Neural Representation Of Three-dimensional Acoustic Space In The Human Cortex

Sound localization is an important function of the human brain, and it is a fundamental question in neuroscience. Physiological studies have found that auditory neurons in mammals have very broad tuning curves to sound locations, and the underlying cortical mechanisms of sound localization remain unclear. Most previous studies examined the neural encoding mechanisms using sounds presented along the horizontal plane, while in reality sounds may come from anywhere within the entirety of three-dimensional(3D)space. In this study, we design behavioral experiment and functional Magnetic Resonance Imaging(fMRI) experiment to explore the underlying mechanisms of spatial sound perception over all of 3D space in the human cortex with machine methods. We discuss the following three questions:(1) Is there any information about sound locations in the human cortex?(2) What is the neural representation for the three-dimensional space in the human cortex?This thesis consist of two parts: experiments design and data analysis. In the experiments design part, we first build a specialized behavioral platform, and record 3D spatial sounds with inner-ear microphones. These recorded sounds are used as subject-specific spatial sounds since speaker-delivered spatial sounds could not be directly presented during fMRI scan. Then we design an fMRI experiment and obtain the brain images related with listening to 3D spatial sounds.In the data analysis part, we use univariate analysis and Multivariate Pattern Analysis(MVPA) to detect the activations evoked by 3D spatial stimuli in the human cortex. We propose a Region-of-Interest(ROI) based two-step voxel selection approach, and then build models to predict brain images related with listening to spatial sounds. We also build brain image classification models and successfully decode sound locations from these images. Our results reveal that sound locations could be decoded from neural activities in the brain cortex. In addition, we define E?ective Voxel(EV) and Significant E?ective Voxel(SEV), and investigate the layout and consistency of spatial patterns in the cortex across subjects. Voxels exhibiting robust activity formed distinct patterns in the temporal lobe that were e?ective in discriminating di?erent sound locations across subjects. Abscissa discrimination(X axis) was more related to the left posterior STG and right mid-STG. Ordinate discrimination(Y axis) was more related to the left mid-MTL and right mid-STG. Finally, elevation discrimination(Z axis) was more related to the left posterior MTL and right STG. The di?erences in classification accuracies and the SEV’s spatial patterns under di?erent conditions suggest dissimilar processing mechanisms for sound location along di?erent dimensions of 3D acoustic space. These results support a distributed representation of acoustic space in human cortex.