Open Collections

Deng, Xinchen

University of British Columbia

Radiation therapy aims to kill tumour cells while sparing healthy tissue. However, a current bottleneck in radiation therapy remains in the area of personalization of radiation dose to match inherent individual tumour and patient radiosensitivity. This dissertation is devoted to developing a radiation response monitoring tool by combining Raman spectroscopy (RS) with machine learning methods to monitor radiation response in cellular and tissue materials. RS is a non-invasive optical technique that provides detailed spectroscopic information on the molecular composition of the sample and can potentially be used to monitor biological radiation response in both healthy and cancerous tissues. This dissertation has three main objectives. The first objective is to apply and evaluate the non-negative matrix factorization (NMF) performance as a dimension reduction tool in RS data analysis. This dissertation shows that the non-negativity constraint of NMF helps the interpretation of decomposed chemical bases. NMF identified new radiation response biomarker such as lipids that were not previously discovered. The second objective is to combine a semi-supervised dimension technique, GBR-NMF, with a classifier method such as random forest (RF) as a data analytical framework (GBR-NMF-RF) to classify the tumour samples into radiation response groups. It is shown that GBR-NMF-RF helped RS identify radiation response biomarkers such as glycogen, lipids, DNA, and amino acids in cancer cell samples. The third objective is to investigate the performance of GBR-NMF-RF in different types of Raman samples (e.g., cells and tissues). Using RS and GBR-NMF-RF, radiation responses are compared in data sets acquired from cancer cells and tissue xenografts from the murine model. It is shown that GBR-NMF-RF and RS also discovered correlations between biochemicals such as glycogen and amino acids with hypoxia in RS xenograft tissue samples. In conclusion, the work conducted in this dissertation demonstrates the potential of RS, combined with machine learning methods, to be used as a tool to monitor radiation response in cellular and tissue environments. These tools may aid in the overall goal of personalized radiation therapy.

Text

2022-10-03

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/82810

Medical Physics

University of British Columbia

UBCO

http://creativecommons.org/licenses/by-nc-nd/4.0/