2004

Editor's Note

This article has been retracted. The retraction notice can be found here.

The Retraction Note to this article has been published in BioMedical Engineering OnLine 2013 12:113


Functional magnetic resonance imaging (fMRI) is recently developing as imaging modality used for mapping hemodynamics of neuronal and motor event related tissue blood oxygen level dependence (BOLD) in terms of brain activation. Image processing is performed by segmentation and registration methods. Segmentation algorithms provide brain surface-based analysis, automated anatomical labeling of cortical fields in magnetic resonance data sets based on oxygen metabolic state. Registration algorithms provide geometric features using two or more imaging modalities to assure clinically useful neuronal and motor information of brain activation. This review article summarizes the physiological basis of fMRI signal, its origin, contrast enhancement, physical factors, anatomical labeling by segmentation, registration approaches with examples of visual and motor activity in brain. Latest developments are reviewed for clinical applications of fMRI along with other different neurophysiological and imaging modalities.


Functional magnetic resonance imaging image processing segmentation, registration visual activity motor activity neurophysiological activity


Recent investigations focused on specific brain regional and functional specificity to delineate the specific distribution of neural activities at a given moment in the brain as a whole. Functional magnetic resonance imaging (fMRI) was introduced to map the changes in brain local blood flow and oxygenation or hemodynamics that correspond to regional neuronal activity of brain accompanying metabolic events. It extended brain anatomical imaging to map structures and specific function of human brain. High resolution, noninvasive neural activity by a blood oxygen level dependent signal has tremendous potentials for assessing the neurological status and neurosurgical risk [14]. Later fMRI applications extended the understanding of neuronal and motor activities associated with different brain regional functions. Presently, fMRI serves as non-invasive imaging of neurophysiological activities of brain that depend more on physiological characteristics of brain.

The paper reviews the physiological basis of fMRI signal origin and contrast mechanisms with state-of-art fMRI segmentation and registration algorithms to identify cortical visual response and event related cortical areas associated with neurophysiological measurements and potential image post-processing directions in future.


The physiological basis of fMRI


Neurovascular and neurometabolic coupling establishes the critical link between a focal change in neuronal activity and MRI-detectable observations. In fact, all task performances such as arousal, attention, alertness, adaptation, sleep, or consciousness that affect the vascular hemodynamics do interfere with oxygenation-sensitive mapping by fMRI techniques.

Historically, these observations initially were supported by reports on local reduction in deoxyhemoglobin due to increased blood flow without change in oxygen extraction [5]. Deoxyhemoglobin acts as paramagnetic endogenous contrast agent and alters the T2* weighted magnetic resonance image signal [69] and serves as the source of the signal for fMRI. Last decade was an excitement for clinical application of 1.5 T-7.0 T clinical scanners to observe functional activity of visual cortex [1216], the motor cortex [1821] and Broca's area of speech and language-related activities [20, 21]. fMRI and conventional neurophysiological techniques have been in use to localize the specific functions of the human brain [2227].

Increased neuronal activity needs the metabolic support. For that, blood flow provides the substrates. Still there is paucity in information of metabolic requirements and hemodynamic response in different brain functions. Recent trend was focused on identification of brain regions involved with characteristic oxygenation-sensitive MRI response function.

The visual response function

The oxygen concentration in brain serves as a tool to map cortical regions responsible for performing various cognitive tasks because oxygenation level in active cortex changes between baseline and tasking conditions i.e. pattered lights protocols affect the spatiotemporal response and characteristics in the visual system. These visual stimulations generate the signal rise due to differences between tonic and phasic MRI hemodynamic responses after the onset of activation i.e. rapid rise in BOLD response due to rapid increase in the blood flow or enhanced oxygen delivery / oxygen consumption. Recently, the delayed upregulation of oxidative glucose consumption in brain and a slow venous blood volume (balloon model) suggested them as two processes. These were relevant for fMRI mapping studies with shorter protocol timings [28]. The link between neuronal activity and blood flow characteristics forms the basis for functional mapping using fMRI. These characteristics such as cerebral blood flow (CBF), cerebral volume (CBV), metabolic regional oxygen (CMRO2), and BOLD signal form an interconnected set of quantities that are coupled during normal brain activation.

Tissue oxygen and framework for BOLD Signal

fMRI images can be made sensitive to local oxygen concentrations in tissue. BOLD signal derives from the local concentration of deoxygenated hemoglobin that is modulated by several factors. The generator of this paramagnetic contrast agent is oxygen metabolism (CMRO2). Blood oxygenation and blood magnetization both depend upon the balance of oxygen flow into and out of a region. The rate of oxygen inflow is proportional to cerebrovascular blood rate (CBR). During functional brain activation, increased CBF produces a washout of Hbr as contrast agent by counteracting the effect of increased CMRO2. Local blood volume fraction determines the deoxyhemoglobin content of a voxel at any level of blood oxygenation. As blood vessels swell, magnetic fields extend further into the tissue, causing a signal loss in the extravascular space. BOLD contrast can be approximated as changes in the BOLD relaxation rate scale with changes in the deoxy hemoglobin concentrations i.e. BOLD contrast (X) = K.A [Hbr], where 'x' depends upon the magnetic field strength and the sample volume.

A BOLD framework is based upon conservation of oxygen mass (Fick's Law) i.e. at the steady-state, unidirectional extraction of oxygen from the blood is the difference between the 'flow' of oxygen 'into' and 'out' of the volume, FO2 IN -, FO2 OUT = dV/dt. The resulting expression takes a form like the following:

ΔR2 = -K [Hbr]o{ΔF/Fo - ΔV/Vo - ΔM/Mo}    (1)

F, V, and M refer to CBF, CBV, and CMR02 respectively. Subscript "o" indicates baseline values prior to stimulation. BOLD signal changes are positive when the quantity in brackets is positive. Positive stimulus-induced BOLD represent the relative changes in CBF that exceed over combined effect of changes in CBV and CMR02. [Hbr]o is proportional to V and Mo and inversely proportional to F as shown in Figure 1 that combine to generate BOLD signal. Equation (1) stands good for small functional changes while intravascular signal contributions affect the linear relationship between ΔR2, and Δ[Hbr] dependence on blood oxygen and blood volume. The term ΔV/Vo is relative change in total venous hemoglobin.
Figure 1

Figure represents the "Oxygen oversupply" hypothesis. Regional deoxy-Hb (thick solid lines) decline is associated with increase in regional blood flow (thin solid line) and oxygen consumption (broken lines). In fMRI, T2* is described as blood flow based functional imaging of brain activation as sum of all activation pixels (yellow color) from all serial images shown in the figure 1 (on right).

A rapid change in CBF produces an effect on BOLD signal that is both delayed and dispersed by transit through the vascular bed. The framework of Equation (1) of BOLD signal does not specify how oxygen is allocated, as long as the total amount is conserved. However, the quantities {ΔF/Fo - ΔV/Vo - ΔM/Mo} do not change arbitrarily during brain activation. The reproducibility of BOLD results across systems and BOLD stimuli poorly represent the coupling.


Regional basal cerebral oxygen and glucose utilization show a molar ratio consistently less than 6(CMRO2/ CMRglu - 5.5), suggesting that the oxidative glucose metabolism (C6H1206 + 6 O2 + 6H20 + 6C02) is the primary source of energy. Tight linear couplings have been shown for regional basal measurements of CBF versus CMRglu and CBF versus CMR02 [29].

Stimulus-induced activation

Blood oxygen is delivered to the brain by diffusion along an oxygen concentration gradient that falls in the tissue. Brain has low oxygen reserves so blood flow increases the oxygen delivery (temporal resolution) resulting with reduced blood transit time through the capillaries' decreased oxygen 'extraction fraction' or restricted capillary area and vascular resistance at the local level. So, the regional trajectories represent the temporal physiological quantities determined from the baseline or average state i.e. the diffusion and regional coupling of CBF and CMRO2 match during brain activation. The current viewpoint of CBF-CMR02 coupling and regional relationship with rate of oxygen delivery explained the events during brain activation by Buxton-Frank diffusion-limited model of oxygen delivery [30]. The exact scaling between relative changes in blood flow (f) = F/Fd and relative changes in oxygen delivery (M) depend upon the baseline value of the extraction fraction (Eo)and extent of capillary dilation as:
  1. (v)

    = V / Vo    (2)


M = f (1 - (l - Eo)v/f)/ Eo    (3)

Empirically, the influence of CMR02 on BOLD signal can be deduced by comparing the responses of CBF and BOLD signal using stimuli that affect CBF and CMR02 i.e. hypercapnia modulates CBF without changing oxygen-utilization. Focal activation of the human visual cortex increases CMR02 [31]. By using graded levels of visual stimulus and hypercapnia, a linear coupling was measured between relative changes in CBF and CMRO2 for flow [31].


Cerebral vascular resistance is defined as the total pressure drop across a vascular bed. In the brain, intravascular pressure drops from mean arterial blood pressure in large arteries to venous pressure in the large veins. The brain activation increases CBF by reducing cerebral vascular resistance corresponding to an increase in CBV. Blood flow and blood volume both exhibit different temporal responses [32]. However, the basal blood level of deoxygenated hemoglobin is determined by the ratio of CMR02 to CBF. Neurophysiological changes in fMRI alter BOLD signal by resetting the ratio of basal CMR02 to CBF, and altering CBV [33].

Sensitivity of fMRI signal

Blood volume fraction, oxygen extraction fraction, distribution of vessels, arterial oxygenation neurophysiological factors and intravascular or extravascular signals depend upon the pulse sequence, field strength, degree of activation and the physiology of the functional variable etc. Sensitivity is the product of the relative change in brain 'activation' (flow, volume, oxygenation, etc.) and 'amplification' factor expressing the intrinsic sensitivity per unit change:

Sensitivity = (activation) × (amplification)    (4)

The amplification factor applies spatial resolution to the pattern of brain activation but it is independent of the degree of activation [32]. For detecting changes in local brain functional activity, fMRI signal-to-noise ratio (fSNR) refers to the time-averaged value of signal divided by the temporal standard deviation of the signal: fSNR = Stt.

Similarly, contrast for fMRI or functional CNR per unit time (fCNR) may be expressed as the ratio of time dependent signal changes (δS) to time-dependent noise:

fCNR = δStt, fSNR × δSt/St    (5)

Basic functional MRI sequences and functional MRI contrast

FLASH or single-shot EPI sequences with prolonged echo times are employed depending on the desired spatial or temporal resolution. These pulse sequences are shown (Figure 2). Typically, EPI sequences acquire all differently phase-encoded gradient echoes required for image reconstruction after a single slice-selective RF excitation pulse. The individual echoes are generated by multiple sinusoidal or trapezoidal reversal of the read or frequency-encoding gradient. Phase encoding is performed by a 'blipped' gradient, whereas the EPI technique uses a 'weaker' constant gradient. Echoes cover a large range of different echo times. The effective TE is given by the Fourier line representing the lowest spatial frequency, i.e. for zero phase encoding, as it dominates the image contrast. Basic emphasis is on high speed yield and image acquisition times of the order of 100 ms and excellent maximum volume coverage by multi-slice fMRI imaging at the expense of limited in-plane resolution.
Figure 2

A basic EPI sequence FLASH (top) and single-shot EPI Sequence (bottom) for functional Imaging is shown with reversible reading gradients and pulses to generate rapid images in less than a minute.

In contrast, FLASH sequences require multiple RF excitations with low flip angles < 90° that normally generate only a single gradient echo per repetition interval. As large TE values also prolong the repetition time, typical imaging times are in the range of several seconds. The ability to select an arbitrary compromise between temporal and spatial resolution is best exploited for gaining access to high-resolution maps at the expense of less volume coverage. However, EPI images also suffer from several unavoidable artifacts.

Neurophysiological factors in functional MR contrast

Various measures are currently used to identify the activated pixels in corresponding fMRI maps i.e. analysis of signal differences, variances, statistical parameters, temporal correlations or frequencies, principal components, clusters, phase information, and noise characteristics. In visual response, time-locked averaging of images and subsequent subtraction across the functional states i.e. summation of images was reported that was acquired during one condition (e.g. lights off) and subtraction of the result from that obtained for a different condition (e.g. lights on) [33]. This robust and sensitive approach exploited the temporal structure of the known stimulation protocol and compared it to the oxygenation-sensitive MRI signal intensity time courses, on a pixel-by-pixel basis. This 'boxcar' function was employed to calculate the color-coded activation map for correlation coefficients identifying the activation centers and optimum area delineation i.e. retinotopic maps on brain V1, V2, V4 and MT (visual cortex) measured by fMRI for establishing the accuracy of visual maps as basis of hemodynamic responses in these two cortical areas [33]. The stimuli used block-alternation design with relatively long intervals of stimulus vs rest state. However, fMRI has been widely used to image ocular dominance and orientation columns within a fraction of a millimeter [3436].

Validation and physical factors in Functional MRI

BOLD and fMRI characteristics are important determinants in validation process of ultrafast fMRI image acquisition of raw data and 'task to functional map' correlation by postprocessing and statistical analysis. We describe these concepts that validate brain activation, fMRI spatial resolution and BOLD events. 'Brain activation' is read as 'statistically significant pixel intensity changes' associated with a given set of tasks to denote the area of neuronal activation. It is a scalar number given by MRI system upon completion of image reconstruction including magnetic correction and other inherent factors. MRI spatial resolution is low in the range of 4 mm3 on high field MR scanners. BOLD events are neuronal events. T2* contrast changes seen in fMRI are an empirically observed biological phenomenon. For fMRI, T2* contrast is 'weak' susceptibility effect of deoxy-hemoglobin (deoxy-Hb).

Susceptibility effects and T2* contrast

In MRI, regional field inhomogeneity is common due to high paramagnetic susceptibility, ferromagnetic susceptibility and air sinuses, Bo inhomogeneity. The susceptibility effect may affect a large area of the image matrix and can introduce image deformity. The smaller susceptibility effects introduce the pixel intensity changes in nearby pixels generating the T2 * contrast. This is the basis of fMRI that represents a 'statistical' method based on pixel intensity changes in the brain placed in high magnetic field B.

Magnet shimming

MRI is performed grossly by using high order gradient 1st and 2nd order shim coils to correct Bo inhomogeneity. Human brain undergoes the continuous motion and hard to shim i.e. the line width of human brain is 200–400 Hz. Slice thickness and slab size over that focused shimming reduces the inhomogeneity. Fast spin echo (FSE) images are insensitive but echo planar images (EPI) are sensitive to inhomogeneity. In fMRI, selective excitation selects the appropriate slice thickness with appropriate inter slice gap between the slices.

Nyquist ghost

The unique k-space trajectory of the EPI sequence results in the appearance of a characteristic artifact termed 'Nyquist ghost'. However, in practice the most common cause of Nyquist ghost is minor field perturbation as shown in Figure 3. Nyquist ghost represents the fictitious activation encountered in fMRI. Direct adaptation of such paradigms to fMRI typically introduces task-correlated Nyquist ghost and fictitious activation.
Figure 3

the figure represents Nyquist ghost in panel A and representative fictitious activation in panel B. Notice the background bright signal as ghost (panel A) and activation areas (panels B) shown in yellow color spots which are not due to brain activation.

Pixel misalignment and limitation of spatial resolution

Pixel misalignment causes the fictitious activation due to subject motion as shown in Figure 4. Many "motion correction" post-processing algorithms have been developed based on the two-dimensional imaging and three-dimensional misalignments. The theoretical limit of the spatial resolution for fMRI is approximately 4 mm3. fMRI image of the normal brain can be used to evaluate the relative intensity of cerebral cortex at various sites relative to CSF as shown in Figure 4. The substantial variation in the intensity of cortex is primarily due to the partial volume phenomenon as shown in Figure 5. The correction algorithms such as 're-slicing', 'standardization', or 'motion correction' image-processing methods as shown in Figure 6 have been reported [37]. Basically, each raw image data is used for statistical analysis. In the following description, some representative examples of application of these physiological principles of fMRI are illustrated.
Figure 4

The figure represents fictitious activation due to pixel misalignment. The bright spots around the bottom brain surface are misaligned that create illusion of active areas shown in yellow color.

Figure 5

An example of an fMRI image of the normal brain is shown. The numbers indicate the relative intensity of cerebral cortex at various sites relative to CSF which assigned a value of 100 (written in red). The substantial variation in the intensity of cortex is primarily due to the 'partial volume' phenomenon.

Figure 6

The figure illustrates the effects of motion correction algorithm SPM 96. The activation maps were obtained for a bilateral hand motion paradigm using a horizontal 3 T MRI system with image voxel resolution of 3 mm × 3 mm × 5 mm. In this setting, acceptable pixel misalignment was determined to be 0.6 mm. Brain motion exceeding 0.6 mm (> 0.6 mm) produced significant pixel misalignment artifact. A motion correction algorithm wiped out these artifacts as well as actual activation. It also wiped out a small cluster of fictitious activation, while 'true' activation remained visible. In contrast, brain motion less than 0.6 mm provided activation maps of bilateral primary motor cortices. Application of motion correction algorithm artificially eliminated true activation areas.

Functional MRI signal of motor and visual stimulation

Two examples of fMRI experiments of motor and visual stimulation are described.

Visual Stimulation and fMRI

The typical BOLD time course (shown in black) shows 4 'active' states and 4 'resting' states are shown in Figure 7. With prior knowledge of the activation timing (shown in red), a statistical test is performed on the data to determine active areas of the brain. In brief, MP-RAGE (magnetization prepared, rapid acquisition gradient echo) sequence generates a 3D anatomic image of the head and brain. fMRI is performed with T2*-weighted gradient recalled EPI. The visual stimuli are created on a visual stimulus generator graphics card. The stimuli are presented dichoptically using polar filters and adjustable right-angle prisms for optical superimposition of the right and left image are shown in Figure 7. The stimuli appear as 'radial checkerboards', in which the high-luminance contrast checks exchanged position as a sinusoidal function of time. During rest (baseline) periods, subjects view a small black fixation mark superimposed onto a homogeneous field. The experimental paradigms contain five different epochs: Alternating monocular stimulation [A]; Simultaneous binocular stimulation [B]; left eye leading-right eye trailing [LR]; Right eye leading-left eye trailing [RL]; and baseline. Each rest epoch is followed by one of the described epochs of checkerboard stimulation. The complete sequence of one repetition is shown in Figures 7 and 8. In a recent report, BOLD contrast in visual cortex related to binocular interactions in primary visual cortex could be revealed by fMRI at high field 4 T MRI. Binocular and monocular stimulations were characteristic of high contrast radial checkerboard pattern-stimulated neurons tuned to high and low spatial frequencies. The different striate cells in ocular dominance columns interacted when they are simultaneously activated and reduced by binocular or monocular stimulation resulting with increased BOLD response [37]. However, binocular rivalry due to disparity appears as a source of error. Fixation of eye and maintaining it throughout scan period reduces the disparity which otherwise is commonly observed in area V3 by random-dot stereogram.
Figure 7

Time course of activation for the four stimulus epochs (B binocular, M monocular RL right eye stimulated first, followed by left eye stimulation, LR left eye stimulation first, followed by right eye stimulation)(left panel). During stimulation period, the subjects perceived a single flickering radial checkerboard, whereas during the rest period they viewed a single black fixation mark at constant mean luminance (right panel). Typical time courses from region of interest (ROI) centered in one hemisphere in the primary visual cortex (VI) or in an extrastriate region. Statistical parametric maps of significant BOLD responses to alternating monocular stimulation compared to the binocular condition (right panel). Voxels in bright regions indicate strong response to alternating monocular stimulus. The cross hairs represent the most active voxel within the cluster used for normalization (bottom panels).

Figure 8

Selected brain regions showing the activation areas observed by fMRI during finger movement. Regions with Z-score higher than threshold of 3.5 are displayed in red color. Stroke occurred in area colored as yellow. The time-course of on-off fMRI signal recorded in a typical voxel responding activation due to the stimulus paradigm(shown as red bold line).

fMRI activation in primary motor and pre-motor regions

Paralyzed patient retains the motor control. fMRI shows the activation in primary motor and premotor regions. An example is illustrated here for event-related finger tapping fMRI image acquisition and analysis. Right-handed subjects paralysed with eye blinking and restricted mouth movement with intact cognition were imaged by fMRI single-shot gradient recalled echo-planar imaging pulse sequence. Patients were simulated for the action of tapping fingers 'stimulus paradigm' as shown in Figure 9. Multiple regressions were applied to do statistical data analysis. Individual variables included 'box-car activation paradigm' and constants for activation signals. A ramp-regressor was used to remove linear-drift in the signal. Standard deviated Z-score map was superimposed on high-resolution anatomical image to display brain activation areas. Time course signal in each voxel was obtained to reveal hemodynamic response to the stimulus paradigm. The regions of activation were mainly in contralateral to the primary motor area (M1) and premotor regions (PM) in right motor cortex with little activation in supplementary motor area (SMA). However, ipsilateral activation in premotor (PM) area of left motor cortex was also observed. The time series of functional MRI signals from the voxel in labeled areas (see Figure 9). The time series corresponded with activation paradigm suggesting time-course on-off binary fMRI signal by simulated motor task due to neuronal or cognition activity. It suggested the association of motor cortex, somatosensory cortex and visual cortex with cerebellum through pontine nuclei during its motor activity and rCBF increases.
Figure 9

Time course of activation for the four stimulus epochs (B binocular, M monocular RL right eye stimulated first, followed by left eye stimulation, LR left eye stimulation first, followed by right eye stimulation)(left panel). During stimulation period, the subjects perceived a single flickering radial checkerboard, whereas during the rest period they viewed a single black fixation mark at constant mean luminance (right panel). Typical time courses from region of interest (ROI) centered in one hemisphere in the primary visual cortex (VI) or in an extrastriate region. Statistical parametric maps of significant BOLD responses to alternating monocular stimulation compared to the binocular condition (right panel). Voxels in bright regions indicate strong response to alternating monocular stimulus. The cross hairs represent the most active voxel within the cluster used for normalization (bottom panels).

Figure 10

Selected brain regions showing the activation areas observed by fMRI during finger movement. Regions with Z-score higher than threshold of 3.5 are displayed in red color. Stroke occurred in area colored as yellow. The time-course of on-off fMRI signal recorded in a typical voxel responding activation due to the stimulus paradigm (shown as red bold line).

3D motion paradigm subtractive approach

It generates activation fMRI maps significant for evaluation of symmetry of activation in the frontal lobes. The cerebrum cortex is not motion physiology sensitive (see Figure 11) but cerebellum cortex is motion physiology sensitive so pixel intensity changes represent its true activation maps as shown in Figure 12. However, paradigm independent structures with high susceptibility effects, partial volume effect become apparent on simultaneously FSE and EPI images as shown in Figure 13. Common examples are air sinuses, air spaces, and ferromagnetic substrates. It is the reason coronal images are not acquired for fMRI imaging but axial images show specific task-related activation areas.
Figure 11

The figure shows activation in the right intrapareital sulcus of cerebrum cortex lobes (as arrow). The raw image fMRI image did not show ghost or susceptibility effect to cause fictitious activation. After segmentation and processing, corresponding time series of activated pixels showed intensity changes. These intensity changes correlated with boxcar type paradigm. The frontal lobes showed fictitous activation while right intraparietal area showed valid activation map.

Figure 12

The figure shows activation in the left cerebellum cortex lobes (as arrow). The raw image fMRI image did not show ghost or susceptibility effect to cause fictitious activation. After segmentation and processing, corresponding time series of activated pixels showed intensity changes. These intensity changes correlated with boxcar type paradigm. The left area showed fictitious activation map due to eye movement.

Figure 13

The figure shows strong susceptibility effects in axial slice. Note the effect of partial presence of air sinus caused significant distortion in EPI image (arrow in left panel). In coronal slice, air sinus occupied larger image volume (arrow in right panel).

Image processing principles

We mainly describe two segmentation and registration methods for fMRI image processing and their applications.


The segmentation task in fMRI is performed by recognition and distinguishing brain areas that respond to a given task or stimulus with high specificity and sensitivity. Two methods 'Bayesian approach' and 'General Linear Model' are common for computation of statistical parameter maps (SPM). The detection of brain activation due to an input stimulus is segmented by statistically comparing images acquired during stimulation (ON state) and those acquired when brain is at rest (OFF state). The results of comparison are expressed by test statistics for each brain voxel in terms of 'likelihood' or 'significant activation' of voxel by the stimulus. Whole brain voxels' likelihood generates SPM map. SPM is an image in which image intensity values represent statistics obtained under null hypothesis of no activation and conform to a certain probability distribution.

'Thresholding' technique using SPM at a significant value can detect brain activation and spatial correlation using 'Gaussian random fields' (GRF) for multivariate Gaussian distribution. This GRF performs the spatial filtering of functional images to minimize pseudo-active brain regions. Alternatively, binary 'Markov random fields' (MRF) models for activation patterns suggested the intensity distribution of SPMs and Bayesian modeling of fMRI time-series inferred the hidden psychological states in fMRI experiments using 'likelihoods' of activation probabilities from these SPMs as shown in Figure 14.
Figure 14

Activation obtained on two axial brain slices of a representative volunteer in the memory retrieval task by (a) thresholding the SPM{z} at a significant P value = 0.01, (b) using the SPM approach on the SPM{z} with a minimum blob size of three voxels and a significance threshold z = 3.5, and (c) using the MRF approach on the SPM{z}. The significance values (z-values) of the activated voxels are shown color-coded. Reproduced with permission from reference [38].

Bayesian approach

A functional brain image is a spatio-temporal signal from brain serial scans taken over time. The posterior probability (Gaussian conditional covariance ηθ/y, where (p(θ/y) is proportional to the obtained data depending on times of prior probability of θ as: p(θ/y) α p(y/θ) p(θ). The Guass-Markov estimator may be presented as:

ηθ/y = (XT Cε -1X)-1(XTCε - 1y)    (6)

First, preprocessing of images is done and then detection of brain activation analysis is performed [38]. For it, the set of brain voxels is identified from image domain, and the brain scans are corrected for baseline intensity variation and person's head motion. In next step, derivation of SPMs and their statistical analysis by GLM, detects regions of significant activation.

F contrasts

Event-related conditions for motor responses are interpreted as hemodynamic response function (HRF) to generate SPM 't' maps as shown in Figure 8. Using design matrix Xo for 'right motor response' regressors look for variance of residuals. The 'F' test computes the sum of squares of "right hand regressors" as following:

with ν1 = tr[(Ro - R)Σi] and ν2 = tr(R Σi); df 1 = tr [Ro - R) Σi (Ro - R) Σi /tr(Ro - R) Σi]2 and df 2 = tr [R Σi R Σi /tr(R Σi]2

Where Ro is projector onto residual space of xo, and Px is orthogonal projector onto X. The 'F' contrasts are one-dimensional, in which case 'F' statistics is simply the square of the corresponding 't' statistics. In SPM interface, 'F' contrasts are displayed as images and 't' statistics is displayed as bars [37, 38].

General linear model

Consider an fMRI experiment involving multiple-input stimuli. Let y(t) and xo(t) denote the values of the fMRI time-series reponse and the input stimulus 'o' at time t, respectively. Let Xo = (Xo(t); t θ)T and the design matrix of experiment by [X1X2....Xn xn+1....xn+m] where X1, X2....Xn represent n stimulus covariates and Xn+m represent 'm' dummy covariates such as age, gender etc. If y = (y(t): t θ)T represents the fMRI time-series, the GLM can be written as y = Xβ + η, β = (β1, β2, ...βn+m)T denotes the regression coefficients relating the input covariates to the fMRI response, the matrix X = [H1X1 H2X2 ....HnXn Xn+1.....Xn+m] represent the design matrix having covariates modified with the modulation matrices Ho = {hk ij}n.m and the components of noise factor η will correlate and distribute normally. The multiplication of 'input' stimulus with 'modulation matrix' both auto-correlate the dispersion in fMRI response. The 'F' statistics estimates the significance of stimulus to generate time-series 'y' and least square estimate of regression coefficients 'β'. Using time-series at voxel site and stimulus condition Xo, the Fo (statistical score) = {Fo(p): p θ} represents 'F' statistical maps for stimulus 'o' and denoted by SPM [Fo]. SPM obtained using one statistics can be converted to another statistics using their grand mean scaling, regressors by parametric modulation, high-resolution basic functions and serial covariance matrix to get cumulative distributions in each voxel. The applications of these smooth and filtered SPM intensity profiles indicate spatial extents of the activated blobs [38].

Markov Random Field (MRF) model

This approach presumes that brain activation patterns form MRF to incorporate contextual information. Let us assume that set ao = {ao(p):p θ} denote a segmentation of an SPM or a configuration of brain activation, where ao(p) denotes the state of the brain voxel at site p and ao(p) = 0 if the voxel is inactive and ao(p) = 1 if the voxel is activated by the stimulus 'o'. Here ao represents MRF or activation pattern [38]. As the brain voxel is either activated or inactive, the MRF is assumed as binary logistic model. For this contextual information, preprocessed images find height threshold for smooth statistical map to detect and distinguish activated areas by using 'Euler Characteristics', Benferroni Correction and contextual clustering algorithm [37, 38]. The Euler characteristics E [EC] is:

E[EC] = R(4 loge 2)(2π)-3/2 Zte-1/2Zt 2    (8)

Where Z-score thresholds between 0–5, R is number of resels. The later is based on the fact that SPM voxel is adjusted with neighborhood information, if differs from expected non-activation value more than a specified decision value. The 'contextual clustering algorithm' estimates cluster parameter, calculates probability distribution and estimates modulation function to classify the voxel as the 'activation' class, otherwise 'nonactivation' class. However, three algorithms viz. 'voxel-wise thresholding', 'cluster-size thresholding' and 'contextual-clustering' have been described earlier [38]. Contextual clustering detects activations in small areas with high probability and voxel-wise specificity. 'Benferroni Correction' is based on probability rules and used for calculating family-wise error (FEW) rates Pfwe for fMRI as Pfwe = 1 - (1 - α)n where α is single-voxel probability threshold.

Computation of a statistical parametric map

For the purpose of this segmentation, SPM in the voxel i is represented as:

SPM{Fx}={Fk(p): p ΩB}    (9)

Where F statistical map of Fk(p) for stimulus k represents F statistical score and indicates significance of predicting time-series of voxel site p. For image data, first spatial low-pass filtration increases signal-to-noise ratio and sensitivity then SPM{Fx} is computed [38].

Applied segmentation methods

Voxel-wise thresholding (VWTH) segmentation method of an SPM applies thresholding to each voxel separately. The voxel at location I is considered as active if and only if zi < T where T is threshold. Cluster-size thresholding (CSTH) explains the cluster 'c' as active if and only if for all voxels within the cluster zi < T and the size of cluster 'c' is at least 'T' size voxels. Other common use 'contextual clustering algorithm' (CC) was described in steps [37, 38] as:
  1. 1)

    Label the voxels with zi <T as active and other voxels as non-active. Voxels outside the image volume are considered as non-active.

  2. 2)

    Compute for all voxels I the number of active neighbor voxels ui.

  3. 3)

    Relabel the voxels for which

    Zi + β/T (ui - N/2) < T    (10)

    Equation 10 represents voxels as active and other voxels as non-active. The number of neighbor voxels equals to 26-connectivity N = 26. The parameter determines the weighing of the contextual information and is usually positive.

  4. 4)

    If the current labeling is same as the labeling in the previous cycle before that, then stop iterations, otherwise return to step 2.


Probability of observing false activation voxels in a whole volume compares the sensitivity of methods by algorithm applied to different data parameter values. The decision parameter value is chosen that gives false activation in approximately 5 % images and 5 % measured false activation voxels.

Sensitivity, segmentation accuracy and robustness

Generally averaged 500 SPMs achieved by different segmentation methods give mean image probability at voxel-level. The less number of false classified voxels as 'active' in the neighborhood of activation represents segmentation accuracy. Noise evaluation by segmentation of different data determines the robustness against spatial autocorrelations. Low robustness is high probability of false activation detection more than the expected probability.



Image registration estimates the mapping between a pair of images. Registration performs for each 3D volume with display of movement parameters in continually updating graph to get matching criteria. Combination of 2D shearing operations and Fourier transform based shifting generate accurate high-speed 2D MR image rotation based on factorization of a general 2D planar rotation matrix. 3D arbitrary orthogonal matrix can be factored in to 3D rotations to accomplish 3D image rotation from nine 2D shears. Other approach of linear-in-frequency phase shift in frequency domain applied to 1D fast Fourier transforms (FFTs) generated the image rotation with polynomial interpolation methods [3941].

However, 3D real time image registration (rotation) algorithm chose the axes ordering that resulted in the least intermediate image distortion (minimum net rotation) at proper flip angle about x, y or z-axes i.e. generalized and windowed sinc interpolation. It applied real-time functional MRI acquisition and activation analysis modules within AFNI package. Functional MRI requires the rigid body transformations: small rotations, translations, zooms, rotating tensors and shears in 1–2 degrees or 1–2 voxel dimensions [39, 41]. So, repeated linearization of weighted least squares penalty functions with respect to motion parameters accomplishes the registration of a base image to a target image. This method minimized the regional influences and intrinsic variability in functionally active voxels in the brain. However, fMRI registration suffers from motion-related artifacts: interpolation errors, spin excitation in slice, spatial distortion by Gy and Nyquist ghosts. Intensity based intermodal registration AIR use variance of intensity ratios (VIR) cost function. Real-time image reconstruction was reported using Vision 3.5 software in communication with AFNI or TCP/IP sockets for intra- or intercomputer communications. These registration and rotation algorithms are available as AFNI registration and Visualization program [39, 42].

Post-processing methods for fMRI images

Several post-processing programs 'BrainVoyager', 'AFNI', 'LOFA', 'AIR' are available as a highly optimized and user-friendly software systems for the analysis and visualization of functional magnetic resonance imaging data [4346]. These combine surface-based and volume-based tools to study the structure and function of the brain to explore the secrets of the active brain by fast and highly optimized 2D and 3D image analysis and visualization routines, as shown in Figure 15. These are built-in-support for major standard and advanced data formats.
Figure 15

BrainVoyager is a highly optimized and user-friendly software system for the analysis and visualization of functional and anatomical magnetic resonance imaging data. It combines surface-based and volume-based tools to study the structure and function of the primate brain. Reproduced with permission from reference [44].

In general, post-processing is completed in following steps:

Volume-based statistical analysis

Methods include conjunction and Random Effects Analysis (RFX) for single and group analysis via Summary Statistics as following:
  1. 1.

    Fit the model for each subject using different GLMs for each subject or by using a multiple subject GLM.

  2. 2.

    Define the effect of interest for each subject with the contrast factor. Each produces a contrast image containing the contrast of the parameter estimates at each voxel.

  3. 3.

    Feed the contrast images into a GLM that implements a one-sample t test.


The RFX analysis is good technique for making inference from representative subjects [47]. In fMRI, block analysis, event-related easy selection of regions-of-interest, display of time courses, integration of volume and surface rendering are powerful tools for creation of high-quality figures and movies.

Advanced methods for automatic brain image-processing

The post-processing offers a comprehensive set of analysis and visualization tools that start its operation on raw data (2D structural and functional matrices) and produces visualization of the obtained results. Now a day, all advanced software features are available via a 'intuitive Windows interface'. Several approaches were performed for surface reconstruction, cortex inflation and flattening; cortex-based statistical data analysis (cbGLM) and inter-subject alignment based on gyral / sulcal pattern; cortexbased Independent Component Analysis (cbICA); creation and visualization of EEG / MEG multiple dipole models (fMRI "seeding"); multi-processor support, for ultimate performance; open architecture via COM interface, including scripting and automation [4851].

Data analysis

It includes data analysis (motion correction, Gaussian spatial and temporal data smoothing, and linear-trend removal, filtering in the frequency domain), correlation analysis, and determination of Talairach coordinates, volume rendering, surface rendering and cortex flattening [47]. Statistical maps may be computed either in the 2D or 3D representation since structural as well as functional 4D data (space × time) is transformed into Talairach space. Talairach transformation is performed in two steps. The first step consists of rotating the 3D data set for each subject to be aligned with the stereotaxic axes. For this step, the locations of the anterior commissure (AC) and the posterior commissure (PC) as well as two rotation parameters for midsagittal alignment have to be specified interactively. In the second step, the extreme points of the cerebrum are specified. These points together with the AC and PC coordinates are then used to scale the 3D data sets into the dimensions of the standard brain of the Talairach and Tournaux atlas [47]. Isolating the brain gray matter and white matter tissues using region-growing methods, filter operations and the application of 3D templates performs segmentation. Segmentation explores a 3D volume with superimposed pseudocolor-coded statistical maps in a four-window representation showing a sagittal, coronal, transversal and oblique section. Based on a (segmented) 3D data set, a 3D reconstruction of the subjects' head and brain can be calculated and displayed from any specified viewpoint using volume or surface rendering. Parametric and non-parametric statistical maps may be computed and superimposed both on the original functional scans as well as onto T1-weighted 2D or 3D anatomical reference scans. Nonparametric-permutatation approaches are alternate options at low degree of freedom (small sample size to determine intersubject variability) for noisy statistic images when random fields are conservative i.e. smooth variances [47].

Volume rendering

It is performed with a fast 'Ray-Casting algorithm'. Lightning calculations are based on 'Phong-shading'. Surface rendering of reconstructed surfaces was performed using OpenGL [48].

The surface reconstruction

The surface reconstruction starts with a sphere (recursively tessellated icosahedron) or a rectangle, which slowly wraps around a (segmented) volume data set. Blood oxygenation level-dependent (BOLD)-based fMRI was performed in the visual cortex, and the foci of fMRI activation utilized as seeding points for 3D fiber reconstruction algorithms, thus provided the map of the axonal circuitry underlying visual information processing [49]. A reconstructed cortical surface may be inflated; cut interactively and slowly unfolded minimizing area distortions. Statistical 3D maps may be superimposed on reconstructed, inflated or flattened cortex. Signal time courses may be invoked by simply pointing to any region of a visualized surface.

Future developments and advances

Recent advances in fMRI research in visual and motor events response are reported mainly to identify localized cortical regions by robust image processing segmentation and registration methods, statistical analysis and better spatial resolution using multimodal approaches as reviewed in following section.

High-field MR scanner system is an advantage in fMRI

For high-field fMRI imaging, paramagnetic susceptibility of spin may be related with gyromagnetic ratio (γ) and represented by the Brillouin equation as:

where k is the Boltzmann constant and T absolute temperature.

Susceptibility effect in fMRI increases exponentially as the hyperbolic tangent associated with an increase in the main field, Bo, of the system. T2* detectable activation (ΔI) is significantly increased as shown in Figure 16. Simultaneously, artifacts inducing perturbations also increase. High-field MRI imager system generates T2* contrast for analysis of complex behavioral tasks as shown in Figure 17. Single subject Ideographic analysis was reported at 3 T systems to locate face-exemplar by regional cortical flat-mapping [50]. For clinical purposes, 1.5 T systems work well. For advanced neuroimaging investigation, higher field MR systems are essential.
Figure 16

A typical time series of an activated pixel in primary cortex is shown to represent the performance of horizontal 3 T system optimized for fMRI. S/N indicated variation of EPI images, while _I, activation induced increase in signal intensity. This time series represents signals from a single voxel volume of 3 mm × 3 mm × 5 mm. The red curves represent raw data and boxcar type model functions shown in blue color.

Figure 17

Figure represents functional maps shown for comprehension tasks: for reading task (eft panel) and hearing task (right panel). Paradigms requiring tasks of different modalities (visual vs auditory) may provide almost identical activation maps based on the identical abstract concept of "comprehension". Using high field fMRI may provide high quality activation maps to distinguish these activation maps.

fMRI Time series analysis

In fMRI, 'time series analysis' by SPM99 is recently used for autocorrection and smoothening. For it, generalized linear model can be expressed as a function of discrete time series, y(t) as:

Y(t) = xc(t)βc + ε(t)    (12)

where xc(t) and ε(t) are function of time, βc are time-invariant parameters. Linear time invariance distinguishes neural activity (event) and post-stimulation (epoch) onsets. In general, the resolution of delta function, dt = Tr/T sec and the number of columns = Nc = NiNjNkNb represent invariance in design matrix. High pass filtering of 'time series' frequency components y(t) get Fourier transformed to remove noise and convolution. Temporal autocorrection in fMRI series is done by 'temporal smoothing' and 'intrinsic autocorrelation' and estimated by 'Auto-Regression' or '1/ f low-pass smoothing' methods to remove bias [51].

SPM99 offers 'finite impulse response' (FIR) sets for increased neural activity increases BOLD response 'amplitude' over few seconds based on BOLD from different brain regions such as V1, S1, A1 and higher cortical regions with different vasculature 'Temporal Basis Functions'. FIR sets consist of Nk contiguous box-car functions of peristimulus time, each of duration TH/Nk. TH is maximum duration of high-pass filter. The Fourier set consists of sine Ns and cosine functions of harmonic periods TH, TH/2...TH/Ns. Linear combination of FIR, Fourier sets captures any shape of response in timescale TH/Nk or Ns/TH respectively [52]. 'Event-related Response' and 'Basis sets' were chosen based on stimulus variability and canonical 'Hemodynamic Response Function' and 'F' contrasts determine contribution of different basis sets. Single event- Multi-event type design minimizes the 'contrast error'. Deterministic, static and dynamic stochastic designs use minimum 'stimulus onset synchrony' (SOAm) and probability of event (for single event design) or transition matrix (for multi-event design) to induce variance over a range of frequencies.

Linear Transform model interprets fMRI signals, origin of the fMRI signals and compares fMRI with neuronal signals. Left and right lateralization for motor cortex stimulation generated visual flash motor response indicative of relationship between different measures of neuronal activity such as single-and multi-unit spiking activity, LFP etc. and reflected neuronal functions. Recently, fMRI signal measured the signal induced by the inputs to a cortical area [53].

Independent component-cross correlation-sequential epoch (ICS) Analysis: Image Processing

The fMRI acquisition time is usually less per paradigm. For multiple task-oriented studies, analysis of complex higher brain functions is based on the 'principle of functional independence' and functional distinct areas (chronoarchitecture). Independent component-cross correlation-sequential epoch (ICS) segregates distinct areas in cerebral and temporal chronoarchitectonic maps. The different exposures of the brain to natural conditions for different 'durations' segregate the different brain areas for their temporal differences. One subset of natural conditions, reflect free-viewing activity of visual, parietal, temporal areas. However, frontal, pre-frontal cortices functional subdivisions or multivariate paradigms were recently illustrated as shown in Figure 18. A sequential epoch paradigm is composed to embed the function which correlates to the target multiple behavioral hypothesis i f i (t)). The simplest function is a 'boxcar function' as employed in many fMRI studies. For it, fMRI time series are subjected to blind separation into independent components by independent component analysis. Subsequently, cross correlation analysis is performed utilizing each embedded function, f(t), as model function to achieve multiple fMRI images as behavioral correlates given by the selected function as an activation map. For the hemodynamic reference function (HRF) following a single sensory stimulation, the time course function represents as model function and ICS becomes a reliable method for event-related fMRI. ICS is useful for event related high-field fMRI where T2* contrast enhances the magnitude of activation than that performed on conventional 1.5 T clinical systems [54].
Figure 18

An example of multivariate analysis in primary motor cortex is represented for revealing the presence of dual representations of constructed multivariate paradigms (MI 4a and MI 4p) in human.

Brain functional areas

Recent trend in fMRI research was to understand the relationship of physiological mechanisms and selective activation of different brain locations using fMRI techniques. However, the knowledge of the independent brain functions and control by different parts is still in infancy. fMRI has long way to answer the physiological stimuli and mechanism of different fMRI BOLD signals. The success of it solely depends on power of fMRI image processing. Recent investigations highlight the fMRI visible different brain areas, new understanding of fMRI sensitive physiological stimuli and use of high field scanners. Visual stimuli and event related neural activity by fMRI and image processing methods are reviewed in recent years [18]. Visual areas are recently identified as the boundaries of visual areas V1, V2, V3, V3A, V4, MT/V5, and TEO/V4A in visual cortex and their distribution within the occipital lobe. Motor related areas are known as M1. These corresponding areas are: Broadmann's area (BA 4), SMA (BA 6) and premotor area (BA 6). These fMRI visible areas as shown in Figure 18, are important to interpret fMRI stimulation and its location [18, 55].

Recently, new understanding of quantitative visual field eccentricity function measurements on visual field maps by fMRI were made in macaque visual cortex visual areas. fMRI estimated the average receptive field sizes of neurons in each of several striate and extrastriate visual areas of the human cerebral cortex. Retinotopic mapping procedures determined the boundaries of the visual areas and visualized on flattened occipital cortex, primary visual cortex V1, V2, V3/VP and V3A and V4. In all these areas, receptive fields increased in size with increasing stimulus eccentricity similar to macaque monkeys [56]. fMRI maps and the visual area maps represent the distribution of cortical signals and computational homologies between human and monkey. Neural activity and the creation of a new memory trace were observed using functional magnetic resonance imaging (fMRI). Event-related fMRI demonstrated the activity in prefrontal and medial temporal lobe areas associated with successful memory storage. Contrast activity was associated with encoding success and encoding effort using a cue in the form of a letter (R or F). These recent studies suggested the prefrontal activation strongly associated with intentional verbal encoding and left medial temporal activation for successful memory on the subsequent test. Cortical regions sensitive to motion processing receive their inputs only via the primary visual cortex (striate cortex).

Recently, fMRI evidenced higher-order motion-processing in primates and humans with damaged primary visual cortex (e.g., "blindsight" for motion in the blind visual hemifield) for the existence of a direct thalamic functional pathway exists to extrastriate visual cortical motion processing areas that bypasses primary visual cortex [56]. Highfield fMRI retinotopic method was reported to map the neural substrate of retinal slip compensation during visual jitter in flattened cortical format. A novel illusion (visual jitter) suggested the compensation mechanism based on retinal motion. fMRI suggested the pathway from V1 to MT+ involved in the compensation stage in stabilizing the visual world [57]. fMRI demonstrated the sensitivity changes controlled within the visual pathway for responses in human visual area V1 to a constant-amplitude, contrastreversing probe presented on a range of mean backgrounds. fMRI signals from probes initiated in the L and M or S cones. Psychophysical tests showed changes in V1 fMRI cortical BOLD signals by 'mean-field adaptation model' within cone photoreceptor classes [58]. A new mechanism of hypercapnia and hypocapnia was described as alveolar oxygen and CO2 gases flux and their effect on BOLD response to visual stimulation. At high magnetic field 7 T, the BOLD signal magnitude and dynamics of hemodynamic response represented the effect of CBF under conditions: hypocapnia, normocapnia, and hypercapnia [59].

Binocular interactions present checkerboard stimuli occurring when subjects view dichoptically. A flickering radial checkerboard stimulation of eyes in binocular or monocular conditions, generate specific responses in striate and extrastriate visual cortex on T2*-weighted images of visual cortex acquired with gradient-echo, echoplanar imaging. The striate area, calcarine fissure BOLD response differed for these stimulation conditions [60]. Recently, a neuron location by color-selective mapping method has attracted to compare the relationships of ocular dominance and orientation with responses to high-contrast luminance stimulus and patchy distribution of color selectivity to locate different functional subdivisions of striate cortex in macaque. These color patches with the cytochrome-oxidase (CO) blobs speculated the ocular dominance (OD) column. For it, "Ice cube" model of color-selective regions predicted the organization of orientation and ocular dominance functional hypercolumns in V1 [61]. Dipole locations in cortical brain (regional visualization) is developed as a new art by fMRI activations. Neural generators of the visual evoked potential (VEP) generate isoluminant checkerboard stimuli. Using Multichannel scalp recordings, retinotopic mapping and dipole modeling techniques estimated the dipole locations of the cortical regions giving rise to C1, P1, and N1 components of VEP [62]. These locations could be matched to both MRI-visible anatomical brain regions and fMRI activations. Several locations are broadly identified as C1 component (striate cortex; area 17), early phase of the P1 component (dorsal extrastriate cortex of the middle occipital gyrus), late phase of the P1 component (ventral extrastriate cortex of the fusiform gyrus), posterior N 150, anterior N 155 (parietal lobe) in relation to visual-perceptual processes. In other development for complex cognitive tasks, neuronal encoding and fMRI processing strategies segregate retention and retrieval phases of visual short-term memory for objects, places and conjunctions in humans. These tasks were associated with spatio-temporal activation of parietal and prefrontal areas during the retention phase and posterior-anterior and right-left dissociation for spatial versus non-spatial memory [63].

The 'perceptual switch' stimulus induces responses in areas calcarine to parieto-occipital and ventral and lateral temporo-occipital cortex to anterior insula. During vection, early motion-sensitive visual areas and vestibular parieto-insular cortex deactivate, whereas higher-order parieto- and temporo-occipital areas respond to optical flow retained identical activity levels. Recent fMRI study showed that these areas displayed transient activations as response to the type of visual motion stimulus and perceptual biostability [64]. fMRI distinguished different neural substrates as 'visual object recognition' sites i.e. lateral occipital and posterior inferior temporal cortex with lower activity for repetitions of both real and non-sense objects; fusiform and left inferior frontal regions with lower activity for repetitions of only real 3D objects; left inferior frontal cortex for different exemplars evidencing dissociable subsystems in ventral visual cortex with distinct view-dependent and view-invariant object representations. Repetition-priming method was proposed for visual stimuli recurring at unpredictable intervals, either with the same appearance or with changes in size, viewpoint or exemplar [65].

Event related potentials and fMRI activation patterns

Combining event-related potentials (ERP) and fMRI activation provide temporal and spatial resolution, functional connectivity of neural processes of same neural networks within the bilateral occipital gyrus, lingual gyrus; precuneus and middle frontal gyrus; and the left inferior and superior parietal lobe; middle and superior temporal gyrus; cingulate gyrus, superior frontal gyrus and precentral gyrus. It evidenced the correlation within the common activity and time-range in a complex visual language task [66]. These tasks comprise specific stimulus-response associations and activate a variety of non-specific cortical regions [67]. Dystonia, a movement disorder involves involuntary coordination of agonist and antagonist muscles, which cause abnormal posture or twisting. Event related fMRI technique revealed impairment of muscle contraction and relaxation. Comparison of activated volume in cortical motor areas in dystonia patients with volunteers showed different muscle relaxation and contraction activation volumes as shown in Figure 20. In these tasks, mainly SM1 and SMA activated areas were reduced contra- laterally in dystonia patients as evidenced by time course of fMRI signal in SMA activation area [68].
Figure 19

An illustration of anatomical connections between motor areas are shown. Interconnections between multiple motor-related cortices are represented with numbers in parentheses. These are: Brodmann's area M1: Primary motor cortex, SMA: Supplementary motor area, dPMA: Dorsal premotor area, vPMA: Ventral premotor area, CMAr: Rostral cingulated motor area, CMAc: Caudal cingulated motor area, PFC, Prefrontal cortex, SPL: Superior parietal lobe, IPL: Inferior parietal lobe.

Figure 20

Figure represents the application of event-related fMRI to dystonia. Comparison of activated volume in motor cortical areas in a patient with dystonia and a normal subject is represented in the muscle relaxation and contraction task. In both tasks, the activated areas in the M1 and SMA were smaller for dystonia while in normal these activated areas were larger (see top 4 panels shown as A). A solid line indicates a mean signal change across ten trials shown as dots. The transient signal change timelocked to EMG offset was observable even in single trial (see panel B). Group data from eight patients and twelve healthy volunteers; demonstrate that activated volumes in the contralateral SM1 and SMA are greater in the healthy volunteers than in the dystonic patients. Figure is reproduced with permission of Oxford University Press from the reference [68].

Sensory and motor systems

Sensory and motor systems interact in complex ways. Voluntary movements with visual attention yield distinct fMRI hemodynamic signals and brain activations i.e. making repetitive finger movements, attending to the color of a visual stimulus or simultaneous finger movement and visual attention. In these processes, the primary motor cortex, supplementary motor area, cerebellum, sparse cerebral cortical and substantial bilateral cerebellar locations get active. Activation-related interactions in the left superior parietal lobule, the right fusiform gyrus, and left insula, recently were indicated their role in visual attention and movement [69].

Different psychological tests have been developed to evaluate frontal tasks of macaque monkeys and humans. Wisconsin Card Sorting Test (WCST) characterized the frontal lobe lesions in macaque monkeys and humans based on behavioral flexibility in the form of cognitive set shifting. Equivalent visual stimuli and task sequence showed transient activation related to cognitive set shifting in focal regions of prefrontal cortex in both monkeys and humans. These functional homologs were located in cytoarchitectonically equivalent regions in the posterior part of ventrolateral prefrontal cortex. This comparative imaging provided insights into the evolution of cognition in primates [70].

Test-retest precision of functional magnetic resonance imaging (fMRI) by student't' mapping (STM) is recently described for independent component analysis (ICA) using two or three iterations of visual and auditory stimuli for fMRI scans. Concurrence ratios of the activated voxels divided by the average number of voxels activated in each repetition showed similar test-retest precision of ICA as STM [71].

High spatial resolution fMRI

High spatial resolution in fMRI showed as dependent on hyperoxic hemodynamic response to neural activity in short duration and it was used to investigate the columnar architecture of ocular dominance within the primary visual cortex [35, 72]. For intensity-based non-rigid registration of medical images was developed for atlas based segmentation and intensity-based geometric correction of functional magnetic resonance imaging (fMRI) images by 'Adaptive bases algorithm' to register the smallest structures in the image [73].

Summary and future directions

The goal of this review was to introduce the neurophysiological factors and image processing principles of fMRI to suggest potential future applications in neuroscience and physiology. These future directions include neurosurgical planning and improved assessment of risk for individual patients, improved assessment and strategies for the treatment of chronic pain, improved seizure localization, and improved understanding of the physiology of neurological disorders. We look ahead to newer algorithms, enhanced fMRI sensitivity and spatial resolution by use of high field systems, ASL and phase array coils or newer contrast agents [74]. Presently, other emerging applications of EEG/MEG, PET and neuropsychological testing along with fMRI are coming up as the benefits of this fMRI technology incorporated into current neuroscience and future patient care. In recent years, lot of future excitement is evident in the following areas of brain information by segmentation and registration methods applied to fMRI. These include mainly automated nonlinear labeling; and automated surface reconstructions. Automated surface reconstruction appears to be possible by: i. cortical surface-based analysis by segmentation and surface reconstruction [75]; ii. cortical surface-based analysis by inflation, flattening, and a surface-based coordinate system [76]. Automated anatomical brain labeling may be performed by: i. whole brain segmentation: automated labeling of neuro-anatomical structures in the human brain [77]; ii. multipatient registration of brain fMRI using intensity and geometric features [78]; iii. automatic detection and labeling of the human cortical fields in magnetic resonance data sets [79].



Authors’ Affiliations

Departments of Medicine and Radiology, Columbia University
Department of Electrical Engineering, Indian Institute of Technology
Department of Electrical Engineering, Jai Narain Vyas University


  1. Tegeler C, Strother SC, Anderson JR, Kin SG: Reproducibility of BOLD based functional MRI obtained at 4 T. Hum Brain Mapp 1999, 7: 267–283. 10.1002/(SICI)1097-0193(1999)7:4<267::AID-HBM5>3.0.CO;2-3Google Scholar
  2. Lee SP, Silva AC, Ugurbil K, Kim SG: Diffusion-weighted Spin-echo fMRI at 9.4T: Microvascular/tissue Contribution to BOLD Signal Changes. Magn Reson Med 1999, 42: 919–928. 10.1002/(SICI)1522-2594(199911)42:5<919::AID-MRM12>3.3.CO;2-#Google Scholar
  3. Singh M, Kim S, Kim TS: Correlation between BOLD-fMRI and EEG signal changes in response to visual stimulus frequency in humans. Magn Reson Med 2003,49(1):108–14. 10.1002/mrm.10335MathSciNetGoogle Scholar
  4. Bandettini PA, Ungerleider LG: From neuron to BOLD: new connections. Nature Neurosci 2001, 4: 864–866. 10.1038/nn0901-864Google Scholar
  5. Zaini MR, Strother SC, Anderson JR, Liow JS, Kjems U, Tegeler C, Kim SG: Comparison of matched BOLD and FAIR 4.0T-fMRI with [ 15 O] water PET brain Volumes". MedicalPhysics 1999, 26: 1559–1567. 10.1118/1.598652Google Scholar
  6. Reber PJ, Siwiec RM, Gitleman DR, Parrish TB, Mesulam MM, Paller KA: Neural correlates of successful encoding identified using functional magnetic resonance imaging. J Neurosci 2002,22(21):9541–8.Google Scholar
  7. Preibisch C, Haase A: Functional MR imaging of the human brain using FLASH: influence of various imaging parameters. J Magn Reson 1999,140(1):162–71. 10.1006/jmre.1999.1838Google Scholar
  8. Ogawa S, Menon RS, Tank DW, Kim SG, Merkle H, Ellermann JM, Ugurbil K: Functional Brain Mapping by blood oxygenation level-dependent contrast magnetic resonance imaging: A comparison of signal characteristics with a biophysical model. Biophys J 1993, 64: 803–812.Google Scholar
  9. Nakai T, Matsuo K, Kato C, Okada T, Moriya T, Isoda H, Takehara Y, Sakahara H: BOLD contrast on a 3T magnet: detectibility of the motor areas. J Comput Assit Tomogr 2001, 25: 436–445. 10.1097/00004728-200105000-00018Google Scholar
  10. Bandettini PA, Cox RW: Event related fMRI contrast when using constant interstimulus interval: Theory and experiment. Magn Reson Med 2000, 30: 161–173.Google Scholar
  11. Turner R, Jezzard P, Wen H, Kwong KK, Le Bihan D, Zeffiro T, Balaban RS: Functional mapping of the human visual cortex at 4 and 1.5 T using deoxygenation contrast EPI. Mag Reson Med 1993, 29: 277–279.Google Scholar
  12. Kwong KK, Chesler DA, Weisskoff RM, Donahue KM, Davis TL, Ostergaard L, Campbell TA, Rosen BR: MR perfusion studies with T1-weighted echo planar imaging. Mag Reson Med 1995, 34: 878–887.Google Scholar
  13. Russ MO, Cleff U, Lanfermann H, Schalnus R, Enzensberger W, Kleinschmidt A: Functional magnetic resonance imaging in acute unilateral optic neuritis. J Neuroimaging 2002,12(4):339–50.Google Scholar
  14. Miki A, Liu GT, Englander SA, Raz J, van Erp TG, Modestino EJ, Liu CJ, Haselgrove JC: Reproducibility of visual activation during checkerboard stimulation in functional magnetic resonance imaging at 4 Tesla. Jpn J Ophthalmol 2001,45(2):151–5. 10.1016/S0021-5155(00)00352-XGoogle Scholar
  15. Shibata K, Osawa M, Iwata M: Visual evoked potentials in cerebral white matter hyperintensity on MRI. Acta Neurol Scand 2000,102(4):230–5. 10.1034/j.1600-0404.2000.102004230.xGoogle Scholar
  16. Fransson P, Kruger G, Merboldt KD, Frahm J: A comparative FLASH and EPI study of repetitive and sustained visual activation. NMR Biomed 1997,10(4–5):204–7. 10.1002/(SICI)1099-1492(199706/08)10:4/5<204::AID-NBM452>3.0.CO;2-JGoogle Scholar
  17. Kim SG, Rostrup E, Larsson HB, Ogawa S, Paulson OB: Determination of relative CMRO2 from CBF and BOLD changes: Significant increase of oxygen consumption rate during visual stimulation. Magn Reson Med 1999, 41: 1152–1161. 10.1002/(SICI)1522-2594(199906)41:6<1152::AID-MRM11>3.0.CO;2-TGoogle Scholar
  18. Nakai T, Matsuo K, Kato C, Okada T, Moriya T, Isoda H, Takehara Y, Sakahara H: BOLD contrast on a 3T magnet: detectability of the motor areas. J Comput Assist Tomogra 2001, 25: 436–445. 10.1097/00004728-200105000-00018Google Scholar
  19. Toma K, Nakai T: Functional studies in human motor control studies and clinical applications. Mag Reson Med Sci 2002,1(2):109–120.Google Scholar
  20. Kim SG: Quantification of relative cerebral blood flow change by flowsensitive alternating inversion recovery (FAIR) technique: Application to functional mapping. Magn Reson Med 1995, 34: 293–301.Google Scholar
  21. Nakada T, Fujii Y, Kwee IL: Brain strategies for reading in the second language are determined by the first language. Neurosci Res 2001, 40: 351–35. 10.1016/S0168-0102(01)00247-4Google Scholar
  22. Logothetis NK, Pauls J, Augath M, Trinath T, Oeltermann A: A neurophysiological investigation of the basis of fMRI signal. Nature 2001, 412: 150–157. 10.1038/35084005Google Scholar
  23. Mayville JM, Bressler SL, Fuchs A, Kelso JA: Spatiotemporal reorganization of electrical activity in the human brain associated with a timing transition in rhythmic auditory-motor coordination. Exp Brain Res 1999, 127: 371–381. 10.1007/s002210050805Google Scholar
  24. Haslinger B, Erhard P, Kampfe N, Boecker H, Rummeny E, Schwaiger M, Conrad B, Ceballos-Baumann AO: Event related functional magnetic resonance imaging in Perkinson's disease before and after levodopa. Brain 2001, 124: 558–570. 10.1093/brain/124.3.558Google Scholar
  25. Kim DS, Duong TQ, Kim SG: High-resolution mapping of iso-orientation columns by fMRI. Nature Neurosci 2000, 3: 164–169. 10.1038/72109Google Scholar
  26. Ogawa S, Menon RS, Kim SG, Ugurbil K: On the characteristics of functional MRI of the brain. Ann Rev Biophy and Biomol Struct 1998, 27: 447–74. 10.1146/annurev.biophys.27.1.447Google Scholar
  27. Jueptner M, Weiller C: Does measurement of regional cerebral blood flow reflect synaptic activity: implication for PET and fMRI. Neuroimaging 1995, 2: 148–156. 10.1006/nimg.1995.1017Google Scholar
  28. Villringer A: Physiological changes during brain activation. In "Functional MRI" (Edited by: Moonen CTW, Bandenitti PA). Springer-verlag, Berlin 1999, 3–13.Google Scholar
  29. Buxton RB, Frank LR: A model for the coupling between cerebral blood flow and oxygen metabolism during neuronal stimulation. J Cereb Blood Flow Metab 1997, 17: 64–72. 10.1097/00004647-199701000-00009Google Scholar
  30. Mandeville JB, Marota JJ, Ayata C, Zaharchuk G, Moskowitz MA, Rosen BR, Weisskoff RM: Evidence of a cerebral post-arteriole windlissel with delayed compliance. J Cereb Blood Flow Metab 1999, 19: 679–689. 10.1097/00004647-199906000-00012Google Scholar
  31. Zaharchuk G, Ledden PJ, Kwong KK, Reese TG, Rosen BR, Wald LL: Multislice perfusion and perfusion territory imaging in humans with separate label with image coils. Magn Reson Med 1999, 44: 92–100.Google Scholar
  32. Disbrow EA, Slutsky DA, Roberts TP, Krubitzer LA: Functional MRI at 1.5 Tesla: A comparison of the blood oxygenation level-dependent signal and electrophysiology. Proc Natl Acad Sci USA 2000, 97: 9718–9723. 10.1073/pnas.170205497Google Scholar
  33. Cheng K, Waggoner RA, Tanaka K: Human ocular dominance columns as revealed by high field functional magnetic resonance imaging. Neuron 2001, 32: 359–374. 10.1016/S0896-6273(01)00477-9Google Scholar
  34. Cox RW: AFNI: software for analysis and visualization of functional magnetic resonance neuroimages. Comput Biomed Res 1996,29(3):162–73. 10.1006/cbmr.1996.0014Google Scholar
  35. Goodyear BG, Menon RS: Brief visual stimulation allows mapping of ocular dominance in visual cortex using fMRI. Human Brain Mapp 2001,14(4):210–217. 10.1002/hbm.1053Google Scholar
  36. Kim DS, Duong TQ, Kim SG: High resolution mapping of iso-orientation columns by fMRI. Nature Neurosci 2000, 3: 164–169. 10.1038/72109Google Scholar
  37. Kiebel S, Holmes A, Poline JB, Kherif F, penny W: The general Linear Model; Contrasts and classical inference. In: Human Brain Function. Chapters 37 and 38 (Edited by: Frackowiak RSJ, Friston KJ, Frith CD, Dolan RJ, Price CJ, Jeki S, Ashburner J, Penny W). Elsevier Academic Press, London 2004, 749–779.Google Scholar
  38. Rajapakse JC, Priyaratna J: Bayesian approach to segmentation of statistical parametric maps. IEEE Trans on Biomed Eng 2001,48(10):1186–1194. 10.1109/10.951522Google Scholar
  39. Cox RW, Jesmanowicz A: Real-time 3D image registration for functional MRI. Magn Reson Med 1999, 42: 1014–1018. 10.1002/(SICI)1522-2594(199912)42:6<1014::AID-MRM4>3.0.CO;2-FGoogle Scholar
  40. Sarkissian E, Bowman KW: Application of a nonuniform spectral resampling transform in Fourier-transform spectrometry. Appl Opt 2003,42(6):1122–31.Google Scholar
  41. Ciulla C, Deek FP: Performance assessment of an algorithm for the alignment of fMRI time series. Brain Topogr 2002,14(4):313–32. 10.1023/A:1015756812054Google Scholar
  42. Nichols T, Holmes A: Non parametric permutation tests for functional neuroimaging. In: Human Brain Function Chapters 46 (Edited by: Frackowiak RSJ, Friston KJ, Frith CD, Dolan RJ, Price CJ, Jeki S, Ashburner J, Penny W). Elsevier Academic Press, London 2004, 887–908.Google Scholar
  43. Gokcay D, Mohr CM, Crosson B, Leonard CM, Bobholz JA: LOFA: software for individualized localization of functional MRI activity. Neuroimage 1999,10(6):749–55. 10.1006/nimg.1999.0511Google Scholar
  44. Gold S, Christian B, Arndt S, Zeien G, Cizadlo T, Johnson DL, Flaum M, Andreasen NC: MRI statistical software packages: a comparative analysis. Hum Brain Mapp 1998,6(2):73–84. 10.1002/(SICI)1097-0193(1998)6:2<73::AID-HBM1>3.3.CO;2-1Google Scholar
  45. Vemuri BC, Ye J, Chen Y, Leonard CM: Image registration via level-set motion: applications to atlas-based segmentation. Med Image Anal 2003,7(1):1–20. 10.1016/S1361-8415(02)00063-4Google Scholar
  46. Friston KJ, Glaser DE, Henson RNA, Kiebel S, Phillips C, Ashburner : Classical and Bayesian inference in neuroimaging : applications. Neuroimage 2002, 16: 484–512. 10.1006/nimg.2002.1091Google Scholar
  47. Moutoussis K, Zeki S: The Chronoarchitecture of the Human Brain: Functional Anatomy Based on Natural Brain Dynamics and the Principle of Functional Independence. In: Human Brain Function. Chapter 13 (Edited by: Frackowiak RSJ, Friston KJ, Frith CD, Dolan RJ, Price CJ, Jeki S, Ashburner J, Penny W). Elsevier Academic Press, London 2004, 201–229.Google Scholar
  48. Hong L, Kaufman AE: Fast Projection-Based Ray-Casting Algorithm for Rendering Curvilinear Volumes. IEEE Trans Visual and Comp Graph 1999,5(4):322–332. 10.1109/2945.817349Google Scholar
  49. Kim DS, Kim M, Ronen I, Formisano E, Kim KH, Ugurbil K, Mori S, Goebel R: In vivo mapping of functional domains and axonal connectivity in cat visual cortex using magnetic resonance imaging. Magn Reson Imaging 2003,21(10):1131–40. 10.1016/j.mri.2003.08.014Google Scholar
  50. Schmitt F, Grosu D, Mohr C, Purdy D, Salem K, Scott KT, Stoeckel B: 3 Tesla MRI: successful results with higher field strengths. Radiologe 2004,44(1):31–47. 10.1007/s00117-003-1000-xGoogle Scholar
  51. Henson RN, Rugg MD, Friston KJ: The choice of basis functions in the eventrelated fMRI. Neuroimage 2001, 15: 83–97. 10.1006/nimg.2001.0940Google Scholar
  52. Rugg MD, Henson RN: Episodic memory retrival: an event-related functional neuroimaging perspective. In the Cognitive Neuroscience of Memory Encoding and Retrieval (Edited by: Parker AE, Wilding EL and Bussey T). Psychology Press, Hove 2002, 150–189.Google Scholar
  53. Meyer FG: Wavelet-based estimation of a semiparametric generalized linear model of fMRI time-series. IEEE Trans Med Imaging 2003,22(3):315–22. 10.1109/TMI.2003.809587Google Scholar
  54. Kiviniemi V, Kantola JH, Jauhiainen J, Tervonen O: Comparison of methods for detecting nondeterministic BOLD fluctuation in fMRI. Magn Reson Imaging 2004,22(2):197–203. 10.1016/j.mri.2003.09.007Google Scholar
  55. Brewer AA, Press WA, Logothetis NK, Wandell BA: Visual areas in macaque cortex measured using functional magnetic resonance imaging. J Neurosci 2002,22(23):10416–26.Google Scholar
  56. Schoenfeld MA, Heinze HJ, Woldorff MG: Unmasking motion-processing activity in human brain area V5/MT+ mediated by pathways that bypass primary visual cortex. Neuroimage 2002,17(2):769–79. 10.1016/S1053-8119(02)91204-8Google Scholar
  57. Sasaki Y, Murakami I, Cavanagh P, Tootell RH: Human brain activity during illusory visual jitter as revealed by functional magnetic resonance imaging. Neuron 2002,35(6):1147–56. 10.1016/S0896-6273(02)00899-1Google Scholar
  58. Wade AR, Wandell BA: Chromatic light adaptation measured using functional magnetic resonance imaging. Neurosci 2002,22(18):8148–57.Google Scholar
  59. Cohen ER, Ugurbil K, Kim SG: Effect of basal conditions on the magnitude and dynamics of the blood oxygenation level-dependent fMRI response. J Cereb Blood Flow Metab 2002,22(9):1042–53.Google Scholar
  60. Buchert M, Greenlee MW, Rutschmann RM, Kraemer FM, Luo F, Hennig J: Functional magnetic resonance imaging evidence for binocular interactions in human visual cortex. Exp Brain Res 2002,145(3):334–9. 10.1007/s00221-002-1121-xGoogle Scholar
  61. Landisman CE, Ts'o DY: Color processing in macaque striate cortex: relationships to ocular dominance, cytochrome oxidase, and orientation. J Neurophysiol 2002,87(6):3126–37.Google Scholar
  62. Di Russo F, Martinez A, Sereno MI, Pitzalis S, Hillyard SA: Cortical sources of the early components of the visual evoked potential. Hum Brain Mapp 2002,15(2):95–111. 10.1002/hbm.10010Google Scholar
  63. Munk MH, Linden DE, Muckli L, Lanfermann H, Zanella FE, Singer W, Goebel R: Distributed cortical systems in visual short-term memory revealed byevent-related functional magnetic resonance imaging. Cereb Cortex 2002,12(8):866–76. 10.1093/cercor/12.8.866Google Scholar
  64. Kleinschmidt A, Thilo KV, Buchel C: Neural correlates of visual-motion perception as object- or self-motion. Neuroimage 2002,16(4):873–82. 10.1006/nimg.2002.1181Google Scholar
  65. Vuilleumier P, Henson RN, Driver J, Dolan RJ: Multiple levels of visual object constancy revealed by event-related fMRI of repetition priming. Nature Neurosci 2002,5(5):491–9. 10.1038/nn839Google Scholar
  66. Jackson GM, Swainson R, Mullin A, Cunnington R, Jackson SR: ERP correlates of a receptive language-switching task. Q J Exp Psychol A 2004,57(2):223–40. 10.1080/02724980343000198Google Scholar
  67. Maclin EL, Gratton G, Fabiani M: Visual spatial localization conflict: an fMRI study. Neuroreport 2001,12(16):3633–6.Google Scholar
  68. Oga T, Honda M, Toma K, Murase N, Okada T, Hanakawa T, Sawamoto N, Nagamine T, Konishi J, Fukuyama H, Kaji R, Shibasaki H: Abnormal cortical mechanisms of voluntary muscle relxation in patients with writer's cramp: An fMRI study. Brain 2002, 125: 895–903. 10.1093/brain/awf083Google Scholar
  69. Indovina I, Sanes JN: Combined visual attention and finger movement effects on human brain representations. Exp Brain Res 2001,140(3):265–79. 10.1007/s002210100796Google Scholar
  70. Nakahara K, Hayashi T, Konishi S, Miyashita Y: Functional MRI of macaque monkeys performing a cognitive set-shifting task. Science 2002,295(5559):1532–6. 10.1126/science.1067653Google Scholar
  71. Nybakken GE, Quigley MA, Moritz CH: Test-retest precision of functional magnetic resonance imaging processed with independent component analysis. Neuroradiology 2002,44(5):403–6. 10.1007/s00234-001-0722-6Google Scholar
  72. Yoo SS, Talos IF, Golby AJ, Black PM, Panych LP: Evaluating requirements for spatial resolution of fMRI for neurosurgical planning. Hum Brain Mapp 2004,21(1):34–43. 10.1002/hbm.10148Google Scholar
  73. Rohde GK, Aldroubi A, Dawant BM: The adaptive bases algorithm for intensity-based nonrigid image registration. IEEE Trans Med Imaging 2003,22(11):1470–9. 10.1109/TMI.2003.819299Google Scholar
  74. Ugurbil K: Magnetic Resonace Studies of Brain Function and Neurochemistry. Annu Rev Biomed Eng 2002, 2: 633–60. 10.1146/annurev.bioeng.2.1.633Google Scholar
  75. Fischl B, Sereno MI, Dale AM: Cortical Surface-based analysis I: Segmentation and Surface Reconstruction. Neuroimage 1999, 9: 195–207. 10.1006/nimg.1998.0396Google Scholar
  76. Fischl B, Sereno MI, Dale AM: Cortical Surface-based analysis II: Inflation, Flattening, and a Surface-Based Coordinate System. Neuroimge 1999, 9: 179–194. 10.1006/nimg.1998.0395Google Scholar
  77. Fischl B, Salat DH, Busa E, Albert M, Dieterich M, Haselgrove C, van der Kouwe A, Killiany R, Kennedy D, Klaveness S, Montillo A, Makris N, Rosen B, Dale AM: Whole brain segmentation: Automated labeling of neuroanatomical structures in the human brain. Neurone 2002, 33: 341–355. 10.1016/S0896-6273(02)00569-XGoogle Scholar
  78. Cachier P, Mangin JF, Pennec X, Riviere D, Papadopoulos-Orfanos D, Regis J, Ayachi N: Multipatient registration of brain MRI using intensity and geometric features. Proceedings of MICCAI (Edited by: Niessan W, Vierever M). 2001, LNCS 2208: 734–742.Google Scholar
  79. Lohmann G, Yves von Cramon D: Automatic detection and labeling of the human cortical fields in magnetic resonance data sets. In Computer Vision, Fifth European Conference, EECV Friburg, Germany (Edited by: Buckhardt H, Neumann B). Springer-Verlag, Berlin 1998, 369–381.Google Scholar


© Sharma and Sharma; licensee BioMed Central Ltd. 2004

This article is published under license to BioMed Central Ltd. This is an Open Access article: verbatim copying and redistribution of this article are permitted in all media for any purpose, provided this notice is preserved along with the article's original URL.