Microcalcification detection using Self Organizing Neuro Glia Network classifier

Microcalcifications (MCs), tiny calcium deposits, are the very first sign of breast cancer in women. Their early detection is the most effective way to recovery. In this paper, we propose a novel approach for microcalcifications detection based on Self Organizing Neuro Glia Network (SONG-Net) classifier, a combination of supervised and unsupervised neural network. It models the role of glial cells in the neural network activity. The detection process consists of contrast enhancement, textural features extraction, then, region classification. The proposed method was tested on images from Digital Database for Screening Mammography (DDSM) and on a private database. Obtained results confirm that including astrocyte type glial cells-like behavior to the neural network provides better learning performances than using a Multi-Layer Perceptron (MLP), above all, considerably minimizes the learning time.