Neural Encoding and Interpretation of High-Level Visual Cortices Based on fMRI Using Image Caption Features

On basis of functional magnetic resonance imaging (fMRI), researchers are devoted to designing visual encoding models predict the neuron activity human in response presented image stimuli and analyze inner mechanism cortices. Deep network structure composed hierarchical processing layers forms deep by learning features data on specific task through big dataset. have powerful representation data, brought about breakthroughs for encoding, while revealing structural similarity with manner information However, previous studies almost used those pre-trained classification construct models. Except structure, or corresponding dataset is also important models, but neglected studies. Because a relatively fundamental task, it difficult guide master high-level semantic representations which causes into that performance cortices limited. In this study, we introduced one higher-level vision task: caption (IC) proposed model based IC (ICFVEM) encode voxels Experiment demonstrated ICFVEM obtained better than task. addition, interpretation was realized explore detailed characteristics visualization words, comparative analysis implied behaved correlative content.