Dementia with Lewy bodies (DLB) is characterised by marked clinical heterogeneity, which may not be fully captured by regional metabolic abnormalities on FDG-PET. We investigated whether different representations of FDG-PET data capture complementary aspects of DLB phenotype variability.

Subject-specific metabolic connectomes were derived from FDG-PET and compared with regional uptake measures and whole-brain network descriptors for predicting cognitive, motor, behavioural and disease-stage outcomes. Across clinical domains, mesoscale metabolic relationships between cortical regions provided the most robust predictive information, outperforming both regional and global network representations.

Predictive subnetworks highlighted the importance of distributed interactions between somatomotor and higher-order associative systems, suggesting that DLB clinical expression is encoded by large-scale network reorganisation rather than isolated regional abnormalities.

These findings support the use of network-based metabolic connectome approaches as a sensitive framework for characterising individual variability in DLB.