Research Repository
LU|ZONE|UL distributes and preserves the scholarly work of LU faculty. It is a space for faculty to support the dissemination of knowledge created at Laurentian.
Electronic Theses and Dissertations (ETD) Repository This section preserves Master's theses and doctoral dissertations accepted at Laurentian University and is a mechanism for making this form of scholarly work widely accessible.
Communities in LU|ZONE|UL
Select a community to browse its collections.
Recent Submissions
Prediction and survival analysis of head and neck cancer in patients using epigenomics data and advanced machine learning methods
(2023-08-22) Chaudhary, Vikaskumar
Epigenomics is the field of biology dealing with modifications of the phenotype that do
not cause any alteration in the sequence of cell DNA. Epigenomics adds something to the top of
DNA to change the properties, which eventually prohibits certain DNA behavior from being
performed. Such modifications occur in cancer cells and are the sole cause of cancer. The main
objective of this research is to perform prediction and survival analysis of Head and Neck
Squamous Cell Carcinoma (HNSCC) which is one of the biggest reasons of death and accounts
for more than 650,000 cases and 330,000 deaths annually worldwide. Tobacco use, alcohol
consumption, Human Papillomavirus (HPV) infection (for oropharyngeal cancer), and Epstein-
Barr Virus (EBV) infection are the main risk factors associated with head and neck cancer (for
nasopharyngeal cancer). Males, with a proportion ranging from 2:1 to 4:1, are slightly more
affected than females. Four different types of data are used in this research to predict HNSCC in
patients. The data includes methylation, histone, human genome and RNA-Sequences. The data is
accessed through open-source technologies in R and Python programming languages. The data is
processed to create features and with the help of statistical analysis and advanced machine learning
techniques, the prediction of HNSCC is obtained from the fine-tuned model. The optimal model
was determined to be ResNet50 utilizing the Sobel feature selection method for image data and
ReliefF-based feature selection for clinical features, achieving a test accuracy of 97.9%. The
model's precision score was 0.929, its recall score was 0.930, and its F1 score was 0.930.
Additionally, the ResNet101 model demonstrated the best performance using the Histogram of
Gradients feature selection method for image data and mutual information-based feature selection
for clinical features, yielding a test accuracy of 96.1%. Its precision score, recall score, and F1
score were identical to the aforementioned ResNet50 model. The research also utilized Kaplan-
Meier survival analysis to investigate the survival rates of patients based on various factors,
including age, gender, smoking status, tumor size, and location of site. The results obtained from
this analysis yielded the effectiveness of the method in providing valuable insights for risk
assessment.
Supporting pairwise comparisons method by internet services
(2023-06-28) Xue, Songwen
The pairwise comparisons method helps us with decision-making that involves
multiple criteria. The consistency-driven pairwise comparisons method has
been proven to be especially helpful with inconsistent data. By combining the
pairwise comparisons method with nowadays’ advanced web technology, the
decision-making process is simplified. This thesis integrates the theory of the
pairwise comparisons method, and Koczkodaj’s inconsistency measurement,
and reduction algorithms into an online implementation. The implementation
utilizes JavaScript and its technologies, CSS, and HTML. It also uses version
control technologies like GitHub and GitHub Pages. GitHub is a platform and
cloud-based service for software development and version control. GitHub
pages provide online access to develop software.
Investigating the role of hydrogen sulfide in the regulation of glucagon-like peptide-2 secretion and gut physiology
(2023-08-29) Hammond, Joel
Hydrogen sulfide (H2S) is an endogenously produced gasotransmitter which regulates a variety
of physiological processes including hormone regulation. In individuals with Crohn’s disease or
colitis, H2S is elevated. Whether H2S plays a role in gut disease development or protection is not
clear. Glucagon-Like Peptide-2 (GLP-2) is a gut health-promoting peptide hormone produced in
the enteroendocrine L-cells (EECs) located primarily in the distal gastrointestinal (GI) tract. In
the gut, GLP-2 increases cell proliferation, enhances barrier function, and decreases
inflammation. In this study, it is hypothesized that local H2S produced in the distal GI tract and
in EECs can regulate GLP-2 secretion and downstream gut physiology. GLP-2 secretion and gut
physiology was examined in mice lacking the H2S producing enzyme cystathionine gamma-lyase
(CSE). Additionally, H2S production and GLP-2 secretion was examined from mouse
enteroendocrine L-cells (GLUTag). CSE knockout (KO) mice had significantly reduced
intestinal H2S production and claudin-7 expression. In cells, H2S was also produced directly
from EECs which could be partially blocked by CSE inhibitors. When EECs were treated with
H2S donors (NaHS and GYY 4137), the effect on GLP-2 secretion was variable with a slight
suppression with high dose NaHS and no effect with GYY 4137. Finally, when cells were treated
with CSE inhibitor, GLP-2 secretion was significantly reduced. Our work indicates that the EEC
L-cell can produce the gasotransmitter H2S. Adding additional H2S has a variable effect while
inhibiting endogenous production suppresses GLP-2 secretion. These findings suggest that GLP-
2 may play a role in the interplay between H2S and gastric health, which could provide support
for the potential use of H2S drugs in the treatment of GI diseases.
RoBERTa: a machine reading comprehension for climate change question answering in natural language processing
(2023-06-27) Mohasina , Mohasina
With the advancement of artificial intelligence technology in various domains in the past few
years, the question-answering system has brought important changes to the knowledge acquisition
process. When compared to the conventional retrieval question-answering system, the question-
answering system that uses machine reading comprehension can provide short and accurate
answers. This thesis proposes an intelligent question-answering approach based on information
retrieval and machine reading comprehension. To begin, a two-stage information retrieval
approach is developed. To produce a contextualized vector representation, the first dense vector
technique (SRoBERTa) is utilized to roughly collect relevant climate change material. Second, an
algorithm (DPR) is used, which precisely uses and organizes related paragraphs in order to obtain
replies to the paragraph that are incredibly relevant to the question at hand. The model is then
improved using a mechanism known as RoBERTa during the machine reading comprehension
stage. It is carried out by utilizing these texts and then looking for concise and to-the-point
solutions. When compared to other common methods, the results of information retrieval and
reading comprehension show that the models developed in this study perform well.
Super-resolution image reconstruction from multiple low-resolution images
(2023-06-29) Vavadiya, Bhavinkumar
This study explores a novel approach for super-resolution image reconstruction from
multiple low-resolution images, employing frequency domain motion estimation technique
(FMT), Keren-based image interpolation, and bicubic interpolation (BI). The method
performs well in estimating scaling parameters, but accuracy decreases as shift distance or
rotation angle increases. Compared to Vandewalle's algorithm, the proposed method shows
better accuracy in estimating scaling parameters but similar accuracy for rotation and
translation parameters. Differences are observed in estimated values for each parameter
between both methods. The study underscores the need for further research to improve the
accuracy of the proposed method in motion estimation and interpolation optimization.
Additionally, Generative Adversarial Networks (GANs) outperform Bicubic and Wavelet
Domain Super-Resolution (WDSR) algorithms in image quality improvement, indicated
by higher Peak Signal-to-Noise Ratio (PSNR) values. This superior performance is
attributed to GANs' ability to leverage deep learning algorithms to capture complex image
features. The research validates the potential of the proposed method for super-resolution
image reconstruction, and the power of deep learning-based algorithms, specifically
GANs, in enhancing low-resolution images. More advanced motion estimation algorithms
and interpolation technique optimization could further improve the accuracy of this
method.