Date of Award
Doctor of Philosophy
Electrical and Computer Engineering
Dr. Ikhlas Abdel-Qader
Dr. Dionysios Kounanis
Dr. Liang Dong
Microcalcifications are residual calcium deposits that are often the first signs of developing breast abnormalities that may lead to breast cancer. Up to 30% of cancerous lesion in diagnosed breast cancer cases could have been detected earlier through mammogram screenings if the right tools were available. While the detection of calcifications may be easier in fatty backgrounds, it is challenging in dense parenchyma, suggesting the need for more sensitive tools for accurately identifying suspicious regions in mammograms and propping a computer-aided system for further target classification. Therefore, the objective of the research work in this dissertation is to develop a novel highly sensitive method for the detection of microcalcification that is independent of the characteristics of background tissue.
Continuous wavelet transform is employed to detect singularities in mammograms by tracking modulus maxima along maxima lines. This work is based on convolving the mammogram with Gaussian kernel to detect and extract microcalcifications that are modeled as smoothed impulse functions. Two significant characteristics of the local modulus maxima of the wavelet transform with respect to the smoothed impulse function are investigated: magnitude of general maximum and fractal dimension of the detected sets of singularities. It is also essential to select the suitable computation parameters such as thresholds of magnitude, argument, and frequency range in accordance with spatial and numerical resolution of the analyzed mammogram. This detection approach is independent of the background tissue and is complementary to a computer-aided diagnosis system based on shape, morphology, and spatial distribution of individual microcalcifications.
Experimental work is performed on a set of images with empirically selected parameters for 200 urn/pixel spatial and 8 bits/pixel numerical resolution. Results are indicating that in abnormal regions the selected general maxima have larger magnitudes and tend to have higher fractal dimension than in surrounding normal regions. Findings are promising since they can be integrated into any framework for breast cancer detection and diagnosis.
Bujanovic, Tomislav, "Spatial Frequency Localization in Mammograms Using Wavelets" (2009). Dissertations. 650.