Brain decoding of the Human Connectome Project Tasks in a Dense Individual fMRI Dataset.

Brain decoding aims to infer cognitive states from patterns of brain activity. Substantialinter-individual variations in functional brain organization challenge accurate decodingperformed at the group level. In this paper, we tested whether accurate brain decodingmodels can be trained entirely at the individual level. We trained several classifiers ona dense individual functional magnetic resonance imaging (fMRI) dataset for which sixparticipants completed the entire Human Connectome Project (HCP) task battery >13times over ten separate fMRI sessions. We evaluated nine decoding methods, fromSupport Vector Machines (SVM) and Multi-Layer Perceptron (MLP) to GraphConvolutional Neural Networks (GCN). All decoders were trained to classify singlefMRI volumes into 21 experimental conditions simultaneously, using ~7h of fMRI dataper participant. The best prediction accuracies were achieved with GCN and MLPmodels, whose performance (57-67% accuracy) approached state-of-the-art accuracy(76%) with models trained at the group level on >1k hours of data from the originalHCP sample. Our SVM model also performed very well (54-62% accuracy). Featureimportance maps derived from MLP —our best-performing model— revealedinformative features in regions relevant to particular cognitive domains, notably in themotor cortex. We also observed that inter-subject classification achieved substantiallylower accuracy than subject-specific models, indicating that our decoders learnedindividual-specific features. This work demonstrates that densely-sampledneuroimaging datasets can be used to train accurate brain decoding models at theindividual level. We expect this work to become a useful benchmark for techniques thatimprove model generalization across multiple subjects and acquisition conditions.