Representation Learning of Resting State fMRI with Variational Autoencoder

AbstractResting state functional magnetic resonance imaging (rs-fMRI) data exhibits complex but structured patterns. However, the underlying origins are unclear and entangled in rs-fMRI data. Here we establish a variational auto-encoder, as a generative model trainable with unsupervised learning, to disentangle the unknown sources of rs-fMRI activity. After being trained with large data from the Human Connectome Project, the model has learned to represent and generate patterns of cortical activity and connectivity using latent variables. Of the latent representation, its distribution reveals overlapping functional networks, and its geometry is unique to each individual. Our results support the functional opposition between the default mode network and the task-positive network, while such opposition is asymmetric and non-stationary. Correlations between latent variables, rather than cortical connectivity, can be used as a more reliable feature to accurately identify subjects from a large group, even if only a short period of data is available per subject.

Download Full-text

Evaluating the sensitivity of functional connectivity measures to motion artifact in resting-state fMRI data

10.1101/2020.05.04.072868 ◽

2020 ◽

Author(s):

Arun S. Mahadevan ◽

Ursula A. Tooley ◽

Maxwell A. Bertolero ◽

Allyson P. Mackey ◽

Danielle S. Bassett

Keyword(s):

Functional Connectivity ◽

Resting State ◽

Partial Correlation ◽

Motion Artifact ◽

Resting State Fmri ◽

Alternative Methods ◽

Fmri Data ◽

Retest Reliability ◽

Human Connectome Project ◽

Test Retest Reliability

AbstractFunctional connectivity (FC) networks are typically inferred from resting-state fMRI data using the Pearson correlation between BOLD time series from pairs of brain regions. However, alternative methods of estimating functional connectivity have not been systematically tested for their sensitivity or robustness to head motion artifact. Here, we evaluate the sensitivity of six different functional connectivity measures to motion artifact using resting-state data from the Human Connectome Project. We report that FC estimated using full correlation has a relatively high residual distance-dependent relationship with motion compared to partial correlation, coherence and information theory-based measures, even after implementing rigorous methods for motion artifact mitigation. This disadvantage of full correlation, however, may be offset by higher test-retest reliability and system identifiability. FC estimated by partial correlation offers the best of both worlds, with low sensitivity to motion artifact and intermediate system identifiability, with the caveat of low test-retest reliability. We highlight spatial differences in the sub-networks affected by motion with different FC metrics. Further, we report that intra-network edges in the default mode and retrosplenial temporal sub-networks are highly correlated with motion in all FC methods. Our findings indicate that the method of estimating functional connectivity is an important consideration in resting-state fMRI studies and must be chosen carefully based on the parameters of the study.

Download Full-text

Computing personalized brain functional networks from fMRI using self-supervised deep learning

10.1101/2021.09.25.461829 ◽

2021 ◽

Author(s):

Hongming Li ◽

Srinivasan Dhivya ◽

Zaixu Cui ◽

Chuanjun Zhuo ◽

Raquel E. Gur ◽

...

Keyword(s):

Deep Learning ◽

Resting State ◽

Brain Function ◽

Resting State Fmri ◽

Fmri Data ◽

Functional Networks ◽

Brain Functional Networks ◽

Human Connectome Project ◽

Functional Homogeneity ◽

Decoder Architecture

ABSTRACTA novel self-supervised deep learning (DL) method is developed for computing bias-free, personalized brain functional networks (FNs) that provide unique opportunities to better understand brain function, behavior, and disease. Specifically, convolutional neural networks with an encoder-decoder architecture are employed to compute personalized FNs from resting-state fMRI data without utilizing any external supervision by optimizing functional homogeneity of personalized FNs in a self-supervised setting. We demonstrate that a DL model trained on fMRI scans from the Human Connectome Project can identify canonical FNs and generalizes well across four different datasets. We further demonstrate that the identified personalized FNs are informative for predicting individual differences in behavior, brain development, and schizophrenia status. Taken together, self-supervised DL allows for rapid, generalizable computation of personalized FNs.

Download Full-text

Human Connectome Project resting state fMRI data organized into 60 fine-scaled retinotopically organized regions in cortical areas V1, V2 and V3

10.1101/340174 ◽

2018 ◽

Cited By ~ 2

Author(s):

Debra Ann Dawson ◽

Zixuan Yin ◽

Jack Lam ◽

Amir Shmuel

Keyword(s):

Visual Cortex ◽

Resting State ◽

Resting State Fmri ◽

Regions Of Interest ◽

Fmri Data ◽

Cortical Areas ◽

Human Connectome Project ◽

Public Data ◽

Visual Areas ◽

Data Release

AbstractThe data comprises 60 regions of interest (ROIs) from V1, V2, and V3 of the human visual cortex. Preprocessed data from the Human Connectome Project (HCP) 900 subjects public data release were utilized: 220 subjects were randomly selected, each with 4 scans of resting state fMRI data. Given that these subjects did not have retinotopy scans performed, the visual areas were defined using an anatomical template from Benson et al. (2014). Visual areas from each hemisphere were further divided along dorsal-ventral lines into quadrants, resulting in 4 quadrants per subject. Within each quadrant, fine scaled ROIs were defined by subdividing each visual area into 5 regions according to eccentricity. These data may be useful for studying retinotopically organized functional connectivity in the visual cortex using the HCP 3 Tesla dataset.

Download Full-text

A multi-measure approach for assessing the performance of fMRI preprocessing strategies in resting-state functional connectivity

10.1101/837609 ◽

2019 ◽

Cited By ~ 2

Author(s):

Michalis Kassinopoulos ◽

Georgios D. Mitsis

Keyword(s):

Resting State ◽

Signal To Noise Ratio ◽

Resting State Fmri ◽

Motion Artifacts ◽

Fmri Data ◽

Quality Of Data ◽

Resting State Functional Connectivity ◽

Physiological Processes ◽

Human Connectome Project ◽

Low Pass

AbstractIt is well established that head motion and physiological processes (e.g. cardiac and breathing activity) should be taken into consideration when analyzing and interpreting results in fMRI studies. However, even though recent studies aimed to evaluate the performance of different preprocessing pipelines there is still no consensus on the optimal strategy. This is partly due to the fact that the quality control (QC) metrics used to evaluate differences in performance across pipelines have often yielded contradictory results. Furthermore, preprocessing techniques based on physiological recordings or data decomposition techniques (e.g. aCompCor) have not been comprehensively examined. Here, to address the aforementioned issues, we propose a framework that summarizes the scores from eight previously proposed and novel QC metrics to a reduced set of two QC metrics that reflect the signal-to-noise ratio and the reduction in motion artifacts and biases in the preprocessed fMRI data. Using this framework, we evaluate the performance of three commonly used practices on the quality of data: 1) Removal of nuisance regressors from fMRI data, 2) discarding motion-contaminated volumes (i.e., scrubbing) before regression, and 3) low-pass filtering the data and the nuisance regressors before their removal. Using resting-state fMRI data from the Human Connectome Project, we show that the best data quality, is achieved when the global signal (GS) and about 17% of principal components from white matter (WM) are removed from the data. Finally, we observe a small further improvement with low-pass filtering at 0.20 Hz, but not with scrubbing.

Download Full-text

Spatiotemporal Trajectories in Resting-state FMRI Revealed by Convolutional Variational Autoencoder

10.1101/2021.01.25.427841 ◽

2021 ◽

Author(s):

Xiaodi Zhang ◽

Eric Maltbie ◽

Shella Keilholz

Keyword(s):

Functional Connectivity ◽

Resting State ◽

Brain Activity ◽

Resting State Fmri ◽

Spatiotemporal Patterns ◽

Network Activity ◽

Fmri Data ◽

Variational Autoencoder ◽

Spatiotemporal Trajectories ◽

The Brain

AbstractRecent resting-state fMRI studies have shown that brain activity exhibits temporal variations in functional connectivity by using various approaches including sliding window correlation, co-activation patterns, independent component analysis, quasi-periodic patterns, and hidden Markov models. These methods often model the brain activity as a discretized hopping among several brain states that are defined by the spatial configurations of network activity. However, the discretized states are merely a simplification of what is likely to be a continuous process, where each network evolves over time following its unique path. To model these characteristic spatiotemporal trajectories, we trained a variational autoencoder using rs-fMRI data and evaluated the spatiotemporal features of the latent variables obtained from the trained networks. Our results suggest that there are a relatively small number of approximately orthogonal whole-brain spatiotemporal patterns that capture the most prominent features of rs-fMRI data, which can serve as the building blocks to construct all possible spatiotemporal dynamics in resting state fMRI. These spatiotemporal patterns provide insight into how activity flows across the brain in concordance with known network structures and functional connectivity gradients.

Download Full-text