Comparative evaluation of graph construction methods for individual brain metabolic network from FDG-PET images: an ADNI study in healthy subjects
European Journal of Nuclear Medicine and Molecular Imaging, vol.53, no.2, pp.1139-1154, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 53 Issue: 2
- Publication Date: 2026
- Doi Number: 10.1007/s00259-025-07462-1
- Journal Name: European Journal of Nuclear Medicine and Molecular Imaging
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Biotechnology Research Abstracts, CINAHL, EMBASE, MEDLINE
- Page Numbers: pp.1139-1154
- Keywords: ADNI, Connectivity, Dynamic time warping, Effect size, FDG-PET, Healthy control, Individual graph construction, Kullback-Leibler divergence, Wasserstein
- Galatasaray University Affiliated: No
Abstract
Purpose: Connectivity analyses of fluorodeoxyglucose positron emission tomography (FDG-PET) static images provide a valuable means of investigating brain network organization by capturing metabolic activity at rest. Graph theory is emergently applied to model these networks at individual level; however, the choice of graph construction method can significantly impact analytical outcomes. Methods: In this study, we systematically evaluate and compare methods for building individual graphs from FDG-PET images in healthy control subjects. Specifically, we assess five methods, categorized into mean-based graphs and probability density function (PDF)-based graphs, using two criteria: structural similarity between individual and group-level graphs, and their hub topology structure analysis. Results: Our findings indicate that the Effect Size-based (ES) method best preserves group-level graph structure, achieving 98.9% similarity for the averaged graph while also maintaining around 84% similarity for individual graphs. Among PDF-based approaches, the Wasserstein (WA) method, with its adaptability in PDF-based settings, provides the highest similarity across both averaged (82.5%) and individual (79.1%) graphs, with its adaptive in PDF-settings, making it the most effective for multi-scale network analysis. Meanwhile, Dynamic Time Warping (DTW) captures the highest individual variability, as reflected by its largest variation among individual graphs (11.5%). Conclusion: This analysis highlights the unique strengths and limitations of each method, emphasizing the critical importance of careful method selection tailored to specific research objectives. Additionally, our study suggests a framework for selecting the appropriate methods, with implications for further both research and clinical applications.