COMPUTER AIDED SYSTEM FOR BREAST CANCER DIAGNOSIS USING CURVELET TRANSFORM

• Main Author: MOHAMED ELTOUKHY, MOHAMED MESELHY
• Format: Thesis
• Language: English
• Published: 2011
• Online Access: http://utpedia.utp.edu.my/2817/1/PhD_Thesis_EE_Mohamed_Meselhy_Eltoukhy.pdf; http://utpedia.utp.edu.my/2817/

Breast cancer is a leading cause of death among women worldwide. Early detection is the key for improving breast cancer prognosis. Digital mammography remains one of the most suitable tools for early detection of breast cancer. Hence, there are strong needs for the development of computer aided diagnosis (CAD) systems which have the capability to help radiologists in decision making. The main goal is to increase the diagnostic accuracy rate. In this thesis we developed a computer aided system for the diagnosis and detection of breast cancer using curvelet transform. Curvelet is a multiscale transform which possess directionality and anisotropy, and it breaks some inherent limitations of wavelet in representing edges in images. We started this study by developing a diagnosis system. Five feature extraction methods were developed with curvelet and wavelet coefficients to differentiate between different breast cancer classes. The results with curvelet and wavelet were compared. The experimental results show a high performance of the proposed methods and classification accuracy rate achieved 97.30%. The thesis then provides an automatic system for breast cancer detection. An automatic thresholding algorithm was used to separate the area composed of the breast and the pectoral muscle from the background of the image. Subsequently, a region growing algorithm was used to locate the pectoral muscle and suppress it from the breast. Then, the work concentrates on the segmentation of region of interest (ROI). Two methods are suggested to accomplish the segmentation stage: an adaptive thresholding method and a pattern matching method. Once the ROI has been identified, an automatic cropping is performed to extract it from the original mammogram. Subsequently, the suggested feature extraction methods were applied to the segmented ROIs. Finally, the K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) classifiers were used to determine whether the region is abnormal or normal. At this level, the study focuses on two abnormality types (mammographic masses and architectural distortion). Experimental results show that the introduced methods have very high detection accuracies. The effectiveness of the proposed methods has been tested with Mammographic Image Analysis Society (MIAS) dataset. Throughout the thesis all proposed methods and algorithms have been applied with both curvelet and wavelet for comparison and statistical tests were also performed. The overall results show that curvelet transform performs better than wavelet and the difference is statistically significant.