Author: Dexter  G  Canning

Major: Mathematics

Graduation Year: 2019

Thesis Advisor: Andre Khalil

Description of Publication: Improving computer-aided early detection techniques for breast cancer is paramount because current technology has high false positive rates. Existing methods have led to a substantial number of false diagnostics, which lead to stress, unnecessary biopsies, and an added financial burden to the health care system. In order to augment early detection methodology, one must understand the breast microenvironment. The CompuMAINE Lab has researched computational metrics on mammograms based on an image analysis technique called the Wavelet Transform Modulus Maxima (WTMM) method to identify the fractal and roughness signature from mammograms. The WTMM method was used to color code the mammograms based on the type of tissue present and assign the Hurst exponent (H) value to corresponding tissue: dense tissue with H greater than 0.55, fatty tissue with H less than 0.45, and disrupted tissue with H between 0.45 and 0.55, with the latter being a key trait in tumorous tissue. This analysis on the full breast was performed on 127 cases for the Medio Lateral Oblique (MLO) view. We are revisiting these data by analyzing the region behind the nipple for the MLO view and the region outside the nipple area. After performing the WTMM analysis on each breast, non-parametric statistical analysis methods were performed to determine the level of significance between normal, benign, and cancerous cases. Furthermore, we utilized logistic models to assess the predictability of these metrics for future datasets.

Location of Publication:

URL to Thesis: https://digitalcommons.library.umaine.edu/honors/568/