fMRI signal processing

Details fMRI signal processing fMRI signal processing

1998-11-25

2002-04-09

Lateef, Marvin M. (Department: 3737)

Surgery

Diagnostic testing

Detecting nuclear, electromagnetic, or ultrasonic radiation

C128S922000, C382S128000

Reexamination Certificate

active

06370416

ABSTRACT:

FIELD OF THE INVENTION
The invention relates in general to functional MR imaging, and in particular to fMRI signal analysis.
BACKGROUND OF THE INVENTION
Changes in neuronal activity responsive to the accomplishment of mental and/or physical tasks, such as touching a finger to thumb, are accompanied by physiological changes in regions of the brain associated with and/or controlling the activity. Physiological changes such as cerebral blood flow, blood volume, blood oxygenation and/or metabolism, occurring in such a region of the brain are made visible by functional MR imaging (fMRI).
A typical fMRI session comprises the following steps: (a) stimulating a subject (e.g. by asking him to perform, preferably, a periodic task, usually a task targeting a particular brain region, or be subjected to a periodic visual/audio/tactile stimuli), (b) MR imaging a region of the brain supposed to be involved in the accomplishment of that task, and (c) analyzing a time series of acquired images to determine physiological changes in the brain region.
Often, the signal to noise ratio of fMRI images is poor, so a synchronized detection method of image analysis is preferably used. It is expected that the physiological changes mirror the activity that the subject performs. Thus, the images are analyzed based on expected correlation between variations in pixel values in the analyzed regions and the activity performed by the subject. Generally, however, there is a delay between the performance of the activity and a change in the physiological variable. This delay, as well as the exact response of the pixel value to the activity, are generally not known in advance.
Reference is now made to
FIG. 1
which shows a flow chart for a general fMRI data analysis. Due to the repetitive nature of the activity performed by the subject and the dependence of the physiological changes on the activity, many aspects of the process are best described as a time varying periodic function. An “on-off-on-off-on” stimulation (or activity) paradigm
20
, such as touching a finger to thumb, may be represented by an activation function, where on's are 1s and off's are 0s. Activity
20
induces neuronal activity
22
. As a result of neuronal activity
22
, certain brain tissues react (
24
), for example blood vessels open to bring in oxygenated blood and/or neurons use up oxygen in a blood stream. These effects are represented by F
&agr;
(r, t
k
) (
24
), where F
&agr;
(r, t
k
) is the possibly non-linear response of the brain, as a function of location and time, to the activity. F
&bgr;
is the possibly non-linear response of a pixel, in the image, to the response of the brain F
&agr;
(r, t
k
) as imaged by the MR imager (
30
). q
ij
(t
k
)=F
&bgr;
(F
&agr;
(r, t
k
)) (
34
), is a pixel, (pixel ij); response function which relates variations in (ij)th pixel intensity, as imaged by the MR imager in step
30
, and/or other image parameters, such as pixel phase, to physiological changes
24
. In the above described brain and pixel responses, t
k
represents the discretization of time, which corresponds to the instant at which each MR image is acquired. Typically, t
k
=k*&Dgr;t, where &Dgr;t is the time difference between consecutive images. Alternatively, the images may be non equally spaced in time. Subscript ij represents the discretization of locality r. Typically, image registration (alignment) is performed so that a same image pixel corresponds to a same brain volume over an entire series of images. Typically, the pixel response function is difficult to detect and/or otherwise analyze because of the above mentioned low signal-to-noise ratio. A small signal and two different noise sources, instrumental noise
38
, and physiological fluctuations
40
, contribute to this low signal-to-noise ratio.
In a typical fMRI study, the activity of a brain tissue may be assessed by comparing the pixel response function q
ij
(t
k
)
34
, and the activation function. As it is difficult to detect directly pixel activity from the collected data because of the very low signal-to-noise ratio, the detection is performed synchronously by correlation,
42
, between the pixel response q
ij
(t
k
), embedded in noise and a certain detection (or reference) function,
36
, which is thought to best fit the detection task. The synchronously detected pixel response function, q
ij
(t
k
), is then compared with the activation function.
According to the correlation results
44
, which are reflected by the correlation coefficient(s) &rgr;
ij
, the analyzed pixel is said to have no activity
49
, if the correlation is poor, or to be active
50
, if the correlation is high. Intermediary correlation values may point to different levels of activity of the analyzed biological tissues contained in a voxel.
In conventional fMRI, the detection (or reference) function,
36
, which is thought to best fit the detection task is guessed at. The most commonly guessed detection functions are square pulses, which in fact are identical to the repetitive part of the activation function describing an “on-off-on-off” task, a sinusoidal pulse or an exponential pulse which are close to the square pulse. The drawback of this method is that there is no physiological or other objective basis for guessing a detection function, that a guessed detection function may be delayed relative to the activation function and that the activation function is usually not a square function.
“Time Course EPI of Human Brain Function during Task Activation” by Peter A. Bandettini et al., Magnetic Resonance in Medicine, Vol. 25, p. 390-397 (1992); “Processing Strategies for Time-Course Data sets in Functional MRI of the Human Brain”, by Peter A. Bandettini et al., Magnetic Resonance in Medicine, Vol. 30, p. 161-173 (1993) and “Real-Time Functional Magnetic Resonance Imaging”, by Robert W. Cox et al., Magnetic Resonance in Medicine, Vol. 33, p. 230-236 (1995) all of which are incorporated herein by reference, deal with fMRI signal processing using a guessed detection function.
In “Processing Strategies for Time-Course Data sets in Functional MRI of the Human Brain”, Bandettini et al. use a time-course function in a given pixel as detection (reference) function in fMRI signal analysis. This detection (reference) function may be an experimental time-course function of some particular pixel or a time-course function which is linearly filtered from several experimental time-course functions. This detection (reference) function is then correlated with time-course functions in other pixels. Linear filtering such as performed by Peter, A. Bandettini may eliminate local differences. But as Bandettini's detection function is an experimental time-course function f in some particular pixel or an average of several experimental f's, the linear filtering, as applied by Bandettini, does not ensure the obtaintion of a detection function which is most appropriate in the synchronous detection of a pixel response q
ij
(t
k
), embedded in noise.
In the claims and specification of the present application each of the verbs, “comprise” and “include” and conjugates thereof are used to convey that the object or objects of the verb are not necessarily a listing of all the components, elements or parts of the subject or subjects of the verb.
SUMMARY OF THE INVENTION
One aspect of some preferred embodiments of the present invention relates to calculating a detection function for use in synchronous detection in fMRI signal analysis. Preferably, the detection function for fMRI data analysis is derived from the fMRI data itself. Some preferred embodiments of the present invention relate to a method of determining a detection function by non-linear filtering, preferably, from response functions of a plurality of pixels situated in a region of interest whose behavior is studied. More preferably, the non-linear filtering method uses eigen analysis in order to separate at least two subspaces, namely signal subspace (one or more) and noise subspace, within the space of all the fMRI time course signals.

Affiliated with

Also associated with

Rating

fMRI signal processing does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with fMRI signal processing, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and fMRI signal processing will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2829473