Efficient Neural Representations in Spatial Navigation and Pharmacokinetic Imaging: Information-Theoretic and Machine Learning Approaches

Author

Jared Deighton, The University of TennesseeFollow

Date of Award

5-2025

Degree Type

Dissertation

Degree Name

Doctor of Philosophy

Major

Mathematics

Major Professor

Vasileios Maroulas

Committee Members

Catherine Schuman, Ioannis Sgouralis, Piotr J. Franaszczuk

Abstract

Mammalian spatial navigation relies on specialized neurons, such as place and grid cells, which encode position based on self-motion and environmental cues. While extensive research has explored the computational role of grid cells, the principles underlying efficient place cell coding remain less understood. Existing spatial information measures primarily assess single-neuron encoding, limiting insights into population-level representations. To address this gap, we introduce novel information-theoretic measures that quantify the encoding efficiency of multiple neurons, including the joint stimulus information rate for neuron pairs and the spectral-stimulus information for arbitrary populations. The spectral-stimulus information, defined as the leading eigenvalue of the stimulus information matrix, is maximized when neurons exhibit localized, non-overlapping firing fields—mirroring place cell and head direction cell activity in biological systems. We demonstrate that these measures can be used to train recurrent neural networks (RNNs) via self-supervised learning, leading to the emergence of place cells and head direction cells. Our findings highlight how neural populations collectively encode stimuli, offering a more comprehensive framework for understanding place cell formation and optimizing artificial navigation systems in novel environments. Additionally, functional ultrasound imaging (fUSI) provides high spatiotemporal resolution for monitoring cerebral blood volume (CBV) by detecting backscattered echoes from red blood cells. While fUSI has been widely used in preclinical neuroscience, many studies focus on predetermined regions of interest (ROIs), potentially overlooking relevant brain activity. To address this, we combined fUSI with three machine learning models—convolutional neural networks (CNNs), Vision Transformers (ViTs), and Support Vector Machines (SVMs)—to analyze the pharmacokinetics of Dizocilpine (MK-801), an NMDA receptor antagonist. CNNs demonstrated superior performance in detecting and localizing drug-induced changes in brain hemodynamics. Using class activation mapping (CAM), CNNs identified anatomically specific patterns in cortical and hippocampal regions that align with known NMDA receptor distributions. Quantitative analysis confirmed significant drug-induced CBV reductions in CNN-identified regions. While all models distinguished between drug and control conditions, CNNs uniquely maintained anatomical precision while tracking drug effects over time. By integrating advanced machine learning with fUSI and spectral-spatial measures, our work provides novel insights into both neural representation and pharmacological effects on brain function. These approaches enhance our ability to model spatial coding in artificial systems and improve the detection of drug-induced neural changes, advancing both computational neuroscience and biomedical imaging.

Recommended Citation

Deighton, Jared, "Efficient Neural Representations in Spatial Navigation and Pharmacokinetic Imaging: Information-Theoretic and Machine Learning Approaches. " PhD diss., University of Tennessee, 2025.
https://trace.tennessee.edu/utk_graddiss/12347