Adaptive method for mammography computer aided diagnoses designed for embedded system / Mostafa Ahmed Mohammed Elsayed Eltager ; Supervised Yasser M. Kadah

Thesis (M.Sc.) - Cairo University - Faculty of Engineering - Department of Systems and Biomedical Engineering

This thesis introduces a novel methodology for CAD system aside to the conventional method used. The novel method deals with the case as whole, so that it classify the case to normal-normal case, abnormal-abnormal case or normal-abnormal depending on the state of the both sides of the breast. This is in contrast to the conventional method that deals with the sides of the breast independently, and classify the breast image to either normal or abnormal, regardless the over side of the case. Also in this thesis an implementation of the CAD on the low-power embedded systems is introduced, in replace to the PC-based CAD systems. The system is used to detect the presence on any abnormality in the mammography images. In feature extraction phase, 395 features is investigated in each of the spatial, Fourier and wavelet domains of the picture this adds up to 1975 feature. These features have been tested using the two methodologies on the (mini-MIAS) database. In feature selection method, the enormous readings generated from the both methods undergoes different feature selection methods to select the most relevant feature set. Then a number of machine learning algorithms is applied to the results. The successes rate varies depend on the machine learning algorithm. It reaches 85% with a good potential towards the novel methodology

Issued also as CD