Advanced signal processing methods for high quality funtional magnetic resonance imaging
Functional Magnetic Resonance Imaging (fMRI) is a tool for noninvasively exploring the functionality of the human brain. The method has already provided many insights to the function of the brain. Since examinations can be performed on widely available clinical MR scanners and without using exogenous contrast agents, the potential use for preoperative investigations and following up neurological diseases are important goals within reach. The proposed project aims at developing analysis methods that are able to extract relevant information from the large amount of data acquired in an fMRI examination. This includes many challenging problems such as compensating for patient motion, modeling brain hemodynamics, fusion between different MR images for neurological interpretation and efficient filtering approaches to locate active brain areas. To achieve the robustness required for routine clinical use of fMRI, advanced signal and image processing procedures that solve the above issues need to be developed and evaluated.
- Former Staff:
- Project Description:
The functionality of the human brain is still relatively unknown although much effort has been put into revealing its secrets. A new tool for this purpose is functional Magnetic Resonance Imaging (fMRI). The aim in fMRI is to map cognitive, motor and sensor functions to specific areas in the brain. The physical basis for the method is the fact that oxygenated and deoxygenated blood have different magnetic properties. When a neuron in the brain is active it consumes oxygen, which is supplied by the blood. To compensate for the increased rate of oxygen consumption in an active brain area, the bloodflow and the oxygenation level of the blood to this area are increased. This increase can be measured in a magnetic resonance scanner. Thus, we can locate areas of brain activity indirectly by locating areas with elevated blood oxygen levels.
To map e.g. the sensory function of the right index finger, one can for instance stimulate the finger on a volunteer with a brush, while images of the brain continuously are acquired by the MR-scanner. The scanner produces image-slices like the one below. During the stimulation of the finger there is an increase in image intensity (i.e. the active area becomes brighter) compared to a resting state. Thus, to detect activity we need to compare an image where the finger is stimulated by the brush to an reference image acquired in a resting state. The areas where the "activated" image are brighter than image acquired in the "rest"-state are areas activated when the brush stimulates the finger.
To capture the dynamics of the brain we must acquire each image-slice rapidly. Unfortunately, this makes the images heavily contaminated by random noise. Hence, it is not enough to acquire just one image in activity and one in rest, since it is likely that we can not detect any significant change in intensity due to the high noise level. For this reason typically 75 to 100 slices are acquired in each state of activity and rest respectively, in order to obtain a more reliable result.
The experiment is often divided into blocks where the volunteer alternately is asked to rest and to perform an activity. This is called the paradigm and it is viewed graphically to the right. At each dot an image is acquired. In our finger stimulation example above, the activity periods consist brushing the finger with the brush and the rest periods consist of just pure relaxation.
Now, we have collected a series of images where the volunteers right index finger was brushed and where the volunteer rested, according to the paradigm. We look for areas where the intensity in the image follows the paradigm. The image below illustrates the intensity values at a pixel when we look at it through time. Clearly, the pixel marked by green has an intensity-timecourse which has no resemblance at all to the paradigm, and we conclude that the finger-brushing did not affect this area of the brain. However, if we look at the pixel marked by blue we see a close match to the paradigm, and thus this area is activated when the brush stimulates the finger. It is not possible to inspect every pixel manually, we need a computer to do this automatically and to classify each pixel as "activated" or "nonactivated". The image below shows the final result from an experiment involving hand movement. Areas active during right hand movement are marked by green and areas active during left hand movement are marked by red.
The researchThe research aims at developing reliable methods for detecting activity in the fMRI images. Currently we are working on methods based on Canonical Correlation Analysis and Bilateral Filtering which adaptively filter the fMRI-images to extract the square-wave signal induced by the paradigm. Publications related to fMRI can be found here.
Rydell J, Borga M, Knutsson H. Robust correlation
analysis with an application to functional MRI. In Proceedings of IEEE
International Conference on Acoustics, Speech, & Signal Processing.
Las Vegas, Nevada, USA: IEEE, 2008;
Farnebäck G, Rydell J, Ebbers T, Andersson M, Knutsson H. Efficient computation of the inverse gradient on irregular domains. In IEEE Computer Society Workshop on Mathematical Methods in Biomedical Image Analysis (MMBIA'07). Rio de Janeiro, Brasil, 2007;.Rydell J, Knutsson H, Borga M. Rotational invariance in adaptive fMRI data analysis. In Proceedings of the IEEE International Conference on Image Processing. Atlanta, USA: IEEE, 2006;.
Rydell J, Knutsson H, Borga M. Adaptive filtering of fMRI data based on correlation and BOLD response similarity. In 2006 IEEE International Conference on Acoustics, Speech, and Signal Processing. Toulouse, France, 2006;.Rydell J. Adaptive spatial filtering of fMRI data. Lic. Thesis LiU-Tek-Lic-2005:55, Linköping University, Sweden, 2005. Thesis No. 1200, ISBN 91-85457-43-4.
Rydell J, Knutsson H, Borga M. Correlation controlled adaptive filtering for fMRI data analysis. In Proceedings of the 13th Nordic-Baltic conference on biomedical engineering and medical physics (NBC'05). Umeå, Sweden: NBC, 2005;.Rydell J, Borga M, Lundberg P, Knutsson H. Dimensionality and number of parameters in adaptive filtering of fMRI data. In Proceedings of the SSBA Symposium on Image Analysis. 2004;.
Rydell J, Borga M, Lundberg P, Knutsson H. Dimensionality and degrees of freedom in fMRI data analysis - a comparative study. In IEEE International Symposium on Biomedical Imaging (ISBI'04). Arlington, Virginia, USA, 2004;.