3-D BREAST CANCER DETECTION USING SUPPORT VECTOR MACHINES AND FINITE ELEMENT METHODS

Abbasi Layegh, M and Ghobadi, CH and Nourinia, J and Muhammadi, B (2014) 3-D BREAST CANCER DETECTION USING SUPPORT VECTOR MACHINES AND FINITE ELEMENT METHODS. The Journal of Urmia University of Medical Sciences, 25 (6). pp. 540-548.

Preview

Text 10 AbbasiLayegh.pdf Download (455kB) | Preview

Abstract

Breast cancer is one of the most prevalent non-skin-related malignancies among women in the world. Thus, many countries have commenced screening test in early stages in order to diagnose breast cancer. Buried object detection is performed in the present work to detect 3-D breast cancer applying SVM classifier. Some transmitters and receivers are located above the breast. Each transmitter radiates monochromatic electromagnetic wave. Materials & Methods: When a patient is positioned in a supine position, the breast is naturally flattened. The breast tissue is considered as 3-D lattice of classification cells. Since breast tissue can be considered with high order of accuracy as a linear medium, the received signals are monochromatic. Results: Some particular ranges of parameters have been covered during model selection. Almost all samples in the training set were converted to support vectors. It is worthy to remind that single radiation source (dipole) has been used for the present work. Conclusion: The simulation has been carried out on synthetic electromagnetic data obtained by means of Finite Element Method and Perfectly Matched Layers techniques. Noisy environments have been considered as well in order to simulate realistic conditions of signal measurement. The real positions of the tumor are marked by white contour squares. The dependence of the prediction quality on the depth of the tumor location is not observable unless the tumor is directly under the skin layer. Probability maps obtained demonstrate that the region around the tumor location usually clearly stands out against the background of overall probability values

Actions (login required)

View Item