2020/09/23 by Zeynep Gürler, Gurler, Zeynep, Ahmed Nebli +3
Computer Science · Medicine · Neuroscience · #Advanced Graph Neural Networks #Advanced Neuroimaging Techniques and Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Functional Brain Connectivity Studies #Image and Video Processing (eess.IV) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2009.11166
openalex publication_date 2020/09/23 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Foreseeing the brain evolution as a complex highly inter-connected system,\nwidely modeled as a graph, is crucial for mapping dynamic interactions between\ndifferent anatomical regions of interest (ROIs) in health and disease.\nInterestingly, brain graph evolution models remain almost absent in the\nliterature. Here we design an adversarial brain network normalizer for\nrepresenting each brain network as a transformation of a fixed centered\npopulation-driven connectional template. Such graph normalization with respect\nto a fixed reference paves the way for reliably identifying the most similar\ntraining samples (i.e., brain graphs) to the testing sample at baseline\ntimepoint. The testing evolution trajectory will be then spanned by the\nselected training graphs and their corresponding evolution trajectories. We\nbase our prediction framework on geometric deep learning which naturally\noperates on graphs and nicely preserves their topological properties.\nSpecifically, we propose the first graph-based Generative Adversarial Network\n(gGAN) that not only learns how to normalize brain graphs with respect to a\nfixed connectional brain template (CBT) (i.e., a brain template that\nselectively captures the most common features across a brain population) but\nalso learns a high-order representation of the brain graphs also called\nembeddings. We use these embeddings to compute the similarity between training\nand testing subjects which allows us to pick the closest training subjects at\nbaseline timepoint to predict the evolution of the testing brain graph over\ntime. A series of benchmarks against several comparison methods showed that our\nproposed method achieved the lowest brain disease evolution prediction error\nusing a single baseline timepoint. Our gGAN code is available at\nhttp://github.com/basiralab/gGAN.\n