Cascaded Correlation Neural Network Based Microcalcification Detection in Mammographic Images

This paper presents a novel approach for classification of microcalcification (MC) clusters in mammograms. This cluster is the significant indication of breast cancer in women at the early stage. Diagnosis of these clusters at the early stage is a very difficult task as the cancerous tumors are embedded in normal breast tissue structures. This paper proposes an artificial intelligent neural network algorithm - Cascaded Correlation Neural Network (CCNN) - for detection of tumors in mammograms. CCNN has a distinct feature that it does not use a predefined set of hidden units, instead the hidden units gets added up one by one until the error is minimized. By exploiting this distinct feature of the CCNN, a computerized detection algorithm is developed that are not only accurate but also computationally efficient for microcalcification detection in mammograms. Prior to MC detection texture features from the Region of Interest (ROT) of the mammmographic Image is extracted using gabor features. Then CCNN classifier is used to determine whether the input data is normal/benign/malignant. The performance of this scheme is evaluated using a database of 322 mammograms from MIAS database and real time clinical mammograms. The result shows that the proposed CCNN algorithm has good performance.