Functional Connectivity MultiVariate Pattern Analysis – A Novel Method for the Analysis of Human Connectome

In the article released in PLoS Computational Biology, researcher Alfonso Nieto-Castanon of Boston University describes a technique called fc-MVPA (functional connectivity Multivariate Pattern Analysis). The novel technique uses classic MVPA to evaluate each voxel in a three-dimensional space and traces functional connections to the brain. Monte Carlo simulations are also used in the study to demonstrate the accuracy and sensitivity of fc-MVPA. It is a powerful and potentially useful tool for researchers to further explore the complexities of the human connectome due to its theoretical and practical advantages.

A connectome is a comprehensive map of the brain’s neuronal connections that can be thought of as its “wiring diagram.” The nervous system of an organism is made up of neurons that communicate with one another via synapses. A connectome is created by following the path of a neuron in the nervous system and mapping where neurons are connected via synapses. Functional connectivity Magnetic Resonance Imaging (fcMRI) is used to define the state of the human connectome during rest or individual cognitive tasks. The whole human connectome state can be represented as a matrix of functional connectivity values in the most basic format, restricted only by the spatial resolution of the MRI acquisition procedure.

But with such brain-wide connectome inferences consisting of 60 billion individual statistical tests analyzing the complicated functional connection is taxing. Existing approaches addressed these issues by either narrowing the focus of the analyses to connectivity with one or a few regions in the brain (e.g., connectivity with the amygdala) and then performing seed-based connectivity analyses (SBC), or by limiting the analysis units from voxels to a larger region of interests finally performing ROI-to-ROI connectivity analyses (RRC). An alternate method compares functional networks between queried groups using Principal Component Analysis (PCA) or Independent Component Analysis (ICA). Both PCA and ICA, like seed-based techniques, can dramatically decrease the multiple-comparison problem by focusing on specific networks, each consisting of a set of functionally connected areas, and then analyzing measures of within- and between-network connectivity. 

However, PCA and ICA techniques have the same limitations as SBC. There is a large probability of false negatives, notably when differences in finer functional connectivity on a network level are neglected.

Therefore, to address these issues, functional connectivity Multivariate Pattern Analysis (fc-MVPA) was devised to solve the challenges of brain-wide connectome investigations utilizing multivariate pattern analysis approaches. Here, fc-MVPA uses a basic searchlight approach, where instead of focusing on the activity surrounding each voxel, it evaluates the full multivariate pattern of functional connections between a particular voxel and the rest of the brain.

Fc-MVPA Assesses Gender Differences in Connectivity

The gender disparities in resting state functional analysis were investigated using the Cambridge 1000 connectome dataset. To determine whether there are any variations between male and female subjects over the full voxel-to-voxel functional connectome. The results of fc-MVPA demonstrate a large number of regions with substantial gender-related disparities in connectivity. Given the number of areas displaying substantial gender effects, with the most significant impacts in female participants’ Inferior Frontal Gyrus pars triangularis and male subjects’ A left hemisphere cluster center.

Therefore, this analysis of gender variations in functional connectivity demonstrated the fc-MVPA inferential statistics’ with greater sensitivity to discover relevant effects across the whole human connectome. 

Validity and Sensitivity Estimation by Monte Carlo Simulation 

A set of simplified simulations were built to test the validity and sensitivity of the fc-MVPA inferential technique. All of the simulations took a dataset of 50 patients. Nearly 10,000 simulations and the respective Receiver Operating Characteristic (ROC) curves describing the true positive rate as a function of different prescribed false positive rates were employed to produce estimations of the validity. Each of the 10,000 simulations was repeated over 50 times to ensure sensitivity and validity.

The sensitivity analysis revealed that the sensitivity was high across the full range of patterns evaluated, only diminishing significantly as the number of patterns reached their maximum conceivable value. The validity tests revealed that the reported voxel-level p-values from fc-MVPA inferences matched the empirically observed false positive rates extremely precisely, with all investigated situations displaying accurate diagonal ROC curves.

Strengths of Fc-MVPA

Fc-MVPA exceeds other MVPA methodologies routinely utilized in neuroimaging in three ways: 

•  Instead of the activation patterns surrounding each voxel considered by many MVPA applications, fc-MVPA uses the patterns of connection between each voxel and the rest of the brain to characterize a subject’s mental state.

•  Second, rather than using a backward model to decode the known properties of a subject, fc-MVPA employs a forward model to test a researcher’s hypothesis about the subject’s connectivity state across the entire connectome. 

•  Third, in addition to the aforementioned inferential framework, fc-MVPA also provides a model-free characterization of the heterogeneous sources. 

The Monte Carlo simulations demonstrated that fc-MVPA conclusions are valid over the whole spectrum of tested scenarios, including the use of any arbitrary number of pattern scores, variable sample sizes, and scanning session durations. 

The main practical advantages of fc-MVPA in the context of brain-wide connectome inferences is that it combines the advantages of pattern analysis techniques, such as increased interpretability and reduced noise of lower-dimensional projections, with the advantages of a classical statistical framework, such as the ability to use popular approaches to group-level analyses. 

With this study, the author hopes the fc-MVPA can be a powerful and valuable tool for researchers worldwide to explore further the connectome’s complexity based on its theoretical and practical advantages.

Article Sources: Reference Paper

Learn More: