M.Sc. Dissertation

Abstract: Digital holographic microscopy (DHM) provides Quantitative phase images (QPI) significant when imaging transparent (e.g., biological) samples. This method of imagining requires no damage to the sample due to toxic, chemical staining, leading to a non-invasive and label-free technique. Common-path DHM systems, which are based on self-interference, are usually more robust than double-path DHM systems (based on Mach-Zehnder and Michelson configurations), being less exposed to external fluctuations. Common-path DHM systems usually require fewer optical elements which reduce the cost of the system. In this work, a 3D-printed common-path DHM system using a Fresnel biprism has been design and evaluated using both a star and USAF target samples from Benchmark Technologies. A common issue in these DHM systems is that the self-interference causes an overlay between the two sample’s images. Therefore, common-path DHM systems are restricted for dense biological and material science samples, limiting their use for only imaging sparse samples. To resolve the overlay issue in common-path systems, one can reduce the sample’s field of view using half of the imaging area or insert a spatial filter. In this work, we have also implemented two DHM systems that employ an optical pinhole to spatial filter one of the samples’ image replicas. The optimal pinhole size is evaluated by analyzing the frequency content of the reconstructed phase images of a transmissive star target.

Abstract: Motion capture systems are widely used for improving athletic performance and as a diagnostic tool in sports medicine. Standard motion capture systems capture the three-dimensional movement using an array of cameras and specialized sensors to capture segment locations in space. The major drawbacks of using commercial motion capture systems include a high initial cost, limited portability, and limited laboratory space. Inertial measurement unit-based motion capture systems offer greater environment freedom but require the attachment of many small sensors with specific orientations and placements. Currently, no motion capture system is portable, affordable, and easily accessible. In this Master thesis, we present a marker-less motion capture system using a commercially available sports camera (GoPro Hero8) and the OpenPose human pose estimation algorithm. The OpenPose algorithm detects 135 human body key points such as knee and foot key points from images. We validated the proposed marker-less system by comparing running and jumping kinematics measured using the proposed low-cost system compared to a research-grade marker-based motion capture system (240 Hz, Qualisys).In jumping, our analysis focused on detecting knee valgus position as knee valgus is a risk factor for a traumatic knee injury. Both running and jumping data revealed that the performance of the marker-less system (i.e., a low-cost system) is similar to the research-grade motion capture system. The findings of this study demonstrate that pairing a low-cost sports camera with artificial intelligence allows for high-quality analysis of human movement. However, there are improvements to be made for pose estimation algorithm accuracy. Although the OpenPose algorithm provides good causal usages, a gait analysis will require more precise body detection results. Future work will involve the improvement of the pose estimation algorithm during gait data.

Abstract: Glucose monitoring technologies allow users to monitor glycemic fluctuations (e.g., current glucose levels in their blood, also known as glycemia). This is particularly important for individuals who suffer from diabetes mellitus (DM), commonly referred to as diabetes. Traditional self-monitoring blood glucose (SMBG) devices require the user to prick their finger and extract a blood drop to measure the blood glucose based on chemical reactions with the blood. Unlike traditional glucometer devices, non-invasive continuous glucose monitoring (NICGM) devices aim to solve these issues by consistently monitoring users’ blood glucose levels (BGL) and without invasively acquiring a sample. This Master Thesis aims to investigate the feasibility of a novel approach to NICGM via the use of off-the-shelf wearable sensors and the integration of learning-based models (i.e., machine learning). Several sensors were purchased to generate our own dataset with an increased feature set for studying possible relationships between glucose and non-invasive biometric measurements. Two datasets were collected for this study: (1) the OhioT1DM dataset, which is a publicly available dataset that can be obtained by contacting Ohio University; and (2) the UofM dataset, which was created by this research team. Both the Ohio dataset and our UofM dataset are passed through a machine learning pipeline that tests several models to determine whether the features are sufficient for predicting blood glucose concentrations. While preliminary results seem optimistic, a larger dataset is required to make conclusions about the feasibility of this approach.