Abstract
Organizing visual experience into semantic categories, such as objects and actions, and their hierarchical relationships, provides a powerful lens for studying how the human brain encodes the visual world. Early neuroimaging work established that discrete semantic categories such as faces, places, and bodies evoke distinct and reliable cortical responses. More recently, functional MRI (fMRI) has leveraged high-dimensional semantic feature spaces and voxelwise encoding models to reveal large-scale maps of semantic organization across cortex. fMRI has pioneered this work with its exceptional spatial resolution and ability to collect multi-session data. However, its scanner environment limits ecological validity and restricts semantic mapping to laboratory settings. While functional Near-Infrared Spectroscopy (NIRS) studies have expanded neuroimaging into real-world settings, sparse optode arrays, limited spatial coverage, and poor reconstruction fidelity limit its use for advanced semantic mapping paradigms. High-Density Diffuse Optical Tomography (HD-DOT) is the tomographic extension of fNIRS and enables functional mapping with fidelity closer to fMRI. This dissertation advances visual semantic mapping in two directions: establishing that HD-DOT can support advanced semantic mapping and extending that capability toward fully unconstrained wearable optical neuroimaging. I collected 3.5 hours of naturalistic movie-viewing data from six participants using the highest performing fiber-based DOT system. Voxelwise encoding models with 1,708 semantic categories robustly predicted HD-DOT responses, yielding single-category cortical maps consistent with the fMRI literature. Decoding analyses identified which clips participants viewed, and clustering approaches revealed shared higher-order semantic organization across cortex. Translating this capability to real-world environments requires a high-performance wearable system. To this end, we developed and validated an untethered, whole-head wearable HD-DOT system against functional localizers, naturalistic audiovisual movie viewing, and live piano performance using both single-feature mapping and template-based decoding approaches. Applying the same voxelwise encoding and decoding framework to an initial subset of three participants collected with the wearable system, I demonstrated semantic model-based category maps and above-chance decoding performance, comparable to the fiber-based system. Together, these studies lay the foundation for bringing visual semantic mapping into real-world environments.
Committee Chair
Joseph Culver
Committee Members
Alexander Huth; Christine O'Brien; Deanna Barch; Joseph O'Sullivan; Song Hu
Degree
Doctor of Philosophy (PhD)
Author's Department
Interdisciplinary Programs
Document Type
Dissertation
Date of Award
5-27-2026
Language
English (en)
DOI
https://doi.org/10.7936/xwg3-0k89
Recommended Citation
Fehner, Wiete, "Advancing High-Density Diffuse Optical Tomography for Visual Semantic Encoding and Decoding in Naturalistic Settings" (2026). McKelvey School of Engineering Graduate Student Theses & Dissertations. 1365.
The definitive version is available at https://doi.org/10.7936/xwg3-0k89