Decodability of huamn brain activity measured with fMRI

Abstract

In this dissertation, we explored about decodability of human brain activity non-invasively measured with fMRI signals. Recent brain decoding paradigms have mostly been used for well-known visual modalities. Therefore, we investigated the feasibility of decoding the brain signal from human motor cortex. First, we performed a simulation study to understand how orientation selectivity, a fundamental low-level visual feature, could be successfully decoded from human brain activity measured with fMRI. Such low-level visual features have a finer spatial scale than the resolution of fMRI. Therefore, we investigated how we could decode the orientation selectivity by modeling the functionally distributed map of neurons in primary visual cortex (V1) and characterizing the orientation tuning property. Second, we also performed a simulation study to understand how motor action is represented by spatially distributed patterns of fMRI response in primary motor cortex (M1) and to investigate the feasibility of decoding complex motor action using non-invasive fMRI. To this end, we modeled the functionally distributed map of neurons in M1 and characterizing the direction selectivity which is the basic functional property in M1. Third, we performed the real fMRI experiment to investigate whether fMRI responses from human motor cortex could be used to predict individual motor action. We performed decoding and reconstructing the arm movement direction. Encoding model was used to reconstruct unknown movement directions. Through a series of human brain decoding studies, we discussed the possibility of deciphering the brain activity in human motor cortex and discussed the feasibility of the non-invasive brain decoding techniques on Brain-machine Interface (BMI).