Jan 19, 2017
12:00 PM



Humans interacting face-to-face make use of auditory cues from the talker’s voice and visual cues from the talker’s mouth to understand speech. I will discuss computational models that use Bayesian principles to predict human speech perception. These models make specific predictions about the neural mechanisms of speech perception, which we test with electrocorticography (ECoG) and BOLD fMRI. The anatomical focus of these studies is the human posterior superior temporal gyrus and sulcus (pSTS), a brain region known to be important for speech perception. The pSTS has a complex organization, with some regions responding to specific visual stimuli and others to specific auditory stimuli. Using ECoG, we demonstrate a double dissociation in the pSTG. More anterior regions show a greater neural activity to audiovisual speech with a clear auditory component, as predicted for unisensory auditory regions, while more posterior regions showed similar or greater responses to noisy audiovisual speech, as predicted for multisensory cortex. Using BOLD fMRI, we show for the first time that the natural statistics of human speech, in which voices co-occur with mouth movements, are reflected in the neural architecture of the pSTS. Different pSTS regions prefer visually-presented faces containing either a moving mouth or moving eyes, but only mouth-preferring regions respond strongly to voices. 

About the Speaker

Dr. Beauchamp is a Professor in the Department of Neurosurgery and the Department of Neuroscience at Baylor College of Medicine with adjunct appointments at the University of Texas McGovern Medical School and Rice University. His laboratory examines multisensory integration and visual perception using a variety of techniques, including BOLD fMRI, electrocorticography, and computational modeling. Dr.Beauchamp is the Director of Research in the Department of Neurosurgery and the Director of CAMRI, the Core for Advanced MRI. He graduated from Harvard College in 1992, receiving his PhD from the University of California, San Diego in 1997 before pursuing postdoctoral studies in the National Institute of Mental Health Intramural Research Program.

Jan 26, 2017
11:00 AM
Building 75, room 1103


Calibrated functional MRI was developed to tease apart the hemodynamic and metabolic contributions to the blood oxygenation level-dependent (BOLD) signal using simultaneous measurements of the BOLD signal and cerebral blood flow. While calibrated fMRI has substantially improved our ability to image and understand aspects of brain physiology, it has not been widely adopted due to the need for specialized gas delivery equipment and biophysical confounds associated with the calibration measurements. One such confound is the magnetic susceptibility of dissolved oxygen, which, like deoxyhemoglobin, is paramagnetic. In this talk, I will present work on modelling and measuring the susceptibility and relaxation rates of dissolved oxygen in blood and what impact they have on the hyperoxic BOLD signal and calibration. I will also discuss recent efforts towards improving the accuracy of gas-free BOLD calibration. This work could greatly increase the appeal of calibrated fMRI by eliminating the gas challenge completely. In all, I hope these studies underscore the role that analytical modelling and simulations can play in improving our understanding of the biophysics of the BOLD signal and in guiding imaging strategies to probe brain physiology.