A Matrix Autoencoder Framework to Align the Functional and Structural Connectivity Manifolds as Guided by Behavioral Phenotypes

We propose a novel matrix autoencoder to map functional connectomes from resting state fMRI (rs-fMRI) structural Diffusion Tensor Imaging (DTI), as guided by subject-level phenotypic measures. Our specialized infers low dimensional manifold embedding for the rs-fMRI correlation matrices that mimics canonical outer-product decomposition. The is simultaneously used reconstruct DTI tractography via second alignment decoder and predict inter-subject variability an artificial neural network. validate our framework on dataset of 275 healthy individuals Human Connectome Project database clinical consisting 57 subjects with Autism Spectrum Disorder. demonstrate model reliably recovers connectivity patterns across individuals, while robustly extracting predictive interpretable brain biomarkers in cross-validated setting. Finally, outperforms several baselines at predicting behavioral phenotypes both real-world datasets.