Exploring Default Mode Network Association with Naturalistic Stimuli Using Topological Data Analysis
Poster No:
1552
Submission Type:
Abstract Submission
Authors:
Iqra Ejaz1, Sadia Shakil2
Institutions:
1BiCoNeS lab, Department of Electrical Engineering, Institute of Space Technology, Islamabad, Pakistan, 2Department of Biomedical Engineering, The Chinese University of Hong Kong, Hong Kong
First Author:
Iqra Ejaz
BiCoNeS lab, Department of Electrical Engineering, Institute of Space Technology
Islamabad, Pakistan
BiCoNeS lab, Department of Electrical Engineering, Institute of Space Technology
Islamabad, Pakistan
Co-Author:
Introduction:
The Default Mode Network (DMN) is a sense-making network of the brain influenced by naturalistic stimuli such as movies, stories, and music. Understanding the relation between DMN and naturalistic stimuli can assist in comprehension of such stimuli and our thought process. This study utilizes topological data analysis (TDA) from algebraic topology to extract significant DMN and demographic covariates for distinguishing between a) modalities, and b) contents of the incoming stimuli.
Methods:
The dataset (Zadbood, 2017) consisted of fMRI data of listening (to narration) and watching of two movies (Merlin and Sherlock). Out of 36 participants, 18 listened to the narration of Merlin and watched Sherlock; and the other 18 listened to the narration of Sherlock and watched Merlin. We pre-processed the data using statistical parametric modelling (SPM) and then extracted DMN's regions-of-interest (ROIs) from (Raichle 2015)(Hospital and Chase 2008), consisting of five cortices (posterior cingulate, anterior cingulate and medial prefrontal, inferior parietal, hippocampal formation, lateral temporal).
Pairwise correlation between the DMN cortices was estimated using Pearson correlation coefficient. Mean correlation (mCorr) value of each matrix was computed and TDA was implemented on the correlation matrices to extract the betti-curves for each matrix (Gracia-Tabuenca, 2020). TDA extracts the shape and structure of data by identifying patterns and features such as loops and holes. Betti curves represents the transition of these topological features from being isolated nodes to a single connected component. From these betti curves features such as (area under curve (AUC), slope and kurtosis) were extracted.
Binary logistic regression was implemented to extract covariates association with modalities (listening vs. watching) and contents (Merlin vs. Sherlock). Three demographic (Age, sex, recall score) and four data driven (mCorr, AUC, slope, kurtosis) covariates were used for each participant. Best covariates model selection was based upon lowest AIC (Akaike information criterion) values and covariates with significant association were with p-values <=0.05 (Annas, Aswi, and Abdy 2022).
Using significant covariates, similar analysis was performed on the cortices of DMN to get the information about cortex having positive or negative association with the modalities/contents.
Pairwise correlation between the DMN cortices was estimated using Pearson correlation coefficient. Mean correlation (mCorr) value of each matrix was computed and TDA was implemented on the correlation matrices to extract the betti-curves for each matrix (Gracia-Tabuenca, 2020). TDA extracts the shape and structure of data by identifying patterns and features such as loops and holes. Betti curves represents the transition of these topological features from being isolated nodes to a single connected component. From these betti curves features such as (area under curve (AUC), slope and kurtosis) were extracted.
Binary logistic regression was implemented to extract covariates association with modalities (listening vs. watching) and contents (Merlin vs. Sherlock). Three demographic (Age, sex, recall score) and four data driven (mCorr, AUC, slope, kurtosis) covariates were used for each participant. Best covariates model selection was based upon lowest AIC (Akaike information criterion) values and covariates with significant association were with p-values <=0.05 (Annas, Aswi, and Abdy 2022).
Using significant covariates, similar analysis was performed on the cortices of DMN to get the information about cortex having positive or negative association with the modalities/contents.
Results:
Fig. 1(a) shows the lowest AIC value (left) for mCorr and AUC and significant p-values (right) with positive association of these covariates with the watching data relative to listening data. Fig. 1(b) shows lowest AIC value (left) for ACMP and IP and significant p-values (right) for ACMP (mCorr + AUC) with positive association and IP (AUC only) with negative association with the watching data.
Fig. 2(a) shows lowest AIC (left) for AUC, slope, kurtosis and significant p-values (right) for AUC and slope with positive and negative association respectively with Sherlock. Fig. 2(b) shows lowest AIC value (left) for IP but no significant p-values for it, hence showing no significant association.
Positive association of mCorr with the watching data shows overall high mean correlation among the regions of DMN but does not provide information about the correlations of individual regions. However, higher AUC shows segregation of DMN regions providing information about individual regions. ACMP and IP were found to play a significant role in distinguishing different modalities of the incoming stimuli. AUC and slope being significantly associated with the Sherlock movie shows more segregated activations in the regions of DMN.
Fig. 2(a) shows lowest AIC (left) for AUC, slope, kurtosis and significant p-values (right) for AUC and slope with positive and negative association respectively with Sherlock. Fig. 2(b) shows lowest AIC value (left) for IP but no significant p-values for it, hence showing no significant association.
Positive association of mCorr with the watching data shows overall high mean correlation among the regions of DMN but does not provide information about the correlations of individual regions. However, higher AUC shows segregation of DMN regions providing information about individual regions. ACMP and IP were found to play a significant role in distinguishing different modalities of the incoming stimuli. AUC and slope being significantly associated with the Sherlock movie shows more segregated activations in the regions of DMN.
Conclusions:
We used combination of TDA and logistic regression in a novel manner to explore association of different demographic and data-driven covariates with the modalities and contents of the naturalistic stimuli. Our results suggest that our approach provides a meaningful and deep insight about the relationship of DMN activity with the modalities/contents of the incoming stimuli.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
fMRI Connectivity and Network Modeling 2
Keywords:
Data analysis
FUNCTIONAL MRI
1|2Indicates the priority used for review
Provide references using author date format
Hospital, Massachusetts General, and Chevy Chase. 2008. “The Brain ’ s Default Network Anatomy , Function , and Relevance to Disease.” 38: 1–38.
Raichle, Marcus E. 2015. “The Brain ’ s Default Mode Network.”
Gracia-Tabuenca, Z. a.-P. (2020). Topological data analysis reveals robust alterations in the whole-brain and frontal lobe functional connectomes in attention-deficit/hyperactivity disorder. eneuro.
Zadbood, A. a. (2017). How we transmit memories to other brains: constructing shared neural representations via communication. Cerebral cortex, 4988--5000.