2020/09/23 by Alaa Bessadok, Bessadok, Alaa, Mohamed Ali Mahjoub +3
Biochemistry, Genetics and Molecular Biology · Computer Science · Neuroscience · #Cell Image Analysis Techniques #FOS: Computer and information sciences #Functional Brain Connectivity Studies #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Topological and Geometric Data Analysis
paper · pdf · doi:10.48550/arxiv.2009.11058
openalex publication_date 2020/09/23 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Several works based on Generative Adversarial Networks (GAN) have been\nrecently proposed to predict a set of medical images from a single modality\n(e.g, FLAIR MRI from T1 MRI). However, such frameworks are primarily designed\nto operate on images, limiting their generalizability to non-Euclidean\ngeometric data such as brain graphs. While a growing number of connectomic\nstudies has demonstrated the promise of including brain graphs for diagnosing\nneurological disorders, no geometric deep learning work was designed for\nmultiple target brain graphs prediction from a source brain graph. Despite the\nmomentum the field of graph generation has gained in the last two years,\nexisting works have two critical drawbacks. First, the bulk of such works aims\nto learn one model for each target domain to generate from a source domain.\nThus, they have a limited scalability in jointly predicting multiple target\ndomains. Second, they merely consider the global topological scale of a graph\n(i.e., graph connectivity structure) and overlook the local topology at the\nnode scale of a graph (e.g., how central a node is in the graph). To meet these\nchallenges, we introduce MultiGraphGAN architecture, which not only predicts\nmultiple brain graphs from a single brain graph but also preserves the\ntopological structure of each target graph to predict. Its three core\ncontributions lie in: (i) designing a graph adversarial auto-encoder for\njointly predicting brain graphs from a single one, (ii) handling the mode\ncollapse problem of GAN by clustering the encoded source graphs and proposing a\ncluster-specific decoder, (iii) introducing a topological loss to force the\nreconstruction of topologically sound target brain graphs. Our MultiGraphGAN\nsignificantly outperformed its variants thereby showing its great potential in\nmulti-view brain graph generation from a single graph.\n