Automated detection of differences in treated and untreated breast tissue through analysis of microwave imaging data
Date
2022-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Although microwave imaging has been researched in various applications for about 40 years, its use in biomedical applications is a more recent endeavor. In this thesis, we examine the feasibility of using automated detection on microwave imaging data for providing treatment-related feedback on changes in breast tissue. 16 female patients at the Tom Baker Cancer Centre in Calgary, were recruited from a clinical trial, to be scanned by the University of Calgary's Microwave Imaging Transmission System (MITS) for a maximum of 4 times, over a period of 2 years. Early stage breast cancer treatment typically involves lumpectomy and 1-5 weeks of radiotherapy to the breast that contained the tumour. Our hypothesis is that changes in breast tissue due to cancer treatment may be detected through analysis of microwave imaging data using machine learning algorithms. Data analysis began with exploration of the microwave frequency properties of tissue or tissue permittivity to find differences between treated and untreated tissue. Challenges of identifying specific and consistent changes across the group of patients using permittivity analysis led to switching our approach to analysis of the underlying time-domain signals. Employing wavelet transforms on the time-domain signals resulted in more defined differences between the treated and the untreated breast for feature extraction. Next, classifiers like Support Vector Machine, Random Forest and Gradient Boosting Classifier were used on the extracted features. A final analysis of the frequency domain signals and combined time-frequency domain features was also undertaken to highlight differences and apply classification to the extracted features. This thesis provides a framework for an automated technique to detect changes between treated and untreated breast tissue using the microwave scan data. Our results indicate that this approach to analyzing microwave imaging data may have the potential to extract differences in breast tissue arising from radiotherapy and/or surgery.
Description
Keywords
Microwave imaging, Breast cancer, Machine learning, Classification, Tissue changes
Citation
Garland, A. (2022). Automated detection of differences in treated and untreated breast tissue through analysis of microwave imaging data (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.