Abstract:
Diagnostic ultrasound imaging of the breast is characterized by its non-invasive, radiation free and convenient nature. Such qualities distinguish it from other imaging modalities of the breast, such as mammography and MRI, and has made it an increasingly popular choice for researchers, clinicians as well as patients. Conventional ultrasound imaging of the breast is mainly qualitative in nature, based on analyzing morphological features, and often have a lack of functional and quantitative information. Thus, although conventional ultrasound has a high sensitivity to breast lesions, it often lacks the specificity required to classify such lesions into their benign or malignant counterparts. In this regard, quantitative ultrasound techniques present a viable alternative, as it can provide specific quantitative variables by which to assess tissue features, and thus potentially increase the specificity as well as classification accuracy related to the procedure. This thesis presents a classification approach for benign and malignant tumors of the breast, utilizing parameters extracted from the power spectrum of ultrasound radio-frequency echo signals. Our work shows a clear separation between the two types of lesions with regards to the parameters. Furthermore, the parameters are able to classify lesions with an accuracy of 100% using Linear Support Vector Mechanism, and thus has the potential to assist in the diagnostic procedure associated with the detection of breast cancer.
