Date of Award
Winter 12-19-2023
Degree Name
Doctor of Philosophy (PhD)
Degree Type
Dissertation
Abstract
Modern brain imaging modalities, particularly functional MRI (fMRI), have enabled dramatic advances in cognitive neuroscience, human brain mapping, and neural decoding. For example, in the last decade, fMRI has decoded the identities and categories of thousands of images being viewed by human participants and has even reconstructed visual scenes and sentences. However, standard fMRI employs large, non-portable equipment that cannot be used at the bedside, cannot image many patients with metal implants, is difficult in young children, is not naturalistic, and is impractical for chronic brain-computer interface (BCI) in patients with completely locked-in syndrome whose paralysis prevents communication via ordinary means. Functional near-infrared spectroscopy (fNIRS) images blood dynamics non-invasively, like fMRI, and has potential to address many fMRI limitations. While fNIRS traditionally used sparse source-detector arrays and suffered from poor resolution and distorted point-spread functions, newer high-density diffuse optical tomography (HD-DOT) systems provide higher image quality and a superior surrogate to fMRI at the brain surface. Moving from sparse 30-mm-spaced arrays, with single-distance measurements, to high-density 13-mm-spaced arrays, with multiple-distance measurements, provides better brain specificity, improved resolution, and higher contrast-to-noise ratio. In this dissertation, I first performed simulations indicating that further reducing inter-optode spacing to 6.5 mm (ultra-high density) would further improve image quality and noise-resolution tradeoff. To realize these improvements, I then designed, constructed, and validated an ultra-high-density DOT system, which imaged stimulus-evoked activations with higher spatial resolution and decoded visual stimulus position with lower error than conventional HD-DOT. Finally, I adapted a motion-energy encoding model from previous successful fMRI studies and decoded the identities of up to 40 movie clips outside the decoder’s training set using HD-DOT. Many of the powerful fMRI decoding paradigms leverage these types of encoding models that predict how the brain responds to a generalizable set of stimulus features, yet previous fNIRS and DOT decoding studies were limited to template-based decoding or simpler methods and thus were not able to decode outside the training stimuli. This dissertation research surmounts that limitation and provides a roadmap for translating powerful fMRI decoding capabilities into naturalistic settings with HD-DOT, especially as other fully wearable HD-DOT systems continue emerging.
Language
English (en)
Chair
Joseph Culver