Stable brain PET metabolic networks using a multiple sampling scheme

The human brain's interregional communication is vital for its proper functioning. A promising direction for investigating how these regions communicate relies on the assumption that the brain is a complex network. In this context, images derived from positron emission tomography (PET) have been proposed as a potential source for understanding brain networks. However, such networks are often assembled via direct computation of inter-subject correlations, neglecting variabilities between subjects and jeopardizing the construction of group representative networks. Here, we used [18F]FDG-PET data from 1027 individuals at different syndromal stages (352 CU, 621 MCI and 234 AD) to develop a novel method for constructing stable (i.e. resilient to spurious data points) metabolic brain networks. Our multiple sampling (MS) scheme generates brain networks with higher stability when compared to the conventional approach. In addition, the proposed method is robust to imbalanced datasets and requires 50% fewer subjects to achieve stability than the conventional approach. Our method has the potential to considerably boost PET data reutilization and advance our understating of human brain network patterns in health and disease.