Fig. 5: Communities detected from model-derived and tau SUVr-derived graph networks.
From: AI-driven fusion of multimodal data for Alzheimer’s disease biomarker assessment
The dot heatmap visualizes the detected communities within graph networks constructed from normalized mutual information (NMI): one based on the Shapley values of T1-weighted regional volumetric features (SHAP), and the other based on tau PET SUVr for each of the six regional labels, including medial temporal (med-temp), lateral temporal (lat-temp), medial parietal (med-par), lateral parietal (lat-par), frontal, and occipital. Items within the same column represent a single detected community in the corresponding graph, and communities are order-invariant. Brain regions are grouped into pre-defined Braak stages (I-II, III, IV, V, and VI) on the right for visualization purposes. Statistical annotations denote the results of a one-sided spatial permutation test (n = 5000) on the adjusted mutual information (AMI) between model-based and tau SUVr-derived communities for each regional label (med-temp: AMI = 0.219, p = 1.40 × 10−3; lat-temp: AMI = 0.176, p = 5.60 × 10−3; med-par: AMI = 0.134, p = 4.84 × 10−2; lat-par: AMI = 0.288, p = 1.60 × 10−3; frontal: AMI = 0.138, p = 2.16 × 10−2; occipital: AMI = 0.233, p = 1.00 × 10−3). Significance levels are denoted as * for p < 0.05 and ** for p < 0.01. The corresponding contingency tables are provided in Supplementary Table S20.
