Differentiating Individuals with and without Alcohol Use Disorder Using Resting-State fMRI Functional Connectivity of Reward Network, Neuropsychological Performance, and Impulsivity Measures

Figures

• 

  Figure 1. The study protocol listing the sample, measures, and analytic techniques. The sample consisted of two groups of 30 individuals each, viz., AUD and CTL. The measures used in the prediction model included rs-fMRI functional connectivity (reward network), impulsivity assessed with Barratt impulsiveness scale (BIS), and neuropsychological performance scores. Major analyses were features selection for selecting FC variables, random forest classification method, and correlational analyses, including zero-order and partial correlations.

• 

  Figure 2. Brain reward network regions of interest comprising 16 cortical areas (top panels, C1–C3) and 18 subcortical areas (bottom panels, S1–S3), as listed in Table 1. Bilateral cortical regions: Insula (Ins), Parahippocampal Gyrus (PHG), Anterior Cingulate Cortex (ACC), Mid-Cingulate Cortex (MCC), Mid-Cingulate Cortex (MCC), Posterior Cingulate Cortex (PCC), Orbitofrontal Cortex (OFC), Angular Gyrus (Ang), Superior Parietal Lobule (SPL), Dorsolateral (PFC), and Dorsolateral Prefrontal Cortex (DLP). Bilateral subcortical regions: Ventral Tegmental Area (V), Nucleus Accumbens (N), Amygdala (A), Hippocampus (Hi), Caudate (Cd), Pallidum (Pa), Thalamus (T), and Putamen (Pt). Colors: Blue—Left hemispheric regions; Red—Right hemispheric regions. Views: C1/S1—Axial top view; C2/S2—Coronal front view; C3/S3—Sagittal left view. Directions: A—Anterior; P—Posterior; S—Superior; I—Inferior; L—Left; R—Right.

• 

  Figure 3. Receiver operating characteristic (ROC) curve for the RF model for different true positive rate (sensitivity) and false positive rate (1-specificity) are shown. The area under the ROC curve was 93%. (The further away the curve from the diagonal dashed line and the closer the swelling to the top left corner, the better the classification).

• 

  Figure 4. The distribution of minimal depth among the trees of the forest for the significant variables is shown in different colors for each level of minimal depth. The mean minimal depth in the distribution for each variable is marked by a vertical black bar overlapped by a value label inside a box. Based on the mean minimal depth values, the importance list comprised 2 BIS scores, 13 FC, and 1 neuropsychological score, which contributed to the RF classification of AUD and CTL individuals. The lower mean minimal depth of a feature represents a higher number of observations (participants) categorized in a specific group based on the feature. The number of trees for a feature represents the total number of decision trees in which a split occurs on the feature (see Table 2 for details about the ROI numbers (1–34) represented in the FC variables).

• 

  Figure 5. Multi-way importance plot showing the top significant features (labeled and marked with black circles) that contributed to the classification of alcohol use disorder from control individuals based on the measures Gini decrease (mean decrease in node impurity or classification error), number of trees (total number of decision trees in which a split occurred), and p-value (probability of node splits). The top variables of importance included 2 impulsivity scores, 12 rsFC connections, and 1 neuropsychological variable (see the circled and labeled dots). Notations in the variable labels: BIS–Barratt Impulsivity Scale; MI–Motor impulsivity; NP–Non-planning; TotCor_Fw–Total correct forward. ROIs of FC variables: Refer to Table 2 for details about the ROI numbers (1-34) that are represented in the rsFC variable names.

• 

  Figure 6. Illustration of rankings of features based on correlation between any two random forest (RF) parameters. The panels in the lower triangle of the grid show the distribution of rankings of all predictor variables with black dots along a blue trend line. The panels in the upper triangle of the grid show correlation coefficients across rankings of any two parameters. It is shown that all RF parameters of importance were found to have very high correlations among each other, suggesting the high reliability of each of these parameters in ranking the importance of features for classification. The asterisks (***) represents that the correlations were highly significant (p < 0.001).

• 

  Figure 7. Significant reward network connections that contribute to the random forest classification of alcohol use disorder (AUD) from control (CTL), as listed in Table 3. (A): Hypoconnectivity manifested by the AUD group (cyan lines) across cortical and subcortical ROI regions of the RN, predominantly involving right hemisphere structures; (B): Hyperconnectivity in three connections manifested by the AUD group (orange lines). Images within each panel: Left: axial (top) view; Middle: coronal (front) view; Right: sagittal (left) view.

• 

  Figure 8. Correlation matrix showing associations among the top significant variables based on explorative (descriptive) correlational analysis for the interpretative purpose. Values within each cell represent a bivariate Pearson correlation between the variable on its vertical axis and the variable on its horizontal axis. Correlation coefficients are color-coded (red/pink shades represent negative r-values, blue/cyan shades indicate positive r-values, darker color represent higher magnitude) and significant correlations (before Bonferroni correction) have been marked with asterisks in black font (* p < 0.05; ** p < 0.01; *** p < 0.001). The significant correlations that survived Bonferroni correction have been marked with +++ sign in white font (+++ Significant after Bonferroni correction). Refer to Table 2 for details about the ROI numbers (1–34) that are represented in the rsFC variable names. Acronyms: FC–Functional connectivity, TotCor_Fw–total number of correctly performed forward trials, BIS–Barratt Impulsiveness Scale, NP–Non-planning, MI–Motor Impulsivity.