ADHD diagnostics and severity assessment using topological manifold learning of resting-state functional magnetic resonance imaging (rs-fMRI).

Non-intrusive neuroimaging technology offers fast and robust diagnostic tools for neuro-disorder disease diagnosis, such as Attention-Deficit/Hyperactivity Disorder (ADHD). Resting-state functional magnetic imaging (rs-fMRI) has been demonstrated to have great potential for such applications due to its unique capability and convenience in providing spatial-temporal brain imaging. One critical challenge of using rs-fMRI data is the high dimensionality for both spatial and temporal domains. Thus, direct use of rs-fMRI data for the diagnosis will usually perform poorly due to the "curse of dimensionality." This paper proposes a novel nonlinear dimension reduction technique for rs-fMRI data for easy downstream analysis, such as diagnostics, regression, and visualization. The proposed method integrates the Curvature Augmented Manifold Embedding and Learning (CAMEL) algorithm with key rs-fMRI features, such as Amplitude of Low-Frequency Fluctuations (ALFF), Regional Homogeneity (ReHo), and Functional Connectivity (FC). The ADHD diagnosis problem is formulated as a classification problem in the reduced latent space and is validated with 551 data points from an open fMRI database. Compared to available literature models and results, 13 %-26 % improvement in diagnostic accuracy is observed. Additionally, the proposed methodology also supports individualized ADHA severity assessment by regression analysis in the latent space and provides a potential tool for personalized treatment. Finally, an ADHD sensitivity map is developed, highlighting brain regions associated with ADHD scores and providing interpretable insights into ADHD's neural underpinnings.