ORCID
https://orcid.org/0009-0001-3214-3749
Date of Award
Spring 2025
Language
English
Embargo Period
4-29-2027
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
College/School/Department
Department of Chemistry
Program
Chemistry
First Advisor
Igor Lednev
Committee Members
Igor Lednev, Alexander Shekhtman, Melinda Larsen, Jia Sheng, Neil Gildener-Leapman
Keywords
Raman Spectroscopy, ATR-FTIR Spectroscopy, Forensics Trace evidence, Sjögren's Disease Diagnostics, Chemometrics, Machine Learning
Subject Categories
Biochemistry | Biotechnology
Abstract
This dissertation explores the integration of Raman spectroscopy with machine learning to advance two critical applications: forensic body fluid identification and medical diagnostics. Building on research from the Lednev group, this work seeks to address limitations in current methodologies by providing innovative, non-invasive, and highly accurate solutions.
The first objective focuses on the development of a universal forensic model capable of identifying six main body fluids—peripheral blood, sweat, semen, saliva, vaginal fluid, and urine. Unlike existing methods, this approach is confirmatory, non-destructive, and requires minimal sample preparation, offering significant improvements in forensic investigations. Extending this framework, the research demonstrates the potential of Raman spectroscopy for phenotypic profiling. Specifically, a machine learning model using Random Forest was developed to differentiate urine stains from donors of Caucasian and African American descent with 90% accuracy. This novel capability enables rapid phenotypic profiling at crime scenes, significantly narrowing suspect pools and enhancing investigative efficiency.
The second objective investigates the application of Raman hyperspectroscopy combined with machine learning in medical diagnostics, focusing on Sjögren's disease—a chronic autoimmune disorder affecting exocrine glands. Sjögren’s disease is frequently underdiagnosed due to non-specific symptoms and limitations of current diagnostic techniques, which are invasive, costly, and lack sensitivity and specificity. This research establishes a novel screening method that effectively distinguishes Sjögren's disease patients from healthy controls and individuals with radiation-induced dry mouth. The proposed method offers a rapid, non-invasive, and cost-effective alternative for early detection, using Raman hyperspectral signatures to address diagnostic gaps and improve patient outcomes.
Furthermore, we investigated the same objective utilizing the complementary vibrational spectroscopy technique ATR-FTIR and a machine learning classification model. We attained 90% accuracy for ten external validation samples from various donors.
This dissertation utilizes the flexibility of Raman spectroscopy combined with machine learning to address essential gaps in forensic science and medical diagnostics. The results underscore the opportunity to create universal, dependable, and effective tools for criminal investigations and healthcare, showcasing the transformative influence of these technologies in their respective domains.
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Vyas, Bhavikkumar N., "Machine Learning-Enhanced Raman Spectroscopy for Sjögren’s Disease Diagnostics and Forensic Body Fluid Analysis" (2025). Electronic Theses & Dissertations (2024 - present). 209.
https://scholarsarchive.library.albany.edu/etd/209
