The interconnection between brain regions in neurological disease encodes vital information for the advancement of biomarkers and diagnostics. Although graph convolutional networks are widely applied for discovering brain connection patterns that point to disease conditions, the potential of connection patterns that arise from multiple imaging modalities has yet to be fully realized. In this paper, we propose a multi-modal sparse interpretable GCN framework (SGCN) for the detection of Alzheimer's disease (AD) and its prodromal stage, known as mild cognitive impairment (MCI). In our experimentation, SGCN learned the sparse regional importance probability to find signature regions of interest (ROIs), and the connective importance probability to reveal disease-specific brain network connections. We evaluated SGCN on the Alzheimer's Disease Neuroimaging Initiative database with multi-modal brain images and demonstrated that the ROI features learned by SGCN were effective for enhancing AD status identification. The identified abnormalities were significantly correlated with AD-related clinical symptoms. We further interpreted the identified brain dysfunctions at the level of large-scale neural systems and sex-related connectivity abnormalities in AD/MCI. The salient ROIs and the prominent brain connectivity abnormalities interpreted by SGCN are considerably important for developing novel biomarkers. These findings contribute to a better understanding of the network-based disorder via multi-modal diagnosis and offer the potential for precision diagnostics. The source code is available at https://github.com/Houliang-Zhou/SGCN.

Integrating brain images and genetic data provides a great opportunity to discover potential biomarkers for neurological disorder diagnosis. However, learning genetic information and brain network dysfunction remains a challenging task. In this paper, we propose an interpretable multi-modal imaging and genetic graph convolution network (GCN) for Alzheimer’s disease diagnosis. Our genetic network uses hierarchical GCN to mimic a gene ontology-based graph of biological processes and learn the information flow in this graph. In parallel, our imaging network uses a sparse interpretable GCN with node and edge importance probabilities to learn the brain network from multi-modal images. After multi-modal fusion, the final representation guided by a cluster-based consistency constraint is used to predict the disease-related clinical measures. We evaluate our method on the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. Our result shows that our imaging-genetics framework achieves superior prediction performance compared to all state-of-the-art methods. The interpretation demonstrated that the salient SNPs, and salient regions interpreted by important probabilities were significantly correlated with AD-related clinical symptoms, and considerably important for developing novel biomarkers. The code is available at https://github.com/Houliang-Zhou/IG-GCN.

High annotation costs serve as a significant hurdle in deploying modern deep learning architectures for clinically relevant medical applications, especially when dealing with the inherent heterogeneity of multimodal data, proving the critical need for innovative algorithms that can effectively utilize unlabeled data. In this paper, we propose a model named MCLCA, which integrates multimodal contrastive learning and cross-modal attention to diagnose Alzheimer’s Disease (AD) and identify biomarkers using both labeled and unlabeled multimodal brain imaging genetics data. Through multimodal contrastive learning, MCLCA can effectively learn representations even in the absence of sufficient labels. By utilizing cross-modal attention blocks, the model captures deep connections between different modalities, providing a more comprehensive view of diagnosis. Our proposed MCLCA model is evaluated using the ADNI database with three imaging modalities (VBM-MRI, FDG-PET, and AV45-PET) and genetic SNP data. The results demonstrate that MCLCA can identify important biomarkers with better prediction accuracy compared to the existing methods. The source code is available at https://github.com/MCLCA.