Author ORCID Identifier

https://orcid.org/0009-0005-3073-6777

Semester

Summer

Date of Graduation

2026

Document Type

Dissertation

Degree Type

PhD

College

Statler College of Engineering and Mineral Resources

Department

Chemical and Biomedical Engineering

Committee Chair

Soumya Srivastava

Committee Member

Srinivas Palanki

Committee Member

Timothy Eubank

Committee Member

Donald Adjeroh

Committee Member

Loren Rieth

Abstract

Pancreatic ductal adenocarcinoma (PDAC) carries a five-year survival rate of 13%, driven largely by late-stage diagnosis and the absence of sensitive noninvasive screening tools. Current diagnostic methods lack the sensitivity to detect disease-associated systemic changes at early and precancerous stages, representing a critical clinical gap. This dissertation addresses the central hypothesis that PDAC and its pathological subtypes induce measurable, disease-associated electrophysiological changes in circulating PBMCs that can serve as correlative liquid-biopsy biomarkers. To test this hypothesis, this work asks: can dielectrophoresis (DEP), impedance and zeta potential analysis with AI-assisted classification distinguish benign pancreatic conditions (non-malignant control), precancerous lesions (primarily IPMNs), PDAC, and other pancreatic and periampullary pathologies through label-free electrophysiological profiling of PBMCs? The work advances through four aims: electrical signature identification; immune subpopulation analysis via flow-sorted PBMC subsets and flow cytometry; correlative validation using zeta potential and impedance-based characterization; and AI-based multiparameter classification, including bulk PBMC analysis and PBMC subpopulations (T cells, Monocytes, Neutrophils, and Others).

Using a DEPtech 3DEP analyzer and single-shell biophysical modeling, seven dielectric parameters were quantified per group. Pairwise Welch’s t-tests identified statistically significant differences (p < 0.05), across the four groups – benign, precancerous, PDAC, and others, with no single parameter separating all four; membrane capacitance (Cmem), membrane conductivity (σmem), and crossover frequencies (fx01, fx02) emerged as the most consistently discriminating feature combination (AUC up to 0.969). PDAC-associated PBMCs showed a reduced fₓ₀₁ (≈ 203 kHz) relative to precancerous (≈ 342 kHz) and benign groups (≈ 155 kHz), with Clausius–Mossotti modeling revealing elevated peak polarization in PDAC. Complementary impedance analysis–including phase, Nyquist, imaginary impedance, and Cole–Cole permittivity plots, demonstrated distinct frequency-dependent electrical profiles across disease states, with benign PBMCs showing the most divergent impedance behavior. Among supervised machine learning classifiers, Random Forest demonstrated the strongest and most stable performance (AUC ≈0.967, CV accuracy ≈0.832), outperforming linear and kernel-based models across all four disease groups.

Correlative validation integrated zeta potential characterization of sorted PBMC subpopulations. Resulting data from healthy PBMCs show distinct electrokinetic baseline profiles across subpopulations − monocytes (−25.4 mV), T-cells (−26.6 mV), Neutrophils (-23.9 mV), and other subsets (−17.9 mV). Under PDAC, the changes were subset-specific rather than uniform: monocytes became less negative (−8.6 mV), the most pronounced shift in the dataset, while other subsets varied. This identified monocytes as the most electrically distinct subset, with zeta potential and model-derived impedance converging on the monocyte phenotype as correlative, not mechanistic, support. RNA profiling and mechanistic validation are considered future directions for work. Collectively, this work supports DEP-based immunophenotyping of PBMCs as a rapid, noninvasive adjunct for pancreatic disease evaluation, with future refinement through subpopulation-specific ML model training and clinical validation.

Download

Share

COinS