Design of efficient generative model for human vision reconstruction using fMRI activity profiles

Abstract

Human brain is the primary organ of the central nervous system and it newlinecontrols the many body activities, it is primarily in charge of higher order and newlinemore complicated functions including memory, perception of senses, thought, newlinefocus, emotions, movement and language. In order to, reconstruct the visual data newlinefrom its function by functional Magnetic Resonance Imaging (fMRI). newlineThe term quotfunctional Magnetic Resonance Imagingquot refers to the use of newlinethis technique to identify localized variations in blood flow and blood newlineoxygenation in the brain that take place in built with the intention of mapping newlinethe human brain and used to study brain functions including vision, language, newlinemotor, and cognition. newlineThe principle objective of this thesis was to design and develop an newlineefficient method using generative models for reconstructing the visual stimuli newlinefrom the fMRI activity profiles. However, the challenges involved in this study newlineinclude the complexity of the structure. In order to overcome these difficulties, newlinethe growth of high-quality computational approaches has been executed. In the newlinepresent work, the hybrid form is proposed that combines the features extraction newlinemodel and conditional generative adversarial network techniques. newlineFor mapping fMRI data to visual stimuli, the two hybrid form of newlinenetworks namely Siamese conditional Generative Adversarial Network newline(ScGAN) and Independent Component Analysis conditional Generative newlineAdversarial Network (ICAcGAN) employing is suggested in this research. newline