Search

Programmes for diabetic retinopathy screening have increased the number of patients screened for sight threatening diabetic retinopathy (STDR) and permitted monitoring of people at risk of progression to referable diabetic retinopathy (RDR). A disparity between demand for screening and capacity to provide screening has arisen from the rapid increase in the prevalence of diabetes. Over 350,000 people have diabetes on the island of Ireland; most (~90%) have Type 2 diabetes (T2D). The current mode of universal annual DR screening is unsustainable. A possible solution would be to increase screening intervals for those at "low risk". The means of reliable identification of those at low risk will require identification of the factors associated with disease progression to RDR in Ireland. In chapter 3 of this thesis, we identify the factors associated with progression to RDR in the form of weighted covariates. Chapter 4 herein describes a prospective external validation of pre-existent prediction models for stratifying risk of progression to RDR in people with T2D in Ireland. Inputting into these pre-existent prediction models developed in Iceland and Gloucester (England) involved taking a linear combination of the published coefficients from the models and the corresponding variables in our dataset at the last systemic visit before the index retinal screening. The accuracy of the models in predicting progression to RDR was compared with the outcomes observed in a prospective Irish cohort (n=939). Receiver operating characteristic (ROC) curves were used to assess models' performance. Models developed in Iceland and Gloucester had an acceptable performance with an AUC of ~ 0.70 or above. This signifies that there would be a >70% probability that a randomly selected subject from the screening cohort who did in fact develop RDR would have been allocated to the higher risk score category by each of the models. This suggests relative safety in individualising intervals between screening episodes.

Imaging technology is a growing field that provides solutions in many areas from manufacturing to agriculture. Through previous research, imaging technologies have been studied in livestock farming to monitor the animal’s health and welfare in the production process. However, the feedlot industry is still behind in validating the feasibility to use some of these technologies nor to adopt such technologies to address challenges the industry is facing, such as lack of skilled labor.This work proposes using novel imaging methods to identify feed types and estimate the amount of feed remaining in a typical Midwestern feedlot feed bunk. These methods have promising potential to provide alternative tools to feedlot operations to alleviate labor requirements for tasks like bunk calling, feed sourcing, and feed mixing. This approach, if successful, provides an alternative option that allows existing systems to incorporate these methods into their framework to accurately perform daily tasks.The main contribution of this work is to leverage imaging technologies, specifically, depth imaging and machine learning techniques to build and validate models that can be used in the feedlot production systems in the Midwestern U.S. To date, several studies have explored the use of imaging technologies and machine learning to monitor individual cow intake in dairy production, but there is limited research body comprehensively conducted to explore these technologies for feedlot applications.The proposed methods were used to collect imagery data for eleven common feedlot ingredients and seven diets. Collected images were processed (a) to estimate the weights of residual feed in the bunk, (b) to evaluate the accuracy of depth cameras in estimating residual feed, (c) to characterize the different feed textures, and (d) to classify feed textures using machine learning techniques. Regression models using pixel transformation were developed to correlate image-model-estimated and the scale-measured feed weights, whereas texture analysis techniques and residual neural network model in 10 classes were used to identify the individual ingredients. Methodologies and results are presented in this thesis as a paper format. The major findings indicate that using low-cost depth cameras and machine learning techniques is promising in the development of alternative tools to estimate the amount of residual feed in concrete bunks and identify individual feed ingredients commonly used in commercial feedlots in the U.S. Advisor: Yijie Xiong

New technology developments foster new biology. Beginning from the early observation of cells by Robert Hooke and Anton van Leeuwenhoek, who figured out how to use optical lenses to achieve high magnification, microscopy has made huge advancement and each single discovery in cell biology has been accompanied with the development of microscopy innovations. Recent advances in microscopy have reached a previously unachievable resolution and acquisition speed, allowing people study the heterogenous cellular phenotypes accurately and efficiently. However, not all questions can be addressed with the existing approaches. For example, what is the single molecule basis of translation, how single ribosome translates; can we combine imaging based technologies with other existing approaches like CRISPR to further speed up cell biology discoveries? All of these questions require further effort to develop suitable imaging based technologies to address and this is what I aim to do during my PhD. In Chapter 2 and Chapter3, I will introduce two newly developed imaging based technologies to address the above mentioned questions and together, these technologies provide new opportunities to further facilitate cell biology researches.

Tissue engineering is an interdisciplinary field that combines various aspects of engineering and life sciences and aims to develop biological substitutes to restore, repair or maintain tissue function. Currently, the ability to have quantitative functional assays of engineered tissues is limited to existing invasive methods like biopsy. Hence, an imaging tool for non-invasive and simultaneous evaluation of the anatomical and functional properties of the engineered tissue is needed. In this thesis, we present an advanced in-vivo imaging technology - ultrasound biomicroscopy combined with complementary photoacoustic and elasticity imaging techniques, capable of accurate visualization of both structural and functional changes in engineered tissues, sequential monitoring of tissue adaptation and/or regeneration, and possible assistance of drug delivery and treatment planning. The combined imaging at microscopic resolution was evaluated on tissue mimicking phantoms imaged with 25 MHz single element focused transducer. The results of our study demonstrate that the ultrasonic, photoacoustic and elasticity images synergistically complement each other in detecting features otherwise imperceptible using the individual techniques. Finally, we illustrate the feasibility of the combined ultrasound, photoacoustic and elasticity imaging techniques in accurately assessing the morphological and functional changes occurring in engineered tissue.