Date of Award
Spring 2019
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Mathematics, Statistics and Computer Science
Program
Computational Sciences
First Advisor
Ahamed, Sheikh Iqbal
Second Advisor
Krenz, Gary S.
Third Advisor
Love, Richard R.
Abstract
This dissertation focuses on the non-invasive assessment of blood-hemoglobin levels. The primary goal of this research is to investigate a reliable, affordable, and user-friendly point-of-care solution for hemoglobin-level determination using fingertip videos captured by a smartphone. I evaluated videos obtained from five patient groups, three from the United States and two from Bangladesh, under two sets of lighting conditions. In the last group, based on human tissue optical transmission modeling data, I used near-infrared light-emitting diode sources of three wavelengths. I developed novel image processing techniques for fingertip video analysis to estimate hemoglobin levels. I studied video images creating image histogram and subdividing each image into multiple blocks. I determined the region of interest in a video and created photoplethysmogram signals. I created features from image histograms and PPG signals. I used the Partial Least Squares Regression and Support Vector Machine Regression tools to analyze input features and to build hemoglobin prediction models. Using data from the last and largest group of patients studied, I was able to develop a model with a strong linear correlation between estimated and clinically-measured hemoglobin levels. With further data and methodological refinements, the approach I have developed may be able to define a clinically accurate public health applicable tool for hemoglobin level and other blood constituent assessment.