3D Inverse Electromagnetic Solver Using Deep Neural Network Towards Breast Cancer Detection

Nowadays, Breast cancer is one of the leading health problems in women both in the developed and the developing countries. The frequency of breast cancer cases is rising in the developing world because of the increase in life expectancy, urbanization and modern lifestyles. The success of treatment is well connected with the early detection of this disease. For early detection of malignancy, microwave imaging technique can be used on a regular basis for monitoring, as it is nonionizing, non invasive and economical. The dielectric properties of breast tissues can be used for the detection of cancerous regions. The scattered electromagnetic fields from breast tissues are obtained when it is excited in an electromagnetic field. This field can be analysed to estimate the properties of the scatterer, here the breast tissues, and this turns out to be an inverse, ill posed and computationally intensive problem. In this work, Deep Neural Network is used for cracking the inverse problem, that is to find the complex permittivity from the scattered field and thereby detecting the cancerous voxels. The different structures of malignancies with constant dielectric properties are used to train the network. Different structures of malignancies of breast model are reconstructed using deep neural network and it distinguishes between benign, malignant and skin voxels.