A Connectivity-based Psychometric Prediction Framework for Brain-behavior Relationship Studies

(a) (b) 0.1 0.2 0.3 0.4 NE O F A C -O Au d iti_o n Co n tr_a st S_e n St re_n g th De xte rit_y Ga it S pe ed En du ran_ce Proc _Spe ed Fluid In_t Total Comp Cryst al Com p Fluid Com p Delay Di sc Read ing Pic V ocab Pic Seq List Sor t 2-ba ck A cc 2-b ack Acc Fa ce Lan g t ask Re l A cc Ca rd S o rt Fl a_n ke r Fl u id In t R T 2-ba ck R_T Ga m blin g R_T Re l RT Em ot R ecog _RT Em ot R_ecog + ve Psych + ve Social Rel Self Eff Stress Anger Af fect Fear Affe ct Sadn ess NEO FAC -N NE OFA C-E NE OFA C-A NE OF AC -C 0.1 0.2 0.3 0.4 NE O F A C -O Au d iti_o n Co n tr_a st S_e n St re_n g th De xte rit_y Ga it S pe ed En du ran_ce Proc _Spe ed Fluid In_t Total Comp Cryst al Com p Fluid Com p Delay Di sc Read ing Pic V ocab Pic Seq List Sor t 2-ba ck A cc 2-b ack Acc Fa ce Lan g t ask Re l A cc Ca rd S o rt Fl a_n ke_r Fl u id In t R T 2-ba ck R_T Ga m blin g R_T Re l RT Em ot R ecog _RT Em ot R_ecog + ve Psych + ve Social Rel Self Eff Stress Ange r Affec t Fear Affe ct Sadn ess NEO FAC -N NE OFA C-E NE OFA C-A NE OF AC -C 0.1 0.2 0.3 0.4 NE O F A C -O Au d iti_o n Co n tr_a st S_e n St re_n g th De xte rit_y Ga it S pe ed En du ran_ce Proc _Spe ed Fluid In_t Total Comp Cryst al Com p Fluid Com p Delay Di sc Read ing Pic V ocab Pic Seq List Sor t 2-ba ck A cc 2-b ack Acc Fa ce Lan g t ask Re l A cc Ca rd S o rt Fl a_n ke_r Fl u id In t R T 2-ba ck R_T Ga m blin g R_T Re l RT Em ot R ecog _RT Em ot R_ecog + ve Psych + ve Social Rel Self Eff Stress Ange r Affec t Fear Affe ct Sadn ess NEO FAC -N NE OFA C-E NE OFA C-A NE OF AC -C

Sensorimotor Cognition (general) Cognition (specific) Emotion /

Socioaffective Personality 0.1 0.2 0.3 0.4 NE O F A C -O Au d iti_o n Co n tr_a st S_e n St re_n g th De xte rit_y Ga it S pe ed En du ran_ce Proc _Spe ed Fluid In_t Total Comp Cryst al Com p Fluid Com p Delay Di sc Read ing Pic V ocab Pic Seq List Sor t 2-ba ck A cc 2-b ack Acc Fa ce Lan g t ask Re l A cc Ca rd S o rt Fl a_n ke r Fl u id In t R T 2-ba ck R_T Ga m blin g R_T Re l RT Em ot R ecog _RT Em ot R_ecog + ve Psych + ve Social Rel Self Eff Stress Anger Af fect Fear Affe ct Sadn ess NEO FAC -N NE OFA C-E NE OFA C-A NE OF AC -C

minimal FIX FIX+GSR

MLR

SVR

EN

Jianxiao Wu

_{, Simon B. Eickhoff}

_{, Felix Hoffstaedter}

_{, Kaustubh R. Patil}

_{, Holger}

Schwender

_{, & Sarah Genon}

1 Institute of Systems Neuroscience, Heinrich Heine University Düsseldorf, Düsseldorf, Germany;

2 Institute of Neuroscience and Medicine (INM-7: Brain and Behaviour), Research Centre Jülich, Jülich, Germany 3 Mathematical Institute, Heinrich Heine University Düsseldorf, Düsseldorf, Germany

j.wu@fz-juelich.de

A Connectivity-based Psychometric Prediction

Framework for Brain-behavior Relationship Studies

Methods

Introduction

• Relationships between brain regions and behavioral functions can be studied by relating inter-individual psychometric variability to variability of brain regional connectivity • Recent availability of population-based neuroimaging datasets with extensive psychometric characterisation [1] opens promising perspectives to investigate these relations • The multivariate nature of connectivity-based prediction models severely limits interpretation from a cognitive neuroscience perspective.

• To address this issue, we propose a connectivity-based psychometric prediction (CBPP) framework based on individual region’s connectivity profile.

Preliminary evaluation

Conclusion

connectivity and EN

• Overall, FIX and GSR pre-processing improve prediction accuracy • SVR and EN perform comparably well

Acknowledgments: This study was supported by the Deutsche Forschungsgemeinschaft (DFG, GE 2835/1-1, EI 816/4-1), the Helmholtz Portfolio Theme ‘Supercomputing and Modeling for the Human Brain’, and the European Union’s Horizon 2020 Research and

Innovation Programme under Grant Agreement No. 785907 (HBP SGA2).

References: [1] Van Essen DC, et al. 2012. “The Human Connectome Project: a data acquisition perspective”. Neuroimage. [2] Schaefer A, et al. 2018. Local-Global parcellation of the human cerebral cortex from intrinsic functional connectivity MRI. Cerebral Cortex. [3]

Chatterjee S, Hadi AS. 1986. ”Influential observations, high leverage points, and outliers in linear regression”. Statistical Sciences. [4] Cortes C and Vapnik VN. 1995. “Support-vector networks:. Machine Learning. [5] Zou H and Hastie T. 2005. “Regularization and variable selection via the elastic net”. Journal of the Royal Statistical Society: Series B (Statistical Methodology).

5. Train and test regressors using 10-fold cross-validation

Accuracy = Pearson correlation between actual and predicted psychometric scores

Multiple Linear Regression

(MLR) [3] Linear Support Vector Regression (SVR) [4] Elastic Nets (EN) [4]

4. Adjust for confounds

Age, Gender, Handedness, Brain size, Age2_{, Gender*age, Gender*age}2_{, ICV, Acquisition quarter}

3. Compute functional connectivity (FC)

Pearson correlation Partial correlation (with L2 regularization)

2. Parcellate RS-fMRI data with Schaefer atlas [2]

100-parcel granularity 200-parcel granularity 300-parcel granularity 400-parcel granularity

1. Pre-processing of RS-fMRI data

HCP minimal preprocessing pipeline (“minimal”)

Minimal + ICA-FIX (FIX)

FIX + global signal regression (FIX+GSR)

Parcel-wise prediction

Figure 1. Average prediction accuracy across the 20 most well predicted psychometric variables using whole-brain connectivity. Error bars represent 95% confidence interval across psychometric variables.

Figure 3. Prediction accuracy variation across the brain in left and right hemisphere, for four selected psychometric scores. Negative accuracies were set to zero and shown in gray.

Figure 2. Parcel-wise psychometric profiles for two pairs of selected parcels. Yellow filled contour shows whole-brain prediction pattern, while blue contour shows parcel-based prediction profile.

• Parcel-based predictions allow investigations of the neurobiological relevance of the prediction model and hence of the selected framework

• Future studies should investigate the transferability of this framework to older and clinical populations.

Methods: FIX, 300-parcel granularity, Pearson correlation, SVR

(a) (b) (c) (d) (e) (f) Strength (Age-adjusted)

Crystal Intelligence Composite (Age-adjusted)

2-back Overall Accuracy