Breast cancer is the most common cancer among women worldwide. Mammography is considered as the golden standard and effective screening tool in detecting breast cancer at its early stage. The extensive screening programmes and health care standards lead to the necessity of Computer-Aided Diagnosis tools to support radiologists. In this thesis, we investigate the potential of computer vision and image processing algorithms in developing CAD algorithms for breast density estimation and micro-calcification classification. Many studies have shown that breast density and parenchymal patterns are reliable indicators for breast cancer risk. So, for risk assessment, the thesis proposed a multi-scale elliptical blob modelling for parenchymal pattern representation based on a statistical analysis of breast tissue. To describe the breast density type and risk, we analysed the breast for blob-like dense tissue. The study showed a relation with the BI-RADS density classification based on fatty and dense tissue blob structures. Subsequently, the thesis proposed a new variant to LBP called Mean-Elliptical Local Binary Pattern (M-ELBP), where the intensity and texture features were combined in an elliptical topological structure. The method is more robust to noise as it does not perform a direct comparison between pixels as in traditional LBP and has the benefit of extracting features from multiple orientations. The studies based on density estimation investigated the potential of ROI size, descriptor size, and classifier effect on density classification. The validity of the proposed methods is evaluated using the MIAS and DDSM databases. The obtained classification accuracy was up to 77.4% for MIAS tissue-based class. BI-RADS based classification attained up to 75.4% and 47.65% for MIAS and DDSM. For microcalcification classification, a novel method called connected-chain is developed based on analysis of the topological/distributional pattern of micro-calcifications in MCCs. The connectivity/closeness between individual microcalcification was analysed to discriminate between benign and malignant case. The performance of the method was tested using three datasets: MIAS, DDSM, OPTIMAM obtaining classification accuracies of 82.50%, 86.4% and 76.75%, respectively.
|Goruchwyliwr||Reyer Zwiggelaar (Goruchwylydd) & Bernie Tiddeman (Goruchwylydd)|