Re-working of Jennifer Stark's Information Visualization thesis. The original can be found here.The deliverable was 3+ visualizations for the web.

Visualizing Human Neuroimaging Data

Exploring transient networks in the brain using novel neuroimaging techniques

Project Goals

1: What is the data quality at the subject level?

2: Can we detect transient networks in resting state data?

"Resting state" refers ro MRI scans where the participant is asked to lie still and awake, and think of nothing in particular.

3: Which data-collection method is best? See "Data" for more information on the data collection methods

Data

Multiband EPI is a new method used to collect whole brain volumes in a significantly shorter time frame, and/or finer spatial detail.

We collected high-temporal resolution:

    • 1 second,
    • voxel size 2.23mm

and high-spatial resolution:

    • 2 seconds,
    • voxel size 1.53mm

datasets from each participant.


Analysis and Visualizations

Several types of analysis was performed on this data, including network analysis focussing on the default mode, executive control, and salience networks; and cluster analysis.

Visualization approaches include network graphs, and matrices/heat maps.

Analysis and Visualizations

Individual data

Select Dataset

Project questions 1:

What is the data quality at the subject level?

Goal:

To assess data quality in terms of correlation strength and variance within each network.

Higher mean values indicates that the brain regions comprising that network are communicating well with each other.

Higher standard deviation values (longer lines) means that there is more variance within the network. This could be due to misplaced network nodes, preprocessing errors, excess motion by that subject, or that the regions comprising the network are genuinly not communicating well with each other.

Conclusion:

Subjects 105 and 107 had low means (closer to zero) and high standard deviations. Repositioning the nodes did not improve the outcome. Therefore these subjects' data were excluded from Group analysis for both EPI1 and EPI2.

Default Mode Network (DMN)

Posterior Cingulate Cortex

Angular Gyrus

Parahippocampal Cortex

Salience Network (SN)

Amygdala

Anterior Insula

Temporoparietal Junction

Executive Control Network (ECN)

Anterior Cingulate Cortex

dorsal Posterior Parietal Cortex

dorsolateral Prefrontal Cortex

Caudate Nucleus

EPI1

High Temporal Resolution

Temporal resolution: 1 second

Spatial resolution: 2.23mm

EPI2

High Spatial Resolution

Temporal resolution: 2 seconds

Spatial resolution: 1.53mm

Group data

Chart Key:

Project questions 3:

Which data-collection method is the best?

Goal:

To assess data quality in terms of correlation strength within each network.

Higher mean values indicates that the brain regions comprising that network are communicating well with each other.

Higher standard deviation values (longer lines) means that there is more variance within the network, and that brain regions comprising the network are not communicating well with each other.

Conclusion:

While excluding Subjects 105 and 107 due to low means (close to zero) with high standard deviations, the resulting group means and standard deviations suggest that higher temporal resolution (EPI1) is better at capturing connectivity within these networks.

Default Mode Network (DMN)

Posterior Cingulate Cortex

Angular Gyrus

Parahippocampal Cortex

Salience Network (SN)

Amygdala

Anterior Insula

Temporoparietal Junction

Executive Control Network (ECN)

Anterior Cingulate Cortex

dorsal Posterior Parietal Cortex

dorsolateral Prefrontal Cortex

Caudate Nucleus

EPI1

High Temporal Resolution

Temporal resolution: 1 second

Spatial resolution: 2.23mm

EPI2

High Spatial Resolution

Temporal resolution: 2 seconds

Spatial resolution: 1.53mm

EPI1 Group Network

EPI2 Group Network

Project question 3:

Which data-collection method is the best?

Goal:

To assess data quality in terms of correlation strength within and between networks.

One way to do that is to look at the connections between brain regions. By exploring different thresholds of "connectivity" (correlation z-score), we can compare EPI1 and EPI2 methods. We would expect a superior data collection method will detect more connections within and between networks at any given threshold, while also detecting stronger connections overall.

Conclusion:

While excluding Subjects 105 and 107 due to terrible means (close to zero) and high standard deviations, the resulting group network graphs show that higher temporal resolution (EPI1) is better at detecting connections between brain regions within networks and also between networks.

Default Mode Network (DMN)

Posterior Cingulate Cortex (PCCx)

Angular Gyrus (AngG)

Parahippocampal Cortex (ParaHC)

Salience Network (SN)

Amygdala (AMY)

Anterior Insula (AI)

Temporoparietal Junction (TPJ)

Executive Control Network (ECN)

Anterior Cingulate Cortex (ACC)

dorsal Posterior Parietal Cortex (dPPCx)

dorsolateral Prefrontal Cortex (dlPFCx)

Caudate Nucleus (Caud)

EPI1

High Temporal Resolution

Temporal resolution: 1 second

Spatial resolution: 2.23mm

EPI2

High Spatial Resolution

Temporal resolution: 2 seconds

Spatial resolution: 1.53mm

EPI1 Matrix

EPI2 Matrix

Project question 3:

Which data-collection method is the best?

Goal:

To assess data quality in terms of correlation strength within and between networks.

One way to do that is to look at the connections between brain regions. By exploring different thresholds of "connectivity" (correlation z-score), we can compare EPI1 and EPI2 methods. We would expect a superior data collection method will detect more connections within and between networks at any given threshold, while also detecting stronger connections overall.

Conclusion:

While excluding Subjects 105 and 107 due to terrible means (close to zero) and high standard deviations, the resulting group network graphs show that higher temporal resolution (EPI1) is better at detecting connections between brain regions within networks and also between networks.

Default Mode Network (DMN)

Posterior Cingulate Cortex (PCCx)

Angular Gyrus (AngG)

Parahippocampal Cortex (ParaHC)

Salience Network (SN)

Amygdala (AMY)

Anterior Insula (AI)

Temporoparietal Junction (TPJ)

Executive Control Network (ECN)

Anterior Cingulate Cortex (ACC)

dorsal Posterior Parietal Cortex (dPPCx)

dorsolateral Prefrontal Cortex (dlPFCx)

Caudate Nucleus (Caud)

EPI1

High Temporal Resolution

Temporal resolution: 1 second

Spatial resolution: 2.23mm

EPI2

High Spatial Resolution

Temporal resolution: 2 seconds

Spatial resolution: 1.53mm

Cluster Analysis:

EPI1 Cluster 1

EPI1 Cluster 2

Project question 2 and 3:

Can we detect transient networks in resting state data?

Which data collection method is the best?

Goal and Results:

To identify transient networks, "clusters", "centroids", or "brain states" in grouped EPI1 and EPI2 data.

Clusters, or brain states, were identified using an unsupervised clustering algorithm (Dirichlet Process Gaussian Mixture Model [DPGMM]). It found that EPI1 was best divided into two clusters, or brain states, where as EPI2 only contained one cluster. Therefore, only EPI1 Cluster 1 and Cluster 2 are shown here (threshold >0.3 z-score).

Conclusion:

EPI1 resting state data comprises two recurring brain states over the course of nine minutes. One brain state (Cluster 1) consists of a highly connected framework where connectivity is higher between brain regions within and between networks. The second brain state (Cluster 2) is much less connected overall. Therefore, we are able to detect transient networks using high temporal resoution (EPI1).

Default Mode Network (DMN)

Posterior Cingulate Cortex (PCCx)

Angular Gyrus (AngG)

Parahippocampal Cortex (ParaHC)

Salience Network (SN)

Amygdala (AMY)

Anterior Insula (AI)

Temporoparietal Junction (TPJ)

Executive Control Network (ECN)

Anterior Cingulate Cortex (ACC)

dorsal Posterior Parietal Cortex (dPPCx)

dorsolateral Prefrontal Cortex (dlPFCx)

Caudate Nucleus (Caud)

EPI1

High Temporal Resolution

Temporal resolution: 1 second

Spatial resolution: 2.23mm

EPI1 Cluster 1

EPI1 Cluster 2

Project question 2 and 3:

Can we detect transient networks in resting state data?

Which data collection method is the best?

Goal and Results:

To identify transient networks, "clusters", "centroids", or "brain states" in grouped EPI1 and EPI2 data.

Clusters, or brain states, were identified using an unsupervised clustering algorithm (Dirichlet Process Gaussian Mixture Model [DPGMM]). It found that EPI1 was best divided into two clusters, or brain states, where as EPI2 only contained one cluster. Therefore, only EPI1 Cluster 1 and Cluster 2 are shown here.

Conclusion:

EPI1 resting state data comprises two recurring brain states over the course of nine minutes. One brain state (Cluster 1) consists of a highly connected framework where connectivity is higher (more pink) between brain regions within and between networks. The second brain state (Cluster 2) is much less connected overall (more green). Therefore, we are able to detect transient networks using high temporal resoution (EPI1).

Default Mode Network (DMN)

Posterior Cingulate Cortex (PCCx)

Angular Gyrus (AngG)

Parahippocampal Cortex (ParaHC)

Salience Network (SN)

Amygdala (AMY)

Anterior Insula (AI)

Temporoparietal Junction (TPJ)

Executive Control Network (ECN)

Anterior Cingulate Cortex (ACC)

dorsal Posterior Parietal Cortex (dPPCx)

dorsolateral Prefrontal Cortex (dlPFCx)

Caudate Nucleus (Caud)

EPI1

High Temporal Resolution

Temporal resolution: 1 second

Spatial resolution: 2.23mm