多尺度脑指纹影像组学与图论网络指标整合用于阿尔茨海默病的特征刻画与分类

BACKGROUND: The rising prevalence of Alzheimer's disease (AD) associated with global population aging has become a critical health concern. Accurate and early diagnosis is essential for timely clinical intervention. Structural MRI-based radiomics and graph-theoretical modeling have recently provided new perspectives for identifying neurodegenerative changes.

OBJECTIVE: This study aimed to develop a unified framework integrating morphological brain fingerprinting and graph-theoretical network analysis to classify AD, mild cognitive impairment (MCI), and normal controls (NC), and to characterize the structural alterations across these distinct disease stages.

METHODS: High‑dimensional radiomic features were extracted from multiple AD‑related regions in structural MRI to construct individualized brain fingerprints. A graph structure was established using the inter‑regional correlation matrix, from which network metrics were derived to quantify topological connectivity, including node degree, clustering coefficient, and betweenness centrality. Support vector machine (SVM) and ensemble‑based classifiers were then employed to perform group classification and identify key imaging biomarkers.

RESULTS: In the AD vs NC classification, the model achieved the highest training accuracy (ACC = 0.9563), sensitivity (0.950), specificity (0.9625), and AUC = 0.995 (95% CI: 0.983-1.000); testing performance remained robust (ACC = 0.825, AUC = 0.935). For AD vs MCI and MCI vs NC, the training AUC were 0.972 and 0.952, while the testing AUC were 0.895 and 0.920, respectively. Feature mapping revealed a pattern of structural alteration, with the hippocampus, entorhinal cortex, and posterior cingulate showing significant abnormalities. Graph metrics indicated decreased integration efficiency and compensatory centrality increases within high-order cognitive networks.

CONCLUSION: The proposed fingerprint-graph integration framework enables multilevel quantification of AD‑related structural reorganization, offering interpretable and reliable classification performance. It provides a promising foundation for early diagnosis, individualized assessment, and precision‑oriented neurodegenerative research.