Table 1: Characteristics

Total Mean SD
n 7086
age 9.94 0.63
Sex
Male (%) 50.4
Other (%) 0.01
Race (%)
White 56.8
Black 11.7
Asian 1.9
Other 10.4
Ethnicity (%)
Hispanic 19.2
IQ 102.45 17.38
rsfMRI Motion 0.22 0.21
dMRI Motion 1.27 0.36
CBCL Attention 53.56 5.79
CBCL Externalizing 45.28 10.03
CBCL Internalizing 48.26 10.4
Neighborhood Disadvantage
Area Deprivation Index-ADI 38.18 25.69
Child Opportunity Index-COI 62.51 29.3
Income Levels (%)
Level 1 (< $5,000) 2.9
Level 2 ($5,000-11,999) 3.3
Level 3 ($12,000-15,999) 2.3
Level 4 ($16,000-24,999) 4.2
Level 5 ($25,000-34,999) 6
Level 6 ($35,000-49,999) 8.2
Level 7 ($50,000-74,999) 13.8
Level 8 ($75,000-99,999) 15.2
Level 9 ($100,000-199,999) 32.2
Level 10 (> $200,000) 11.8

Table 2: Primary Model Performance Metrics

Brain Features Only Brain Features with Demographics
Modality Test Train Test Train
Accuracy AUC Accuracy AUC Accuracy AUC Accuracy AUC
Combined Family Income - Low (Levels 1-7) | High Income (Levels 8-10)
sMRI 0.685 0.709 0.694 0.739 0.741 0.784 0.753 0.801
RSFC 0.663 0.667 0.663 0.677 0.742 0.785 0.738 0.785
DTI 0.691 0.719 0.702 0.751 0.741 0.790 0.754 0.809
Multimodal 0.709 0.743 0.746 0.816 0.730 0.783 0.774 0.844
Area Deprivation Index, ADI - (Low/High ADI)
sMRI 0.651 0.663 0.675 0.707 0.666 0.706 0.706 0.749
RSFC 0.632 0.640 0.645 0.658 0.689 0.705 0.690 0.72
DTI 0.669 0.716 0.716 0.767 0.698 0.748 0.735 0.789
Multimodal 0.670 0.725 0.757 0.820 0.682 0.741 0.766 0.834
Child Opportunity Index, COI - (Low/High COI)
sMRI 0.661 0.682 0.701 0.746 0.723 0.748 0.748 0.8
RSFC 0.627 0.625 0.662 0.681 0.732 0.749 0.742 0.777
DTI 0.678 0.709 0.711 0.755 0.725 0.772 0.757 0.812
Multimodal 0.688 0.719 0.761 0.823 0.701 0.753 0.785 0.846

Income models classify groups split by median income level (Level 8) with those at median and above being put in one class and all those below in another. The classes in models based on neighborhood disadvantage indices (ADI, COI) were split by mean. Neural features were derived from the ABCD Study dataset (Data Release 5.1) for neuroanatomy (sMRI), resting-state connectivity (RSFC), and white-matter microstructure (DTI). Multimodal models combined all features.

sMRI: structural MRI, RSFC: Resting-state functional connectivity, DTI: Diffusion Tensor Imaging, AUC: Area Under the Curve

Table 3: Sensitivity Analysis - Classification of Bottom 10-20% of Income Distribution Model Performance

Brain Features Only Brain Features with Demographics
Modality Test Train Test Train
Accuracy AUC Accuracy AUC Accuracy AUC Accuracy AUC
Top 10-20% (Level 10)
sMRI 0.733 0.788 0.802 0.880 0.828 0.900 0.896 0.954
RSFC 0.684 0.734 0.729 0.787 0.849 0.919 0.868 0.934
DTI 0.756 0.799 0.835 0.904 0.796 0.872 0.909 0.961
Multimodal 0.712 0.781 0.975 0.980 0.791 0.868 0.982 0.986
Above Median (Level 9)
sMRI 0.722 0.758 0.781 0.831 0.800 0.860 0.847 0.907
RSFC 0.715 0.726 0.727 0.754 0.824 0.886 0.828 0.891
DTI 0.747 0.793 0.781 0.844 0.826 0.881 0.853 0.913
Multimodal 0.715 0.769 0.859 0.923 0.766 0.825 0.895 0.950
Median (Level 8)
sMRI 0.669 0.728 0.746 0.814 0.742 0.804 0.807 0.874
RSFC 0.638 0.697 0.684 0.749 0.748 0.828 0.783 0.853
DTI 0.684 0.735 0.756 0.833 0.738 0.797 0.814 0.899
Multimodal 0.647 0.683 0.880 0.936 0.674 0.717 0.912 0.952
Below Median (Level 6-7)
sMRI 0.599 0.618 0.675 0.732 0.608 0.669 0.713 0.782
RSFC 0.597 0.619 0.635 0.678 0.648 0.709 0.688 0.755
DTI 0.573 0.611 0.706 0.771 0.631 0.683 0.723 0.789
Multimodal 0.588 0.606 0.779 0.844 0.591 0.626 0.789 0.860

The performance metrics presented in this table reflect the the performance of models that classified the Top 10-20% (Level 10), Above Median (Level 9), Median (Level 8), Below Median (Levels 6-7) classes individually against the bottom 10-20% (Levels 1-5). Neural features were derived from the ABCD Study dataset (Data Release 5.1) for neuroanatomy (sMRI), resting-state connectivity (RSFC), and white-matter microstructure (DTI). Multimodal models combined all features.

sMRI: Structural MRI, RSFC: Resting-state functional connectivity, DTI: Diffusion Tensor Imaging, AUC: Area Under the Curve.

Table 4: Sex-Specific Prediction Using Best-Performing Models

Income COI
Modality 2 Class - High vs. Low Bot. 10-20%-Top 10-20% High vs. Low
Accuracy AUC Accuracy AUC Accuracy AUC
Male
Structural 0.683 0.712 0.677 0.810 0.694 0.705
Functional 0.666 0.678 0.581 0.715 0.629 0.621
DTI 0.723 0.751 0.768 0.870 0.702 0.724
Multimodal 0.727 0.765 0.773 0.871 0.706 0.724
Female
Structural 0.686 0.721 0.704 0.800 0.627 0.657
Functional 0.661 0.661 0.639 0.751 0.626 0.629
DTI 0.659 0.694 0.704 0.778 0.655 0.691
Multimodal 0.691 0.729 0.730 0.816 0.670 0.713
Income COI
Modality Above/Below Median Top/Bottom 10-20% High/Low
Accuracy AUC Accuracy AUC Accuracy AUC
Male
Structural 0.731 0.781 0.833 0.896 0.738 0.760
Functional 0.736 0.788 0.823 0.915 0.740 0.757
DTI 0.736 0.806 0.843 0.927 0.738 0.785
Multimodal 0.737 0.797 0.818 0.900 0.711 0.759
Female
Structural 0.750 0.788 0.833 0.910 0.707 0.736
Functional 0.749 0.780 0.798 0.913 0.724 0.741
DTI 0.746 0.774 0.815 0.888 0.711 0.759
Multimodal 0.722 0.770 0.777 0.878 0.691 0.745

The best performing models dichotomous median income, granular dichotomous bottom and top 10-20% , and COI) were used to predict the respective social markers of males and females individually. Neural features were derived from the ABCD Study dataset (Data Release 5.1) for neuroanatomy (sMRI), resting-state connectivity (RSFC), and white-matter microstructure (DTI). Multimodal models combined all features.

sMRI: Structural MRI, RSFC: Resting-state functional connectivity, DTI: Diffusion Tensor Imaging, AUC: Area Under the Curve.

Figure 2: Model Performance

Performance metrics (Accuracy and AUC) for all models: combined family income split by median income level and neighborhood disadvantage indices (A) and granular family income models with the bottom 10-20% of the income distribution (Levels 1-5) as the reference class (B).

COI: Child Opportunity Index; ADI: Area Deprivation Index; AUC: Area Under the Curve;

Figure 3: sMRI Models

Most important variables for combined family income split by median models for all imaging modalities and both excluding and including demographic information: Structural MRI, sMRI (A), Resting state functional connectivity, RSFC (B), Diffusion Tensor Imaging, DTI (C), Multimodal (D).

STS: Superior Temporal Sulcus; Fiber TV: Fiber Tract Volume; FA: Fractional Anisotropy; Diff: Diffusivity; lh: Left Hemisphere; rh: Right Hemisphere; ex. CC: excluding Corpus Callosum; Tot: Total; Hemi: Hemisphere; Sup: Superior; Avg: Average;Fasc: Fasciculus; Long: Longitudinal;SM:Sensorimotor.

Figure 4: DTI Models

Most important variables for combined family income models classifying the top and bottom 10-20% for all imaging modalities and both excluding and including demographic information: Structural MRI, sMRI (A), Resting state functional connectivity, RSFC (B), Diffusion Tensor Imaging, DTI (C), Multimodal (D).

STS: Superior Temporal Sulcus; Fiber TV: Fiber Tract Volume; FA: Fractional Anisotropy; Diff: Diffusivity; lh: Left Hemisphere; rh: Right Hemisphere; ex. CC: excluding Corpus Callosum; Tot: Total; Hemi: Hemisphere; Sup: Superior; Avg: Average;Fasc: Fasciculus; Long: Longitudinal;SM:Sensorimotor.

Figure 5: Multimodal Models

Most important variables for combined family income models classifying the top and bottom 10-20% for all imaging modalities and both excluding and including demographic information: Structural MRI, sMRI (A), Resting state functional connectivity, RSFC (B), Diffusion Tensor Imaging, DTI (C), Multimodal (D).

STS: Superior Temporal Sulcus; Fiber TV: Fiber Tract Volume; FA: Fractional Anisotropy; Diff: Diffusivity; lh: Left Hemisphere; rh: Right Hemisphere; ex. CC: excluding Corpus Callosum; Tot: Total; Hemi: Hemisphere; Sup: Superior; Avg: Average;Fasc: Fasciculus; Long: Longitudinal;SM:Sensorimotor 

Figure 6

Figure 6 will be the brain image highlighting important regions.

Table S1: Class Distribution by Model (Maybe)

Figure S1: RSFC Models

Most important variables for Child Opportunity Index models split by mean for all imaging modalities and both excluding and including demographic information: Structural MRI, sMRI (A), Resting state functional connectivity, RSFC (B), Diffusion Tensor Imaging, DTI (C), Multimodal (D).

Ex.: Excluding; In.: Including; STS: Superior Temporal Sulcus; Fiber TV: Fiber Tract Volume; FA: Fractional Anisotropy; Diff: Diffusivity; lh: Left Hemisphere; rh: Right Hemisphere; ex. CC: excluding Corpus Callosum; Tot: Total; Hemi: Hemisphere; Sup: Superior; Avg: Average;Fasc: Fasciculus; Long: Longitudinal;SM:Sensorimotor.