Evaluation of GAMMA Generated Biomarkers

GAMMA-generating discriminative ROIs of the most representative ICNs were evaluated in the following steps.

Let ROI^j(ICN^k) denote the jth ROI for the kth ICN, which was selected by C4.5. Let Dⁱ(ICN^k) represent the high connectivity discretization map for subject i of the kth ICN, and Dⁱ(ICN^k) ∩ ROI^j(ICN^k) represents the overlap ROI for ROI^j(ICN^k) and Dⁱ. As a result, N_{Dⁱ(ICN^k)∩ROI^j(ICN^k)} represents the number of voxels that were in both ROI^j(ICN^k) and Dⁱ. Let Fⁱ(ICN^k) represent the functional connectivity map for subject i of the kth ICN, and FC_{Fⁱ(ICN^k)∩ROI^j(ICN^k)} represents the average function connectivity for subject i within the jth ROI of the kth ICN. Both N_{Dⁱ(ICN^k)∩ROI^j(ICN^k)} and FC_{Fⁱ(ICN^k)∩ROI^j(ICN^k)} were two functional integration patterns of the jth ROI for the kth ICN. The two-sample t-test was applied to calculate different sizes (N_{Dⁱ(ICN^k)∩ROI^j(ICN^k)}) and FCs (FC_{Fⁱ(ICN^k)∩ROI^j(ICN^k)}) of discriminative ROIs between MHE and NMHE. A p value < 0.05 was considered to indicate a significant difference.

To investigate the relationship between discriminant results and behavior criteria, we computed correlation coefficients between the GAMMA results and Trail Making Test A (TMT-A), Trail Making Test B (TMT-B), Digit Symbol Test (DST), Block Design Test (BDT), and Child-Pugh scores. In this study, we computed partial correlations between two functional integration patterns of the discriminative ROIs within the selected ICNs and clinical criteria by controlling for movement (the motion parameters were optimized by the band-pass temporal filter according to the previous study⁶⁰), age, gender and years of education. The p-value < 0.05 (Bonferroni corrected) was considered as a significant correlation.