Early identification of subjective cognitive decline using 2D horizontal visibility graph analysis of structural MRI networks

Subjective cognitive decline (SCD) represents a preclinical stage of Alzheimer's disease, yet objective evaluation criteria for early diagnosis remain lacking. Traditional morphometric indicators, such as gray matter volume or density, overlook local voxel features and inter-voxel associations. To address this limitation, this study introduces a two-dimensional horizontal visibility graph (2DHVG) analysis method to identify brain structural abnormalities in SCD patients. Each axial slice of gray matter images was converted into a 2DHVG, and complex network features, including clustering coefficient and betweenness centrality, were extracted to characterize local connectivity and global information transmission capacity. Principal component analysis and light gradient boosting machine were employed for feature selection and classification. The 2DHVG-based method achieved excellent performance in SCD identification, with a mean classification AUC of 0.958, mean accuracy of 89.0%, mean sensitivity of 86.1%, mean specificity of 90.3%, and mean F1-score of 82.9%. Furthermore, significant positive correlations were observed between the mean clustering coefficient and both ADNI working memory (ADNI_MEM, R=0.3339, P=0.0014) and learning ability indicators (RAVLT.learning, R=0.3531, P=0.0007). The mean betweenness centrality similarly correlated with ADNI_MEM (R=0.3018, P=0.0041) and RAVLT.learning (R=0.3149, P=0.0027). The 2DHVG-based structural imaging analysis method demonstrates significant advantages in feature extraction and classification modeling, providing novel insights for graph-theoretic modeling of structural MRI data in early SCD identification.