FMRI Signal Denoising by Dictionary Learning for High-Resolution Functional Connectivity Inference

Stand-By Time

Thursday, June 29, 2017: 12:45 PM - 2:45 PM

Submission No:


Submission Type:

Abstract Submission 

On Display:

Wednesday, June 28 & Thursday, June 29 


Seongah Jeong1, Xiang Li2, Hamed Farhadi3, Quanzheng Li4, Vahid Tarokh3


1Harvard university, Cambridge, MA, 2Massachusetts General Hospital, Boston, MA, 3Harvard University, Cambridge, MA, 4​Massachusetts General Hospital, Boston, MA

First Author:

Seongah Jeong    -  Lecture Information | Contact Me
Harvard university
Cambridge, MA


Functional Magnetic Resonance Imaging (fMRI) is widely applied for delineating the temporal and spatial patterns of functional brain connectivity. One of the major challenges in fMRI anlaysis is the the low signal-to-noise ratio (SNR) caused by various types of noises which limits the identification of the subtle differences in connectivity mappings. The successes of image denoising such as in the field of natural image processing motivates the application of denosing methods on fMRI data for post-processing. In this work, we propose a dictionary learning-based denoising method to support the high-resolution brain functional connectivity analysis. Our basic premise is that there exists a sparse underlying structure of the functional brain organizations, thus the fMRI signals on region can be represented by a limited number of basis learned from the whole signal matrix. By applying the proposed method on the fMRI data during motor task from the Human Connectome Project database¹, we have discovered novel connectivity patterns originated from cerebellum and within premotor/SMA cortex.


The proposed dictionary learning-based denoising method aims at factorizing the fMRI matrix into a suitable dictionary and the corresponding sparse coefficient matrix². The number of dictionary bases and sparsity constraint for the factorization are determined experimentally based on fine-grained parameter tests to achieve the balance between de-noising performance and the preservation of the original information. In order to validate the model performance and evaluate its performance, we applied the proposed method on the Human Connectome Project (HCP) task fMRI (tfMRI) data¹. fMRI images of each subject were registered to the standard MNI152 space using linear registration. Voxel-wise fMRI signals were then mapped into 924 regions of interest (ROI) based on the Talairach atlas, while certain regions were removed from the atlas as the number of grey-matter voxels in those regions is zero. The average of voxel-wise fMRI signals in each ROI was then used as representative signals to infer the functional connectivity using Pearson Correlation, resulting in a 924*924 correlation matrix for each subject. We then performed the t-test on the correlation matrices before and after denoising across multiple subjects to discover the group-wise significant differences. Furthermore, we also tested the method on two groups of data, with each group consisted of 30 subjects. The significant new functional connectivities discovered from the two groups were cross-validated to ensure the consistency of our discoveries.


Totally 686 and 842 novel pair-wise connectives were found in each group, which show significantly increased (p<0.05) correlation value after d-noising and were above the predefined threshold (correlation>0.85). Out of them, 520 (averagely 68%) are identical, validating the reproducibility of our discoveries. Two of the most prominent types of the novel connectivities are shown in Fig. 1. Firstly, we have discovered multiple functional paths connecting the cerebellum area with the frontal lobe (Talairach#809, Inferior Frontal Gyrus), temporal lobe (Talairach#866, #867, right and left precuneus), as well as part of limbic lobe (Talairach #813, Anterior Cingulate). According to our previous studies, signals in cerebellum are usually affected by high-frequency noises and their connectivities are much more difficult to analyze. In such cases the denoising method can be very helpful. Secondly, we have also found increased connectivities within the motor area, especially in the premotor cortex and supplementary motor area (SMA).
Supporting Image: fig1.png
   ·Visualization of the two types of discoveries from functional connectivity analysis based on de-nosed fMRI data.


Preliminary results from testing the proposed method shows that consistent and neuroscientifically meaningful novel connectivities can be discovered by denoising. In addition, this scheme can be readily used with other functional neuroimaging methods to reveal previously hidden patterns and increase the resolution of the analysis.

Imaging Methods:


Modeling and Analysis Methods:

fMRI Connectivity and Network Modeling 1


Data analysis
Machine Learning

1|2Indicates the priority used for review

Would you accept an oral presentation if your abstract is selected for an oral session?


I would be willing to discuss my abstract with members of the press should my abstract be marked newsworthy:


Please indicate below if your study was a "resting state" or "task-activation” study.


By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute the presentation in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels or other electronic media and on the OHBM website.

I accept

Healthy subjects only or patients (note that patient studies may also involve healthy subjects):

Healthy subjects

Internal Review Board (IRB) or Animal Use and Care Committee (AUCC) Approval. Please indicate approval below. Please note: Failure to have IRB or AUCC approval, if applicable will lead to automatic rejection of abstract.

Not applicable

Please indicate which methods were used in your research:

Functional MRI

For human MRI, what field strength scanner do you use?


Which processing packages did you use for your study?


Provide references in author date format

Aharon, M., Elad, M., Bruckstein, A. (2006), ‘K-SVD: An algorithm for designing overcomplete dictionaries for sparse representation’, IEEE Transactions on signal processing, vol. 54, no. 11, pp. 4311-4322.

Barch, D.M., Burgess, G.C., Harms, M.P., Petersen, S.E., Schlaggar, B.L., Corbetta, M., Glasser, M.F., Curtiss, S., Dixit, S., Feldt, C., Nolan, D., Bryant, E., Hartley, T., Footer, O., Bjork, J.M., Poldrack, R., Smith, S., Johansen-Berg, H., Snyder, A.Z., Van Essen, D.C., for the WU-Minn HCP Consortium (2013), ‘Function in the human connectome: task-fMRI and individual differences in behavior’, Neuroimage, vol. 80, 169–189.