Identifying Autism Spectrum Disorder Based on Individual-Aware Down-Sampling and Multi-Modal Learning
Autism Spectrum Disorder(ASD) is a set of neurodevelopmental conditions that affect patients' social abilities. In recent years, deep learning methods have been employed to detect ASD through functional MRI (fMRI). However, existing approaches solely concentrated on the abnormal brain functional connections but ignored the importance of regional activities. Due to this biased prior knowledge, previous diagnosis models suffered from inter-site heterogeneity and inter-individual phenotypical differences. To address this issue, we propose a novel feature extraction method for fMRI that can learn a personalized lowe-resolution representation of the entire brain networking regarding both the functional connections and regional activities. First, we abstract the brain imaging as a graph structure, where nodes represent brain areas and edges denote functional connections, and downsample it to a sparse network by hierarchical graph pooling. Subsequently, by assigning each subject with the extracted features and building edges through inter-individual non-imaging characteristics, we build a population graph. The non-identically distributed node features are further recalibrated to node embeddings learned by graph convolutional networks. By these means, our framework can extract features directly and efficiently from the entire fMRI and be aware of implicit inter-individual differences. We have evaluated our framework on the ABIDE-I dataset with 10-fold cross-validation. The present model has achieved a mean classification accuracy of 85.95% and a mean AUC of 0.92, which is better than the state-of-the-art methods.
READ FULL TEXT